Documentos CEDE

ISSN 1657-7191 Edicin electrnica.

Media Markets, Special Interests, and Voters

Leopoldo Fergusson

JUNIO DE 2012

08

Centro de Estudios sobre Desarrollo Econmico

CEDE

Electronic copy available at: http://ssrn.com/abstract=2104297

Serie Documentos Cede, 2012-08 ISSN 1657-7191 Edicin electrnica. Junio de 2012 2012, Universidad de los AndesFacultad de EconomaCEDE Calle 19A No. 1 37 Este, Bloque W. Bogot, D. C., Colombia Telfonos: 3394949- 3394999, extensiones 2400, 2049, 3233 infocede@uniandes.edu.co http://economia.uniandes.edu.co Ediciones Uniandes Carrera 1 Este No. 19 27, edificio Aulas 6, A. A. 4976 Bogot, D. C., Colombia Telfonos: 3394949- 3394999, extensin 2133, Fax: extensin 2158 infeduni@uniandes.edu.co Edicin y prensa digital: Cadena S.A. Bogot Calle 17 A N 68 - 92 Tel: 57(4) 405 02 00 Ext. 307 Bogot, D. C., Colombia www.cadena.com.co Impreso en Colombia Printed in Colombia El contenido de la presente publicacin se encuentra protegido por las normas internacionales y nacionales vigentes sobre propiedad intelectual, por tanto su utilizacin, reproduccin, comunicacin pblica, transformacin, distribucin, alquiler, prstamo pblico e importacin, total o parcial, en todo o en parte, en formato impreso, digital o en cualquier formato conocido o por conocer, se encuentran prohibidos, y slo sern lcitos en la medida en que se cuente con la autorizacin previa y expresa por escrito del autor o titular. Las limitaciones y excepciones al Derecho de Autor, slo sern aplicables en la medida en que se den dentro de los denominados Usos Honrados (Fair use), estn previa y expresamente establecidas, no causen un grave e injustificado perjuicio a los intereses legtimos del autor o titular, y no atenten contra la normal explotacin de la obra.

Electronic copy available at: http://ssrn.com/abstract=2104297

Centro de Estudios sobre Desarrollo Econmico

CEDE

Media Media Markets, Markets, Special Special Interests, Interests, and and Voters Voters Media Markets, Special Interests, and Voters
Leopoldo Fergusson Leopoldo Fergusson Leopoldo Fergusson June 20, 2012 June 20, 2012 June 20, 2012
Abstract Abstract Abstract This paper examines the role of mass media in countering special interest group inuThis paper examines the role of mass media in countering special interest group inuence. I use examines the concentration ofmass campaign Political Action This paper the role of media contributions in countering from special interest group Cominuence. I use the concentration of campaign contributions from Political Action Committees to proxy special interests capture US Senate candidates from 1980 to Com2002, ence. I use the concentration of campaign contributions from Political Action mittees to proxy special interests capture US Senate candidates from 1980 to 2002, and compare the reaction of voters capture to increases in concentration infrom two di erent mittees to proxy special interests US Senate candidates 1980 to types 2002, and compare the reaction of voters to increases in concentration in two di erent types of media markets in-state media markets and out-of-state media markets. Unlike and compare the reaction of voters increases in out-of-state concentration in two dierentUnlike types of media markets in-state media to markets and media markets. in-state media markets, out-of-state markets and focus on neighboring states politics and of media markets in-state media markets out-of-state media markets. Unlike in-state media markets, out-of-state markets focus on neighboring states politics and elections. Thus, if citizens punish political capture, increases in concentration of special in-state media markets, out-of-state markets focus increases on neighboring states politics and elections. Thus, if citizens punish political capture, in concentration of special interest contributions to a punish particular candidate should reduce his vote share in in-state elections. Thus, if citizens political capture, increases in concentration of special interest contributions to a particular candidate should reduce his vote share in in-state counties relative to the where the candidate receives coverinterest toout-of-state a particular counties, candidate should reduce his vote share less in in-state countiescontributions relative to the out-of-state counties, where the candidate receives less coverage. I nd that a one standard deviation increase inthe concentration of special interest counties relative to the out-of-state counties, where candidate receives less coverage. I nd that a one standard deviation increase in concentration of special interest contributions to a incumbents reduces theirincrease vote share by about 0.5 of to special 1.5 percentage age. I nd that one standard deviation in concentration interest contributions to incumbents reduces their vote share by about 0.5 to 1.5 percentage points in in-state counties relative to the vote out-of-state counties. Results similar contributions to incumbents reduces their share by about 0.5 to 1.5 are percentage points in in-state counties relative to the out-of-state counties. Results are similar in specications that rely solely on to variation in concentration across time are within the points in in-state counties relative the out-of-state counties. Results similar in specications that rely solely on variation in concentration across time within the same county, and when the sample is limited to in-state counties that are contiguous to in specications rely on in concentration time within the same county, andthat when thesolely sample isvariation limited to in-state countiesacross that are contiguous to out-of-state counties and have similar demographic structures. A placebo test where same county,counties and when the sample is limited to in-state counties that are contiguous to out-of-state and have similar demographic structures. A placebo test where in-state counties bordering out-of-state ones are compared to other in-state counties out-of-state counties and have similar demographic structures. A placebo test where in-state counties bordering out-of-state ones are compared to other in-state counties shows no eect, conrming identication thatto the results are not driven in-state bordering the out-of-state ones hypothesis are compared other in-state counties shows nocounties eect, conrming the identication hypothesis that the results are not driven by geographic characteristics oridentication distance from the media center of the state. shows no e ect, conrming the hypothesis that the results are not by geographic characteristics or distance from the media center of the state. driven JEL classication: D72, L82 by geographic characteristics or distance from the media center of the state. JEL classication: D72, L82 Keywords: Elections, media, interests, campaign nance JEL classication: D72, L82 special Keywords: Elections, media, special interests, campaign nance Keywords: Elections, media, special interests, campaign nance

Professor of Economics, Universidad de los Andes, Email: lfergusson@uniandes.edu.co. I Assistant Professor of Economics, Universidad de los Andes, Email:Snyder lfergusson@uniandes.edu.co . I thank Daron Acemoglu, Abhijit Banerjee, James Robinson and James for advice and many useful Assistant Professor of Economics, Universidad de los Andes, Email:Snyder lfergusson@uniandes.edu.co . I thank Daron Acemoglu, Abhijit Banerjee, James Robinson and James for advice and many useful discussions. James Snyder also provided invaluable help with much of the data used in this paper. I am thank Daron Acemoglu, Abhijit Banerjee, James Robinson and James Snyder for advice and many useful discussions. Snyder Cadena, also provided with much of the data used suggestions. in this paper. I am also indebtedJames with Ximena Sarahinvaluable Hamilton, help and Alexei Zakharov for detailed I thank discussions. James Snyder Cadena, also provided invaluable help with much of the data used suggestions. in this paper. I am also indebted with Ximena Sarah Hamilton, and Alexei Zakharov for detailed I thank seminar participants at the 2011 Royal Economic Society and Midwestern Political Science Association also indebted with Ximena Cadena, Sarah Hamilton, and Alexei Zakharov for detailed suggestions. I thank seminar participants the 2011 Royal Society and Midwestern Political Science Association Annual Meetings, theat 9th Workshop on Economic Media Economics (New Economic School and Higher School of seminar participants at the 2011 Royal Economic Society and Midwestern Political Science Association Annual Meetings, the 9th Workshop on Media Economics (New Economic School and Higher School of Economics), Bocconi University, Banco de la Rep ublica de Colombia, Centro de Estudios Monetarios y Annual Meetings, theUniversity, 9th Workshop on de Media Economics (New Economic School and Higher School of Economics), Bocconi Banco la Rep u blica de Colombia, Centro de Estudios Monetarios y Financieros (CEMFI), Instituto Madrile no de Estudios Avanzados (IMDEA), MIT/Harvard Development Economics), Bocconi University, Banco n de la Estudios Rep ublica de Colombia, CentroMIT/Harvard de Estudios Monetarios y Financieros (CEMFI), Instituto Madrile o de Avanzados (IMDEA), Development Seminar and MIT Political Economy Breakfast, Stockholm IIES, Universidad de los Andes, Universidad Financieros (CEMFI), Instituto Madrile n o de Estudios Avanzados (IMDEA), MIT/Harvard Development Seminar and Political Economy Breakfast, Stockholm IIES, Universidad de los Andes, Universidad Aut onoma deMIT Barcelona, Universidad de Barcelona IEB, Universidad del Rosario, Universitata Pompeu Seminar and MIT Political Economy Breakfast, Stockholm IIES, Universidad de los Andes, Universidad Aut o noma de Barcelona, Universidad de Barcelona IEB, Universidad del Rosario, Universitata Pompeu Fabra, and University of Warwick. Aut onoma Barcelona, Universidad de Barcelona IEB, Universidad del Rosario, Universitata Pompeu Fabra, and de University of Warwick. Fabra, and University of Warwick.

Assistant

1 Electronic copy available at: http://ssrn.com/abstract=2104297

Mercados de medios, grupos de inter es, y electores

Leopoldo Fergusson

Resumen En este documento estudio la capacidad de los medios masivos de comunicaci on para contrarrestar la inuencia de los grupos de inter es. Para ello, calculo el grado de concentraci on de las contribuciones a las campa nas electorales para el Senado de los Estados Unidos entre 1980 y 2002, y comparo la reacci on de los electores frente a incrementos en la concentraci on en dos tipos de mercados de medios los mercados -en-el-estado y mercados fuera-del-estado. Contrario a los mercados de medios en-el-estado, los mercados fuera-del-estado se enfocan en la pol tica y elecciones de estados vecinos. As , si los ciudadanos castigan la captura pol tica, incrementos en la concentraci on de las contribuciones por parte de grupos de inter es a un candidato en particular deben reducir su votaci on en los condados con mercados en-el-estado en comparaci on con aquellos fuera-del- estado, donde se da menos cubrimiento al candidato. Los resultados indican que un incremento de una desviaci on est andar en el ndice de concentraci on reduce la votaci on por el senador de turno en aproximadamente 0,5 a 1,5 puntos porcentuales en los condados en-el-estado relativo a los condados fuera-del-estado. Los resultados son similares en especicaciones que se apoyan s olo en la variaci on en la concentraci on a trav es del tiempo en el mismo condado, y cuando la muestra se restringe a condados vecinos que tienen caracter sticas sociodemogr acas similares. En una prueba de placebo, en la que condados que colindan con condados fuera-del-estado se comparan con otros condados en-el-estado, no se encuentran efectos signicativos, conrmando la hip otesis de identicaci on seg un la cual los resultados no son determinados por caracter sticas geogr acas ni por la distancia al centro del mercado de medios en el estado. Clasicaci on JEL: D72, L82 Palabras clave: Elecciones, medios, grupos de inter es, nanciaci on de campa nas

Profesor Asistente, Econom a, Universidad de los Andes, Calle 19A No 1-37 Este, Bloque W, Bogot a, Colombia, Email: lfergusson@uniandes.edu.co, Tel: +571 339 4949 Ext 2439, Fax: +571 332 4492.

Interest groups pursue their dierent strategies to inuence policy. These include gathering information that supports positions, taking their arguments to politicians and the public Interest groups pursue their dierent strategies to inuence policy. These include gathering information that supports positions, taking their arguments to politicians and the public Interest groups pursue their dierent strategies to inuence policy. These include gathering information that supports positions, taking their arguments to politicians and the public to win sympathy, and their undertaking disruptive activities to coerce policymakers into mation that supports positions, taking their arguments to politicians and themaking public to win sympathy, and undertaking disruptive activities to coerce policymakers into making mation that supports their positions, taking their arguments to politicians and themaking public to win sympathy, and and undertaking activities to coerce into concessions Grossman Helpmandisruptive (2001). Yet, the activity that policymakers receives perhaps the most to win sympathy, and undertaking disruptive activities to coerce policymakers into making concessions Grossman and Helpmandisruptive (2001). Yet, the activity that policymakers receives perhaps the most to win sympathy, and undertaking activities to coerce into making concessions Grossman Helpman (2001). Yet, the activity that receives groups perhapscampaign the most attention from media, and the public, academia, and policymakers, is interest concessions Grossman and Helpman (2001). Yet, the activity that receives perhaps the most attention from media, and the public, academia, and policymakers, is interest concessions Grossman Helpman (2001). Yet, the activity that receives groups perhapscampaign the most attention from media, the public, academia, and policymakers, is interest groups campaign contributions to parties and candidates as a and vehicle for inuencing policy. groups campaign attention from media, the public, academia, policymakers, is interest contributions to parties and candidates as a and vehicle for inuencing policy. groups campaign attention from media, the public, academia, policymakers, is interest contributions to parties and candidates as a vehicle for inuencing policy. How can these practices be avoided? In a democracy, elections form the most basic contributions to parties and candidates as a vehicle for inuencing policy. How can these practices be avoided? In a democracy, elections form the most basic contributions to parties and candidates as a vehicle for inuencing policy. How can these potential practices undue be avoided? In of a interest democracy, elections form the most basic safeguard against inuence groups through campaign money. How can these potential practices undue be avoided? In of a interest democracy, elections form the most basic safeguard against inuence groups through campaign money. How can these potential practices undue be avoided? In of a interest democracy, elections form the most basic safeguard against inuence groups through campaign money. If nancial support from special interest groups appears improper, voters may punish the safeguard against potential undue inuence of interest groups through campaign money. If nancial support from special interest groups appears improper, voters may punish the safeguard against potential undue inuence of interest groups through campaign money. If nancialby support from special interest groups appears improper, voters may punish the candidate voting against him. Of course, for this mechanism to be valid voters need to If nancialby support from special interest groups appears improper, voters may punish the candidate voting against him. Of course, for this mechanism to be valid voters need to If nancialby support from special interest groups appears improper, voters may punish the candidate voting against him. Of course, for this mechanism to be valid voters need to be well informed. Since the him. main Of source of political information to for voters is massneed media, candidate by voting against course, for this mechanism be valid voters to be well informed. Since the main source of political information for voters is mass media, candidate by voting against him. Of course, for this mechanism to be valid is voters need to be well informed. Since the main source of political information for voters mass media, the presence of a free, independent mass media is a key component of democratic political be well informed. Since the main source of political information for voters is mass media, the presence of a free, independent mass media is a key component democratic political be well informed. Since the main source of political information forof voters is mass media, the presence of a free, independent mass media is a key component of democratic political institutions. the presence of a free, independent mass media is a key component of democratic political institutions. the presence of a free, independent mass media is a key component of democratic political institutions. In this paper, I examine the role of mass media in countering special interest group institutions. In this paper, I examine the role of mass media in countering special interest group institutions. In this I paper, examine the the role extent of mass counteringsupport special for interest group inuence. do so I examining to media which in county-level candidates In this I paper, Iby examine the the role extent of mass media in counteringsupport special for interest group inuence. do so by examining to which county-level candidates In this I paper, Iby examine the the role extent of mass media in counteringsupport special for interest group inuence. do so examining to which county-level candidates to the United States Senate from the 1980 to 2002 varies as a function support of mediafor exposure and inuence. I do so by examining extent to which county-level candidates to the United States from the 1980 to 2002 a function support of mediafor exposure and inuence. I do so bySenate examining extent to varies which as county-level candidates to the United States Senate from 1980 to 2002 varies as a function of media exposure and candidates campaign nancefrom proles. to the United States Senate 1980 to 2002 varies as a function of media exposure and candidates campaign nance proles. to the United States Senate from 1980 to 2002 varies as a function of media exposure and candidates campaign nance proles. To measure media nance exposure I rely on media market structure. I compare Senate election candidates campaign proles. To measure media nance exposure I rely on media market structure. I compare Senate election candidates campaign proles. To measure media exposure I rely on media market structure. I compare election results for counties located in in-state TV markets (markets centered withinSenate a given state) To measure media exposure I rely on media market structure. I compare Senate election results for counties located in in-state TV markets (markets centered within a given state) To measure media exposure I rely on media market structure. I compare Senate election results for counties located in in-state TV markets (markets centered within a given state) with those located in out-of-state TV markets (markets centered in a city outside of a given results for counties located in in-state TV markets (markets centered within a given state) with those located in out-of-state TV markets (markets centered in a city outside of a state) given results for counties located in in-state TV markets (markets centered within a given with those located in out-of-state TV markets (markets centered in a city outside of a given state). Voters in out-of-state markets receive (markets much lesscentered television coverage of their states with those located in out-of-state TV markets in a city outside of a given state). Voters in out-of-state markets receive much less television coverage of their states with those located in out-of-state TV markets (markets centered in coverage a city outside of astates given state). Voters in out-of-state markets receive much less television of their politics than voters covered by in-state mediamuch markets et al. of (2006). Thus, state). Voters in out-of-state markets receive less Ansolabehere television coverage their states politics than voters covered by in-state media markets Ansolabehere et al. (2006). Thus, state). Voters in out-of-state markets receive much less Ansolabehere television coverage their states politics than voters covered by in-state media markets et al. of (2006). Thus, comparing the behavior of in-state market voters with voters in out-of-state media markets politics than voters covered by in-state media markets Ansolabehere et al. (2006). Thus, comparing the behavior of in-state market voters with voters in out-of-state media markets politics than voters covered by in-state media markets Ansolabehere et al. media (2006). Thus, comparing the behavior of in-state market voters with voters in out-of-state markets provides one possible approach to examine the role of mass media. comparing the behavior of in-state market voters with voters in out-of-state media markets provides one possible approach to examine the role of mass media. comparing the behavior of in-state market voters with voters in out-of-state media markets provides possible approach to examine the role of proxy mass media. I use one campaign contribution data to construct a for capture of politicians by provides one possible approach to examine the role of mass media. I use one campaign contribution data to construct for capture of politicians by provides possible approach to examine the role a of proxy mass media. I use campaign data to on construct a proxy for Political capture of politicians by special interests. In contribution particular, I use data contributions from Action Committees I use campaign contribution data to construct a proxy for Political capture of politicians by special interests. In contribution particular, I use data on contributions from Action Committees I use campaign data to construct a proxy for capture of politicians by special interests. In particular, I use data on contributions from Political Action Committees (PACs) to candidates to build Herndahl concentration indices, at both Action the PAC and more special interests. In particular, I use data on contributions from Political Committees (PACs) to candidates to build Herndahl concentration indices, at both Action the PAC and more special interests. In particular, I use data on contributions from Political Committees (PACs) toindustry candidates to build Herndahl both the PAC and more aggregate levels. The idea behind concentration this approach indices, is that aat more concentrated pattern (PACs) to candidates to build Herndahl concentration indices, at both the PAC and more aggregate industry levels. The idea behind concentration this approach indices, is that aat more concentrated pattern (PACs) to candidates to build Herndahl both the PAC and more aggregate industry levels. The idea behind this approach is that a more concentrated pattern of contributions (i.e., a high Herndahl index) is a good proxy for the extent to which a aggregate industry levels. The idea behind this approach is that a more concentrated pattern of contributions (i.e., a high Herndahl index) is a good proxy for the extent to which a aggregate industry levels. The idea behind this approach is that a more concentrated pattern of contributions (i.e., a high Herndahl index) is a good proxy for the extent to which a candidate is captured by narrow interest groups. Put di erently, candidates with more of contributions (i.e., a high Herndahl index) is a good proxy for the extent to which a candidate is captured by narrow interest groups. Put di erently, candidates with more of contributions (i.e., a high Herndahl index) is a good proxy for the extent to which a candidate is captured by narrow interest groups. Put di erently, candidates with more dispersed contribution sources are preferred by voters because they are less susceptible to candidate is captured by narrow interest groups. Put di erently, candidates with more dispersed contribution sources are preferred by voters because they are less susceptible to candidate is captured by narrow interest groups. Put dierently, candidates with more dispersed contribution sources are preferred voters because are less susceptible to capture by one of the (many) interest groups by sponsoring them. they dispersed contribution sources are preferred by voters because they are less susceptible capture by one of the (many) interest groups by sponsoring them. they are less susceptible to dispersed contribution sources are preferred voters because to capture by one of the (many) interest groups sponsoring them. capture by one of the (many) interest groups sponsoring them. capture by one of the (many) interest groups sponsoring them. 1 1 1 1 1
3

1 Introduction 1 Introduction 1 Introduction 1 Introduction 1 Introduction Interest groups pursue dierent strategies to inuence policy. These include gathering inforInterest groups pursue dierent strategies to inuence policy. These include gathering infor-

My main nding is that an increase in the concentration of campaign contributions leads My main nding is that an increase in the concentration of campaign contributions leads My main nding is that an voters increase in the of campaign contributions leads to a di erential response from across diconcentration erent media markets. The estimates suggest My main nding is that an voters increase in the of campaign contributions leads to a di erential response from across diconcentration erent media markets. The estimates suggest to a di erential response from voters across di erent media markets. The estimates suggest that the share ofresponse the two-party vote for an incumbent with a Herndahl index one-standard to a di erential from voters across dierent media markets. The estimates suggest that the share of the two-party vote for an incumbent with a Herndahl index one-standard that the share of the two-party votebe forabout an incumbent with a Herndahl index one-standard deviation higher than average will 0 . 5 to 1 . 5 percentage points lower in in-state that the share of the two-party votebe forabout an incumbent a Herndahl index one-standard deviation higher than average will 0.5 to 1.with 5 percentage points lower in in-state deviation higher than average will be about 0 . 5 to 1 . 5 percentage points lower in in-state counties relative to out-of-state counties, where the candidate receives less coverage. deviationrelative higher to than average will be about 0.5the to candidate 1.5 percentage points in in-state counties out-of-state counties, where receives less lower coverage. counties relative to out-of-state counties, where the candidate receives less coverage. I address the concern that these results may be driven by the fact that counties in outcounties relative out-of-state counties, where candidate receives less coverage. I address the to concern that these results maythe be driven by the fact that counties in outI address the concern these results mayin be driven by the fact counties in outof-state dominated media that markets are dierent other dimensions to that in-state counties. In I address the concern these results mayin be driven by the fact counties in outof-state dominated media that markets are dierent other dimensions to that in-state counties. In of-state dominated media markets are di erent in other dimensions to in-state counties. In particular, the negative impact of are the di interaction between campaign contribution concenof-state dominated media markets erent in other dimensions to in-state counties. In particular, the negative impact of the interaction between campaign contribution concenparticular, the negative impact of the interaction between campaign contribution concentration and increased media exposure present inbetween specications with contribution county xed eects, particular, negative impact of the is interaction campaign tration andthe increased media exposure is present in specications with county xed conceneects, tration and increased media exposure is present in specications with county xed eects, which rely solely on variation in the Herndahl indices across time within the same county to tration and increased media exposure is present in specications with county xed eects, which rely solely on variation in the Herndahl indices across time within the same county to which rely solely on variation in the Also, Herndahl indices acrosscounties time within the same county to estimate the coe cient of interest. I focus on in-state that are geographically which rely solely on variation in the Also, Herndahl indices acrosscounties time within county to estimate the coe cient of interest. I focus on in-state that the are same geographically estimate the coe cient of counties. interest. Also, Idemonstrating focus on in-state counties that are geographically contiguous to out-of-state After that they are similar to out-of-state estimate the cient of counties. interest. Also, focus on in-state counties that are geographically contiguous tocoe out-of-state After Idemonstrating that they are similar to out-of-state contiguous to out-of-state counties. After demonstrating that they are similar to out-of-state counties on a number of observable characteristics, I reestimate the e ect using solely these contiguous counties. After demonstrating that they are similar to out-of-state counties onto a out-of-state number of observable characteristics, I reestimate the e ect using solely these counties on a number of observable characteristics, I reestimate the e ect using solely these observations and nd similar results. counties on a number of observable characteristics, I reestimate the e ect using solely these observations and nd similar results. observations and nd similar results. I perform and an additional robustness with a placebo test designed to verify whether observations nd similar results. check I perform an additional robustness check with a placebo test designed to verify whether I perform an additional robustness check with a placebo test designed to verify trend whether the Imovement from in-staterobustness to out-of-state captures a general geographic in perform an additional checkcounties with a placebo test designed to verify trend whether the movement from in-state to out-of-state counties captures a general geographic in the movement from in-state to out-of-state that counties captures with a general geographic trend in the response to concentrated contributions is unrelated media exposure. To this the in-state to out-of-state that counties captures with a general geographic in the movement response tofrom concentrated contributions is unrelated media exposure. trend To this the response to concentrated contributions that is unrelated withto media exposure. To this end, I dene the in-state counties that border out-of-state counties be the (placebo) out-ofthe to concentrated contributions that is unrelated withto media exposure. this end,response I dene the in-state counties that border out-of-state counties be the (placebo) To out-ofend, I dene the in-state counties that border out-of-state counties to be the (placebo) out-ofstate counties. I compare to adjacent in-state counties state.out-ofSince end, dene theThen, in-state counties them that border out-of-state counties toin bethe thesame (placebo) stateIcounties. Then, I compare them to adjacent in-state counties in the same state. Since state counties. Then, I compare them to adjacent in-state counties in the same state. Since these sets of counties both in-state with good access to information, state Then,are I compare them counties to adjacent in-state counties in the same there state.should Since these counties. sets of counties are both in-state counties with good access to information, there should these sets of counties are both in-state counties with good access to information, there should be no correlation between increases in concentration andaccess moving across county categories. these of counties are both in-state counties with good to information, there should be nosets correlation between increases in concentration and moving across county categories. be no correlation between increases in concentration and moving across county categories. Indeed, I nd no correlation between increases in concentration and moving across county be no correlation between increases inincreases concentration and moving across county categories. Indeed, I nd no correlation between in concentration and moving across county Indeed, I nd no correlation between increases in concentration and moving across county categories. Indeed, I nd no correlation between increases in concentration and moving across county categories. categories. As noted, I use the pattern of campaign contributions as the main proxy for political capcategories. As noted, I use the pattern of campaign contributions as the main proxy for political capAs noted, I use the pattern of campaign as measure the mainfor proxy forrepresentative political capture by interest groups. Conveniently, I cancontributions build a simple every As noted, I use the pattern of campaign as measure the mainfor proxy forrepresentative political capture by interest groups. Conveniently, I cancontributions build a simple every ture by interest groups. Conveniently, I can build a simple measure for every representative based on campaign nance data alone. IStill, it is important to take afor step back and ask how ture by interest groups. Conveniently, can build a simple measure every representative based on campaign nance data alone. Still, it is important to take a step back and ask how based on campaign nance interest data alone. Still, it is important to take step back treatment and ask how voters think about special money. Historical anecdotes anda the press of based on campaign nance data alone. Still, it is important to take a step back and ask how voters think about special interest money. Historical anecdotes and the press treatment of voters think about special interest money. Historical anecdotes and the press treatment of campaign money seems to indicate a clear answer: most voters dislike campaigns fueled by voters think about special Historical anecdotes and the press treatment of campaign money seems to interest indicate money. a clear answer: most voters dislike campaigns fueled by campaign money seems to indicate a clear answer: most voters dislike campaigns fueled by interest money. answer is not obvious. When special interest groups have by an campaign moneyHowever, seems to the indicate a clear answer: most voters dislike campaigns fueled interest money. However, the answer is not obvious. When special interest groups have an interest money. However, the answer is not obvious. When special interest groups have an informational advantage, contributions may signal candidate results suginterest money. However, their the answer is not obvious. When special quality. interest My groups have an informational advantage, their contributions may signal candidate quality. My results suginformational advantage, their contributions may signal candidate quality. My results suggest, however, that voterstheir assess interest group money Also, I present evidence informational contributions may signalnegatively. candidate quality. My results suggest, however, advantage, that voters assess interest group money negatively. Also, I present evidence gest, however, that voters assess interest group money negatively. Also, I present evidence that they do so in a relatively sophisticated way. Specically, voters punish total evidence interest gest, however, that voters assess interest group money negatively. Also, I present that they do so in a relatively sophisticated way. Specically, voters punish total interest that they do less so in a relatively sophisticated way. voters punish total group money than they punish concentration of Specically, interest group money from a few interest sectors. that they do less so in a relatively sophisticated way. voters punish total group money than they punish concentration of Specically, interest group money from a few interest sectors. group money less thanthe they punish concentration of interest group money from a few interest sectors. These results support idea that voters punish capture of politicians by narrow group thanthe they punish interest group money from a few interest sectors. These money results less support idea that concentration voters punish of capture of politicians by narrow These results support the idea that voters punish capture of politicians by narrow interest These results support the idea that voters punish capture of politicians by narrow interest 2 groups. 2 A related point is whether the suggested2 2mechanism demands an unrealistic degree of knowledge on the part of voters. But even if individuals do not know the approximate 4 pattern of campaign contributions (a reasonable assumption for most voters), a candidate

groups. groups. groups. A related point is whether the suggested mechanism demands an unrealistic degree of A related point is whether the suggested mechanism demands an unrealistic degree of A related is whether the suggested demands unrealistic degree of knowledge on point the part of voters. But even mechanism if individuals do not an know the approximate knowledge on the part of voters. But even if individuals do not know the approximate groups. knowledge on the part of voters. (a But even if individuals do not know the approximate pattern of campaign contributions reasonable assumption for most voters), a candidate pattern of campaign contributions (a reasonable assumption for most voters), a candidate A related point is whether the suggested mechanism demands an unrealistic degree of pattern of campaign contributions (a reasonable assumption for most voters), a journalists candidate with a more concentrated pattern of contributions is prone to being pinpointed by with a moreon concentrated pattern contributions is prone to do being pinpointed journalists knowledge the part of voters.of even if individuals not know the by approximate with a opponent more concentrated pattern of But contributions to being pinpointed by journalists or his as potentially captured. Such is a prone candidate is likely to gure in ads and or his opponent as potentially captured. Such assumption a candidate for is likely to gureain ads and pattern of campaign contributions (a reasonable most voters), candidate or hisas opponent potentially Such a contribute candidate is to gure in ads and news linked toas the economic captured. sectors that heavily to likely his campaign. news as linked to the economic sectors that heavily contribute to his campaign. with a more concentrated pattern of contributions is prone to being pinpointed by journalists news linked to the sectors that heavily contribute his campaign. Toas verify that thiseconomic is the case, I conduct automated news to searches for a sample of US To verify that this is the case, I conduct automated news searches for a sample of and US or his asthis potentially captured. Such a candidate is likely to gure in ads To opponent verify counting that is the case, I conduct automated news searches for a sample of For US newspapers, how frequently are Senate candidates featured in news stories. newspapers, counting how frequently are Senate candidates featured in news stories. For news as linked to the economic sectorsare that heavily contribute to his campaign. newspapers, counting how frequently in news stories. For each candidate, I also count the number of Senate stories candidates that discussfeatured PAC money in the campaign each candidate, I also count the number of stories that discuss PAC money in the campaign To verify that this is thethe case, I conduct automated news PAC searches for in a sample of US each candidate, I also number of stories that money the campaign (PAC stories). Thecount results are reassuring about thediscuss validity of the approach. There is (PAC stories). The how results are reassuring about the validity of thein approach. There is newspapers, counting frequently are Senate candidates featured news stories. For (PAC stories). The results are reassuring the of validity of the for approach. There is a positive and signicant correlation between about the share PAC stories a candidate and a positive and signicant correlation between the share of PAC stories for a candidate and each candidate, I also count the number of stories that discuss PAC money incandidate the campaign a positive and signicant between the share of PAC stories for a and the candidates Herndahlcorrelation concentration index. This suggests that the concentration indices the candidates Herndahl concentration index. This suggests that the concentration indices (PAC stories). The results are reassuring about validity ofthe theconcentration approach. There is the proxy candidates Herndahl concentration index. Thisthe suggests that indices do for media exposure of campaign contribution proles. An additional important do proxy for media exposure of campaign contribution proles. An additional important a andmedia signicant correlation between the share of PAC stories for a candidate and dopositive proxy for exposure of campaign contribution proles. An additional important advantage of this exercise is that it helps address the concern that campaign nance conadvantage of this exercise concentration is that it helps address the concern that campaign nance conthe candidates Herndahl index. This suggests that the concentration indices advantage of this exercise is that it helps address the concern that campaign nance concentration captures some other omitted candidate characteristics which dierentially a ect centration captures some other omitted candidate characteristics which di erentially a ect do proxy for media exposure of campaign contribution proles. An additional important centration captures somecounties other omitted which dierentially a ect their performance across (e.g. if candidate candidatescharacteristics with higher Herndahls are more or less their performance across counties (e.g. if candidates with higher Herndahls are more or less advantage of this across exercise is that (e.g. it helps address the concern that campaign nance their performance counties candidates with higher Herndahls orconless popular). Indeed, by computing the if number of PAC stories as a share of are the more candidates popular). Indeed, by computing the number of PAC stories as a share of the candidates centration captures some other omitted candidate characteristics which erentially aect popular). by I computing the number of PAC stories as a sharedi of the candidates total numberIndeed, of stories, directly normalize by how popular or appealing the candidate is in total number of stories, directly normalize by how popular or appealing theare candidate in their performance acrossI counties (e.g. if candidates with higher Herndahls more oris less total number of stories, I directly normalize by how popular or appealing the candidate is in the media. And, more compellingly, there is in fact no correlation between the candidates the media. And, more compellingly, there is in fact no correlation between the candidates popular). Indeed, by compellingly, computing the number PAC as a between share of the the candidates candidates 1 the media. and And, more there is in of fact no stories correlation Herndahl his total number of news stories 1. . Herndahl and his total number of news stories 1 total number of his stories, directly of normalize by how popular or appealing the candidate is in . of Herndahl and totalInumber news stories Moreover, additional results highlight the role information and the robustness of the Moreover, additional results highlight the role of information and the robustness of the the Moreover, media. And, more compellingly, there is in fact no correlation between the candidates additional the role of information and the robustness of the main results. Given that results a large highlight theoretical and empirical literature suggests an association main results. Given that a large theoretical and 1 empirical literature suggests an association Herndahl and his total of news stories main results. Given that number a large theoretical andI. empirical literature suggests an association between better information and higher turnout, examine whether turnout is higher in-state between better information and higher turnout, I examine whether turnout is higher in-state Moreover, additional results highlight the role of information and the robustness of the between better information and highercounties turnout,do I examine whether turnout higher in-state than out-of-state. I nd that in-state have larger turnout thanis their (otherwise than out-of-state. I nd that in-state counties do have larger turnout than their (otherwise main out-of-state. results. Given that that a large theoretical and empirical literature suggests an association than I nd in-state counties have larger turnout than their (otherwise similar) contiguous, out-of-state neighbors. Mydo estimates imply a 1.7 to 2 percentage point similar) contiguous, out-of-state neighbors. My estimates imply a 1 . 7 to 2 percentage point between better information and higher turnout, I examine whether turnout is higher in-state similar) contiguous, out-of-state neighbors. My estimates imply a 1.7 toon 2 percentage point higher turnout in in-state counties. Furthermore, a regression of turnout a dummy for the higher turnout in in-state counties. Furthermore, a regression of turnout on a dummy for the than out-of-state. I nd that in-state counties doahave larger of turnout than their (otherwise higher turnout in in-state counties. Furthermore, regression turnout on a dummy for the placebo out-of-state counties versus other in-state neighbors nds no eect, which conrms placebo counties versus other in-state neighbors nds no ect, which conrms similar) out-of-state contiguous, out-of-state neighbors. My estimates imply a1 .7 e to 2 percentage point placebo out-of-state counties versus neighbors nds no e ect, which conrms the hypothesis that information (notother otherin-state dierences resulting from moving toward out-ofthe hypothesis that information (not other di erences resulting from moving toward out-ofhigher turnout in in-state counties. Furthermore, a regression of turnout on a dummy for the the hypothesis information other dierences resulting from moving toward out-ofstate counties) that drives dierences (not in turnout. state counties) drives dierences in turnout. placebo out-of-state counties versus other in-state neighbors nds no eect, which conrms state counties) drives dierences in turnout. This set of results opens the discussion concerning whether the concentration of campaign This set of results opens the discussion concerning whether the concentration of campaign the This hypothesis that information (not other dierences resulting from moving toward out-ofset of results opens the discussion concerning whether the concentration of campaign 1 only drawback exercise that online newspaper archives are typically available only since state counties) drivesof dithis erences inis 1 The The only drawback of this exercise isturnout. that online newspaper typically available only since the 1 1990s (at best). Hence, I cannot systematically code the sharearchives of PAC are stories as an alternative measure The only drawback of this exercise is that online newspaper archives are typically available only since the 1990s (at best). Hence, I cannot systematically code the share of PAC stories as an alternative measure This set of results opens the discussion concerning whether the concentration of campaign of capture replicate my main specications with this variable. as indicated the correlations the 1990s (at to best). Hence, I cannot systematically code the share of However, PAC stories as an alternative measure
of capture to replicate my main with this of variable. However, as indicated the correlations where information is available lend specications support to the validity the approach. of capture to replicate my main specications with this variable. However, as indicated the correlations where 1 information is available lend support to the validity of the approach. The only drawback of this exercise is that online newspaper archives are typically available only since where information is available lend support to the validity of the approach. the 1990s (at best). Hence, I cannot systematically code the share of PAC stories as an alternative measure 3 this variable. However, as indicated the correlations of capture to replicate my main specications with 3 3 where information is available lend support to the validity of the approach.

higher turnout in in-state counties. Furthermore, a regression of turnout on a dummy for the placebo out-of-state counties versus other in-state neighbors nds no eect, which conrms the hypothesis that information (not other dierences resulting from moving toward out-ofstate counties) drives dierences in turnout. This set of results opens the discussion concerning whether the concentration of campaign nance money for Senate candidates has negative spillovers for fellow party members running 1 nance money for Senate candidates has negative spillovers for fellow party members running The only drawback of this exercise is that online newspaper archives are party typically availablerunning only since nance money for Senate candidates has negative spillovers for fellow members nance money for Senate candidates has negative spillovers for fellow party members running for other o ces. To testI this idea, I examine if in years where, for example, the Democratic the 1990s (at best). Hence, cannot systematically code the share of PAC stories as an alternative measure nance money for Senate candidates has negative spillovers for fellow party members running for other o ces.To To test this idea, examine in years where, for example, the Democratic for o ces. test this idea, II examine ifif in years where, for example, the Democratic for other o ces. To test this idea, I examine if in years where, for example, the Democratic of other capture to replicate my main specications with this variable. However, as indicated the correlations senator has a relatively high Herndahl, the Democratic Presidential (or House) candidate for other oa ces. To test high this idea, I examine if in years where, for example, the Democratic senator has relatively Herndahl, the Democratic Presidential (or House) candidate where information is available lend support to the validity of the approach. senator has a relatively high Herndahl, the Democratic Presidential (or House) candidate senator relatively high Herndahl, the Democratic Presidential House) candidate tends tohas do a worse in in-state counties relative to out-of-state counties.(or I nd no important senator has a relatively high Herndahl, the Democratic Presidential (or House) candidate tends to do worse in in-state counties relative to out-of-state counties.I I nd no important tends to do worse in in-state counties relative to out-of-state counties. nd no important 2 counties relative to out-of-state counties. I nd no important tends to do worse in in-state spillover ects of this sort2 . counties relative 3 to out-of-state counties. I nd no important tends to e do worse in in-state 2 2. spillover e ects of this sort .2 . spillover e ects of this sort spillover e ects of this sort This non-result, however, rms that the main ndings are not driven by partisan spillover eects of this sort . rea Thisnon-result, non-result, however, rea rmsthat thatthe the mainndings ndingsare arenot notdriven drivenby by partisan This however, rea rms This non-result, however, rea rms that themain main ndings are not partisanship driven bypartisan partisan trends. Indeed, a potentially serious objection to my ndings is that moves This Indeed, non-result, however, rea rms that the to main ndings are not partisanship driven by partisan trends. a potentially serious objection my ndings is that moves trends. Indeed, a potentially serious objection to my ndings is that partisanship moves trends. Indeed, a potentially serious objection to di my ndings is that partisanship moves in dierent directions in dierent counties across erent years. Unfortunately, I cannot trends. Indeed, a potentially serious objection to my ndings is that partisanship moves indi di erentdirections directionsin indi di erentcounties countiesacross acrossdi di erentyears. years. Unfortunately, Unfortunately,I Icannot cannot in erent erent erent in di erent directions in di erent counties across di erent years. Unfortunately, I cannot control for adirections full set of in county-year xed eects, since variation use to estimate key in dierent dierent counties across dithe erent years. I Unfortunately, I the cannot control for a full set of county-year xed eects, ects, since the variation I use to estimate the key control for a full set of county-year xed e since the variation I use to estimate the key control for a full set of this county-year xed ethe ects, since the variation I use to estimate the key interaction terms is at level. However, regressions for Presidential and Congressional control for a full set of county-year xed e ects, since the variation I use to estimate the key interaction terms at this level.However, However, the regressions for Presidential and Congressional interaction terms isis at this level. the regressions for Presidential and Congressional interaction terms is at this level. However, the regressions for Presidential and Congressional races can interms fact be thought of as placebo tests to rule out this interpretation. If important interaction is at this level. However, the regressions for Presidential and Congressional races can in fact be thought of as placebo tests to rule out this interpretation. If important races can in fact be thought of as placebo tests to rule out this interpretation. If important races cantrends in factwere be thought as placebo tests to rule with out this important partisan driving of the results, specications voteinterpretation. share of fellow If party memraces can in fact be thought of as placebo tests to rule out this interpretation. If important partisan trends were driving the results, specications with vote share of fellow party mempartisan trends were driving the results, specications with vote share of fellow party mempartisan trends were driving theshould results, specications with vote share of fellow party of members as the dependent variable mirror the e ects observed for the vote share the partisan trends were driving theshould results, specications with vote share of fellow party of members as the dependent variable mirror the e ects observed for the vote share the bers as the dependent variable should mirror the e ects observed for the vote share of the bers as the dependent variable should mirror the eects observed for the vote share of the incumbent. bers as the dependent variable should mirror the e ects observed for the vote share of the incumbent. incumbent. incumbent. In sum, the overall evidence presented in this paper supports the idea that mass media, incumbent. In sum, the overall evidence presented in this paper supports the idea that mass media, In sum, the overall evidence presented in this paper supports the idea that mass media, In sum, the overall evidence presented in this supports idea that mass media, by informing voters, may reduce the inuence ofpaper special interestthe groups in policy. Better In sum, the overall evidence presented in this paper supports the idea that mass media, by informing voters, may reduce the inuence of special interest groups in policy. Better by informing voters, may reduce the inuence of special interest groups in policy. Better by informing voters, may reduce the inuence of special interest groups in policy. Better access to mass media allows voters to react to potentially negative information about their by informing voters, may reduce the inuence of special interest groups in policy. Better access to mass media allows voters to react to potentially negative information about their access to mass media allows voters to react to potentially negative information about their access to mass media allows voters to react to potentially negative information about their candidates, and specically to the possible inuence of narrow interests in the about politicians access to mass media allows voters to react to potentially negative information their candidates, and specically to the possible inuence of narrow interests in the politicians candidates, and specically to the possible inuence ofof narrow interests in the politicians candidates, and specically to the possible inuence narrow interests in the politicians agenda. candidates, and specically to the possible inuence of narrow interests in the politicians agenda. agenda. agenda. The results also rearm the idea that patterns of special interest campaign contributions agenda. The results also rea rm the idea that patterns special interest campaign contributions The results also rea rm the idea that patterns ofof special interest campaign contributions The results also rea rm the idea that patterns of special interest campaign contributions signal political capture of representatives to voters. The fear of politicians capture has long The results also rea rm the idea that patterns ofThe special interest campaign contributions signal political capture of representatives to voters. fear of politicians capture has long signal political capture ofof representatives to voters. The fear ofof politicians capture has long signal political capture representatives to voters. The fear politicians capture long been prevalent in the United States politics and mass media, and some of the mosthas famous signal political capture of representatives to voters. The fear of politicians capture has long been prevalent in the United States politics and mass media, and some of the most famous been prevalent in the United States politics and mass media, and some ofof the most famous 3 been prevalent in the United States politics and mass media, and some the most famous political scandals have involved the alleged use ofmass contributions in exchange for most favors 3 . The been prevalent in have the United States politics and media, and some of the 3famous .The The political scandals involved the alleged use contributions in exchange for favors .3 political scandals have involved the alleged use ofof contributions in exchange for favors . The political scandals have involved the alleged use of contributions in exchange for favors political scandals have involved the alleged use of contributions in exchange for favors3 . The
I also in nd little evidence of other types of information spillovers. For example, it does not appear that voters out-of-state counties rely on partisan cues and punish the candidate of the party with a high Herndahl index in the neighboring state (from which they receive information), nor that they rely on that voters in out-of-state counties rely on partisan cues and punish the candidate of the party with a high Herndahl index in the neighboring state (from which they receive information), nor that they rely on that voters in out-of-state counties rely on partisan cues and punish the candidate of the party with a high Herndahl index in the neighboring state (from which they receive information), nor that they rely on comparisons between the campaign nance proles of candidates in their state with those in neighboring high Herndahl index in the neighboring state (from which they receive information), nor that they rely on comparisons between the campaign nance proles of candidates in their state with those in neighboring high Herndahl index in the neighboring state (from which they receive information), nor that they rely on comparisons between the campaign nance proles of candidates in their state with those in neighboring states. comparisons between the campaign nance proles of candidates in their state with those in neighboring states. 3 comparisons between the campaign nance proles of candidates in their state with those in neighboring states. states. 3 The topic has been at the center of the policy debate since at least 1957, when allegations of improper 3 3 The topic has been at the center of the policy debate since at least 1957, when allegations of improper states. The topic has been atat the center ofof the policy debate since atat least when allegations ofof improper inuence of this sort led Senate to establish the Special Committee to 1957, Investigate Political Activities, Lobhas been the center the policy debate since least 1957, when allegations improper 3 The topic inuence of this sort led Senate to establish the Special Committee to Investigate Political Activities, LobThe topic has been at the center of the policy debate since at least 1957, when allegations of improper inuence of this sort led Senate to establish the Special Committee to Investigate Political Activities, Lobbying, and Campaign Contributions. A famous earlier attempt to control interest money was President inuence of Campaign this sort led Senate to establish the Special Committee to Investigate Political Activities, Lobbying, and Contributions. A famous earlier attempt to control interest money was President inuence of this sort led Senate to the Special Committee to Investigate Political Activities, Lobbying, and Campaign Contributions. AA famous earlier toto control interest money was President Theodore Roosevelts argument, inestablish 1905, for a ban on attempt all political contributions by corporations, and his bying, and Campaign Contributions. famous earlier attempt control interest money was President Theodore Roosevelts argument, in 1905, for a ban on all political contributions by corporations, and his bying, and Campaign Contributions. A famous earlier attempt to control interest money was President Theodore Roosevelts argument, in 1905, for a ban on all political contributions by corporations, and his call, in 1907, for public nancing of federal candidates via candidates political parties. In 1989, the Keating Theodore Roosevelts argument, infederal 1905, candidates for a ban on all political contributions by corporations, and his call, in 1907, for public nancing of via candidates political parties. In 1989, the Keating Theodore Roosevelts argument, in 1905, for a senators banvia on all political contributions by corporations, and his call, inin 1907, for public nancing ofof federal candidates candidates political parties. In 1989, Keating Five scandal (so named because it involved ve accused of improperly intervening in the 1987 on behalf call, 1907, for public nancing federal candidates via candidates political parties. In 1989, the Keating Five scandal (so named because it involved ve senators accused of improperly intervening in 1987 on behalf call, in 1907, for public nancing of federal candidates via candidates political parties. In 1989, the Keating Five scandal (so named because it involved ve senators accused of improperly intervening in 1987 on behalf of Charles H. Keating, Jr.) attracted public attention. Keating provided substantial political contributions Five scandal (so named because it involved ve senators accusedprovided of improperly intervening in 1987 on behalf Charles H. Keating, Jr.) attracted public attention. Keating substantial political contributions Five scandal (so named because involved ve senators accused ofLoan improperly intervening in contributions 1987 on behalf ofof Charles H. Keating, Jr.) attracted public attention. Keating provided substantial political to each involved senator and was it chairman of Lincoln Savings and Association, an institution targeted of Charles H. Keating, Jr.) attracted public attention. Keating provided substantial political contributions each involved senator and was chairman of Lincoln Savings and Loan Association, an institution targeted of Charles H. Keating, Jr.) attracted public attention. Keating provided substantial political contributions toto each involved senator and was chairman of Lincoln Savings and Loan Association, an institution targeted in investigations by the Federal Home Loan Bank Board in the midst of the Savings and Loan crisis of the to investigations each involved senator and wasHome chairman ofBank Lincoln Savings and Loan Association, an institution targeted by the Federal Loan Board in the midst of the Savings and Loan crisis of the to each involved senator and was chairman of Lincoln Savings and Loan Association, an institution targeted inin investigations by the Federal Home Loan Bank Board in the midst of the Savings and Loan crisis of the late 1980s and early 1990s. in investigations by the Federal Home Loan Bank Board in the midst of the Savings and Loan crisis of the late 1980s and early 1990s. in investigations by1990s. the Federal Home Loan Bank Board in the midst of the Savings and Loan crisis of the late 1980s and early late 1980s and early 1990s. late 1980s and early 1990s.
2 2 I also nd little evidence of other types of information spillovers. For example, it does not appear 2 2 I also nd little evidence of other types of information spillovers. For example, it does not appear II also nd evidence ofof other types ofof information spillovers. For example, it does not appear that voters in little out-of-state counties rely on partisan cues and punish the candidate of the party with a also nd little evidence other information spillovers. For example, it does not appear that2 voters in out-of-state counties relytypes on partisan cues and punish the candidate of the party with a

4 44 4 4 6

subject most recently hit the headlines in connection with the recent nancial meltdown. subject most recently hit the headlines in connection with the recent nancial meltdown. Analysts have emphasized the political force of Wall Street banks in Washington as the Analysts have emphasized the political force of Wall Street banks in Washington as the fundamental cause of the crisis. While certainly not their sole source of inuence, campaign fundamental of the While certainly not their sole source of nancial inuence,meltdown. campaign subject most cause recently hit crisis. the headlines in connection with the recent contributions are among the perceived sources of power for special interests. Johnson and contributions among the perceived of Wall power for special Johnson Analysts haveare emphasized the politicalsources force of Street banks interests. in Washington as and the Kwak (2010) express it clearly: Kwak (2010) cause express clearly: fundamental of it the crisis. While certainly not their sole source of inuence, campaign contributions are among the perceived sources of power for special Johnson and Financial sector money poured into the campaign war chestsinterests. of congressional Financial sector money poured into the campaign war chests of congressional Kwakrepresentatives (2010) express it clearly: (...) Campaign contributions and the revolving door between representatives (...) Campaign contributions and the revolving door between the private sector and government service gave Wall Street banks inuence in the private sector and government service gave Wall Street banks inuence in Financial sector money poured into the campaign war chests of congressional Washington, but their ultimate victory lay in shifting the conventional wisdom in Washington, but theirCampaign ultimate victory lay in shifting therevolving conventional wisdom in representatives (...) contributions and the door between their favor (...) Of course, when cracks appeared in the consensus, such as in the their favor (...) Of course, when cracks appeared theStreet consensus, such as in the the private sector and government service gave in Wall banks inuence in aftermath of the nancial crisis, the banks could still roll out their conventional aftermath of but the their nancial crisis, the banks could stillthe rollconventional out their conventional Washington, ultimate victory lay in shifting wisdom in weaponry campaign money and lobbyists (p. 5). weaponry campaign money and lobbyists (p. 5). their favor (...) Of course, when cracks appeared in the consensus, such as in the aftermath the nancial crisis, the banks before could still out their conventional Similar concernsofabout the inuence of banks and roll during the nancial crisis and Similar concerns about the inuence of banks before and during the nancial crisis and weaponry campaign money lobbyists (p. 5). subsequent bank reform arose in theand recent mid-term elections. To cite one example, one subsequent bank reform arose in the recent mid-term elections. To cite one example, one perspective noted that Spencer Bachus (R.-Ala), the man most likely to become chairman of perspective noted about that Spencer Bachus of (R.-Ala), the man most likelythe to become chairman of Similar concerns the inuence banks before and during nancial crisis and the House Financial Services Committee received well over a million dollars from political the House Financial Services received wellelections. over a million dollars political subsequent bank reform aroseCommittee in the recent mid-term To cite one from example, one action committees representing banks, insurance companies and auditors over the past two action committees representing banks, (R.-Ala), insurance the companies and auditors over the past two perspective noted that Spencer Bachus man most likely to become chairman of election cycles. And wasting no time, on Wednesday, reports the Financial Times, Bachus election cycles. And Services wasting Committee no time, on received Wednesday, reports Financial Bachus the House Financial well over a the million dollarsTimes, from political sent a letter to the Financial Stability Oversight Council (...) that reads as if dictated by sent a committees letter to therepresenting Financial Stability Oversight companies Council (...) that readsover as ifthe dictated by action banks, insurance and auditors past two bank lobbyists4 4. . bank lobbyists election cycles. And wasting no time, on Wednesday, reports the Financial Times, Bachus In short, in the US as in many countries there is a general perception that special interest In as in many countries there is a general perception that interest sent a short, letter in tothe theUS Financial Stability Oversight Council (...) that reads asspecial if dictated by groups can exercise to some degree the levers of power, using campaign contributions as a 4 groups can exercise . to some degree the levers of power, using campaign contributions as a bank lobbyists primary instrument to do so. Yet, there also exist key institutions, free media among them, primary instrument to as doin so. Yet,countries there also exist institutions, free media among them, In short, in the US many there iskey a general perception that special interest that exert control over such inuence. For instance, many argue that an active, informative that exert over For instance, many argue that an active, informative groups cancontrol exercise to such some inuence. degree the levers of power, using campaign contributions as a press reduced corruption in the Progressive Era5 5 . Discussing the period of antitrust reform . Discussing the period of antitrust press reduced corruption the Progressive primary instrument to doin so. Yet, there alsoEra exist key institutions, free media amongreform them, in the early XXth Century, Acemoglu and Robinson (2012) highlight the role of a free press in the early XXth over Century, (2012) highlight role of informative a free press that exert control such Acemoglu inuence. and For Robinson instance, many argue that the an active, to expose the excesses of Robber Barons as well as corruption in local and federal politics. 5 as corruption in local and federal politics. to expose the corruption excesses of in Robber Barons as Era well . Discussing the period of antitrust reform press reduced the Progressive Among the examples discussed, one of the most telling about the potential of a free press to Among the examples discussed, one of and the most telling about the potential of a press to in the early XXth Century, Acemoglu Robinson (2012) highlight the role offree a free press
begin carving up bank reform, by Andre Leonard. In Salon, November 4, 2010. 4 Republicans to expose the excesses of Robber Barons asby well as Leonard. corruption in local and federal politics. Republicans begin carving up bank reform, Andre In Salon, November 4, 2010. http://www.salon.com/technology/how the world works/2010/11/04/spencer bachus wall streets http://www.salon.com/technology/how the world works/2010/11/04/spencer bachus wall streets Among the examples discussed, one of the most telling about the potential of a free press to man in washington/index.html
4

man 5 in washington/index.html Along these lines, Glaeser et al. (2004) argue that the US newspaper industry between 1870 and 1920, 5 4 Along these lines, Glaeser et al. (2004) argue that the US newspaper industry between 1870 and 1920, Republicans carving upbecame bank reform, by Andre Leonard. In Salon,partly November 4, the 2010. which expanded itsbegin circulation and more informative and less partisan, caused decline of which expanded its circulation and Moreover, became less partisan, partly caused the decline of the more world works/2010/11/04/spencer bachus wall streets http://www.salon.com/technology/how political corruption in that period. the informative entry of USand daily newspapers from 1869-1928 produced a political corruption in that period. Moreover, the entry of US daily newspapers from 1869-1928 produced a man washington/index.html more in active electorate, increasing voter turnout Gentzkow et al. (2009). 5 active electorate, increasing voter turnout Gentzkow et al. (2009). more Along these lines, Glaeser et al. (2004) argue that the US newspaper industry between 1870 and 1920, which expanded its circulation and became more informative and less partisan, partly caused the decline of 5 of US daily newspapers from 1869-1928 produced a political corruption in that period. Moreover, the entry 5 more active electorate, increasing voter turnout Gentzkow et al. (2009).

5
7

that exert control over such inuence. For instance, many argue that an active, informative press reduced corruption in the Progressive Era5 . Discussing the period of antitrust reform in the early XXth Century, Acemoglu and Robinson (2012) highlight the role of a free press to expose the excesses of Robber Barons as well as corruption in local and federal politics. Among the examples discussed, one of the most telling about the potential of a free press to 4 counter special interest inuence is reform, that of famous Ira November Tarbell. Tarbell Republicans begin carving up bank Andre muckraker Leonard. In Salon, 4, 2010. wrote counter special interest inuence is that of by famous muckraker Ira Tarbell. Tarbell wrote counter special interest inuence is that of famous muckraker Ira Tarbell. Tarbell wrote http://www.salon.com/technology/how the world works/2010/11/04/spencer wall streets the History of the Standard Oil Company, which played a key role bachus in moving public opinion the History of the Standard Oil Company, which played a key role in moving public opinion man in washington/index.html This paper this line research by studying the rolerole of free media public on improving the History of follows the Standard Oilof Company, which played aOil. key in moving opinion 5 against Rockefeller hiset business interests in Standard Along these lines, and Glaeser al. (2004) argue that US newspaper against Rockefeller and his business interests inthe Standard Oil. industry between 1870 and 1920, political accountability, and in became particular in changing theless electoral for certain types against Rockefeller and hisand business interests in Standard Oil. which expanded its circulation more informative partisan, partly caused the decline of As Acemoglu and Robinson (2012) note, the US and experience in support the rst half of the 20th As Acemoglu and Robinson (2012) the note, the US experience in the rst half ofproduced the 20th political corruption in that period. Moreover, entry of US daily newspapers from 1869-1928 a of candidates. However, unlike previous work, it emphasizes the role of half mass in As Acemoglu and Robinson (2012) the US experience in the rst ofmedia the 20th century emphasizes that free media is anote, key component of the set of inclusive institutions more active electorate, increasing voter turnout Gentzkow et al. (2009). century emphasizes that free media is a key component of the set of inclusive institutions countering special interest group inuence. In terms of the the dior erences century emphasizes that free media is a key component of empirical the set ofstrategy, inclusive institutions that determine economic success. Absent these institutions, under absolutism under that determine economic success. Absent these institutions, under absolutism or under between in-state and out-of-state counties were rst studied by Ansolabehere et al. (2006) that determine economic success. Absent these under absolutism or under 5 notinstitutions, dictatorships, the US public probably would have mobilized against the power and dictatorships, the US public probably would not have mobilized against the power and to examine the implications of television for the incumbency advantage in the US. dictatorships, the US public probably would not have mobilized against the power and abuses of Robber Barons and their trusts. abuses of Robber Barons and their trusts. The paper is also related tosupports a trusts. vast empirical literature on campaign abuses Robber Barons and their Theof evidence in this paper this view and thattheoretical mass media, by informing voters, The evidence in this paper supports this view that mass media, by informing voters, contributions. Much of the theoretical work assumes that there are informed voters who The evidence in this paper supports this view that mass media, by informing voters, constitutes a vital component of the inclusive institutions which promote political and constitutes a vital component of the inclusive institutions which promote political and vote for candidates based on their policy positions, and uninformed voters who can be constitutes a vital component of the institutions which promote political and economic development by reducing theinclusive inuence of special interest groups in policy. economic development by reducing the inuence of special interest groups in policy. swayed bydevelopment campaign advertising. Funds for advertising, provided bygroups interest groups, economic by reducing the inuence of special interest in policy. depend upon the positions taken by the candidates, and these positions take their implications for 1.1 Related literature 1.1 Related contributions and votes into account (Morton and Cameron (1992) oer an early review)8 . 1.1 Related literature literature This paper is related to several strands of literature, most notably, to the relatively recent While this work incorporates the response of voters to overall campaign expenditures, less This paper is related to several strands of literature, most notably, to the relatively recent This paper is related to several strands of literature, most notably, to the relatively recent but fast-growing economics on the political economy of mass media (see This Prat research investigates how theliterature sources of money raised by politicians inuence voters. but fast-growing economics literature on the political economy of mass media (see Prat but economics literature onin the political of the mass media (seemedia Prat and fast-growing Strmberg (2010)). Aapproach central topic this line assumes of economy research role of free occurs partly because the described above thatis voters do and Strmberg (2010)). A central topic in this line of research isuninformed the role of free media 6 and Strmberg (2010)). A central topic in this line of research is the role of free media Severalifempirical show in a ecting policy expectations. and improving political accountability not have rational As emphasized by Coate they did,papers they could 6 . (2004), in a ecting policy and improving political accountability 6 . Several empirical papers show . Several empirical papers show in a ecting policy improving political accountability how availability of and information voters and aects Str omberg (2004b) realize that a party involved inempowers advertising must distort itspolicy policy(e.g. platform to obtain the how availability of information empowers voters and aects policy (e.g. Str omberg (2004b) how availability of information empowers voters and a ects policy (e.g. Str o mberg (2004b) and Besley and Burgess (2002)). Snyder and Str ombergs (2010) work, which estimates the necessary funds, and switch their votes against the advertised party. Similarly, the empirical and Besley and Burgess (2002)). Snyder and Str ombergs (2010) work, which estimates the and Besley and coverage Burgess (2002)). Snyder and Str ombergs actions, (2010) work, which closely estimates the impact offocused press on citizen knowledge, politicians andon policy, relates work has mostly on the e ect of gross campaign expenditures electoral outcomes impact of press coverage on citizen knowledge, politicians actions, and policy, closely relates impact of pressThe coverage on exploit citizen knowledge, politicians actions, and policy, closely relates to this paper. authors the match, or congruence between markets and in altering policy positions or securing favors (see Stratmann (2005)newspaper for a review). to this paper. The authors exploit the match, or congruence between newspaper markets to this The authors exploit the match, or congruence between newspaper markets and USpaper. congressional districts. Varying congruence causes news coverage oftheir politics to vary More recent work reviewed Prat (2006) assumes that voters update beliefs raand US congressional districts. in Varying congruence causes news coverage of politics to vary and US congressional districts. Varying congruence causes news coverage of politics to vary across districts, unlike other measures media availability, does not directly aect tionally. Privatebut campaign nance creates of a trade-o between a it policy distortion and an across districts, but unlike other measures of media availability, it does not directly aect across districts, but unlike other measures of supportive media availability, it does not directly aect key outcome variables of interest. Results are of the role of media in facilitating informational benet. In equilibrium, qualied candidates more contributions than key outcome variables of interest. Results are supportive ofreceive the role of media in facilitating 7 interest. Results are supportive of the role of media in facilitating key outcome variables of . political accountability unqualied candidates. 7 However, candidates may need to distort their policy choices (away political accountability 7. political accountability . from 6 voters interests) in order to attract private donations. When voters are aware of this, 6 Several theoretical contributions imply a role of media in countering special interest group inuence. Several theoretical contributions implytheory, a role of media in countering special interest group inuence. the amounts patterns of campaign contributions may inuence their voting decision. For 6 instance, inand Str ombergs (2001, 2004a) mass media counteracts special interest group inuence Several theoretical contributions implytheory, a role of media in countering special interest group inuence. For instance, in Str o mbergs (2001, 2004a) mass media counteracts special interest group inuence for two reasons. First, the increasing returnstheory, to scalemass nature of their technology and their need for advertising For instance, in Str o mbergs (2001, 2004a) media counteracts special interest group inuence An empirical exercise inspired by these theories is of conducted by Prat et al. (2006). Using for two reasons. First, the increasing returns to scale nature their technology and their need for advertising revenue induce media outlets to provide news to large groups rather than smalland interests. Thus, mass media for two reasons. First, the increasing returns to scale nature of their technology their need for advertising revenue induce media outlets to provide tothe large groups rather than small interests. Thus, mass media a survey-based dataset about the news eectiveness of state legislators in North Carolina, they entices politicians to pay more attention to better-informed, larger audience. Second, by reducing revenue induce media outlets to provide news to large groups rather than small interests. Thus, mass media entices politicians to pay more attention to the better-informed, larger audience. Second, by reducing the share of uninformed voters, free media reduces the eectiveness oflarger advertising purchased with campaign entices politicians to pay more attention to the better-informed, audience. Second, by reducing ask whatof voters can learn about the candidate characteristics from the amount with and campaign pattern the share uninformed voters, free media reduces the e ectiveness of advertising purchased contributions. the share of uninformed voters, free media reduces the e ectiveness of advertising purchased with campaign contributions. 7 of campaign contributions. They nd that the total amount a candidate receives is a weak contributions. 7 Along these lines but in a developing country context, Ferraz and Finan (2008) nd that news about these as lines but in a by developing country context, Ferraz andon Finan (2008) nd that news about 7 Along local corruption uncovered (randomly assigned) audit reports Brazils municipal mayors hurts Along these lines but in a by developing country context, Ferraz andon Finan (2008) nd that news about predictor of that candidates e ectiveness, and that small-sized (large) contributions from local corruption as uncovered (randomly assigned) audit reports Brazils municipal mayors hurts incumbents electoral performance especially where local radioreports is present to divulge the information. In a local corruption as uncovered by (randomly assigned) audit on Brazils municipal mayors hurts incumbents electoral performance especially where local radio is present to divulge the information. In a organizations positively (negatively) signal e ectiveness. Experimental results by Houser eld experiment, Banerjee et al. (2010) provide Indian slum with containing report incumbents electoral performance especially where local radiodwellers is present to newspapers divulge the information. In a
eld Banerjee et al. and (2010) provide Indian slum dwellers is with newspapers containing cardsexperiment, on candidate qualications performance. Better information found to increase turnout, eld Banerjee et al. and (2010) provide Indian slum dwellers is with newspapers containing cardsexperiment, on candidate qualications performance. Better information found to increase turnout, cards on candidate qualications and performance. Better information is found to increase turnout, cash-based vote buying, and cause electoral gains for better performing incumbents. 8 6 A much-cited contribution is Grossman and Helpman (1996), which builds on Baron (1994). report reduce report reduce reduce

6 6 8 7

This paper follows this line of research by studying the role of free media on improving This paper follows this line of research by studying the role of free media on improving This paper follows this of research studying the role of support free media improving political accountability, andline in particular inby changing the electoral for on certain types political accountability, and in particular in changing the electoral support for certain types political accountability, and in particular in changing the electoral support for certain types This paper follows thisunlike line of previous research work, by studying the role of free media on improving of candidates. However, it emphasizes the role of mass media in of candidates. However, unlike previous work, it emphasizes the role of mass media in This paper follows this line of research by studying the role of free media on improving of candidates. However, unlike previousinwork, it of emphasizes the role offor mass media in political accountability, and in particular changing the support certain types countering special interest group inuence. In terms theelectoral empirical strategy, the di erences countering special interest group inuence. In terms ofthe theelectoral empirical strategy, the dierences political accountability, and in particular in changing support for certain types countering special interest group In terms the empirical strategy, the erences of candidates. However, unlike inuence. previous it of emphasizes role of mass media in between in-state and out-of-state counties work, were rst studied by the Ansolabehere et di al. (2006) between in-stateHowever, and out-of-state counties work, were rst studied by the Ansolabehere et al. (2006) of candidates. unlike previous it emphasizes role of mass media in between in-state and out-of-state counties were rst studied by Ansolabehere et di al. (2006) countering special interest group inuence. In terms of the empirical strategy, the erences to examine the implications of television for the incumbency advantage in the US. to examine special the implications of television for the incumbency advantage in the US. countering interest group inuence. In terms of the empirical strategy, the dierences to examine the is implications of to television the incumbency advantage in the US. between in-state and counties were rst studied by Ansolabehere et al. (2006) The paper also out-of-state related a vast for empirical and theoretical literature on campaign The paper is and also out-of-state related to a vast empirical and theoretical literature on campaign between in-state counties were rst studied by Ansolabehere et al. (2006) The paper is alsoofrelated a vast empirical and theoretical campaign to examine theMuch implications of to television for the incumbency advantage in the on US. contributions. the theoretical work assumes that there are literature informed voters who contributions. of the theoretical work assumes that there are informed voters who to examine theMuch implications of television for the incumbency advantage in the US. contributions. Much of the theoretical work assumes that there are informed voters who The is also related to a policy vast empirical and theoretical literature campaign vote for paper candidates based on their positions, and uninformed voterson who can be vote for paper candidates based on their policy positions, and uninformed voterson who can be The is also related to a vast empirical and theoretical literature campaign vote forby candidates their policy uninformed voters who depend can be contributions. Muchbased of theon theoretical work assumes and that there are informed voters who swayed campaign advertising. Funds forpositions, advertising, provided by interest groups, swayed by campaign advertising. Fundswork for advertising, provided by interest groups, depend contributions. Much of the theoretical assumes that there are informed voters who swayed by campaign advertising. Funds forpositions, advertising, provided by interest groups, vote candidates based onthe their policy and uninformed voters who depend can for be upon for the positions taken by candidates, and these positions take their implications upon for thecandidates positions taken by the candidates, and these positions take their implications for vote based on their policy positions, and uninformed voters who can be 8 upon the taken by the candidates, and these positions take their for swayed bypositions campaign advertising. Funds for advertising, provided by interest groups, depend contributions and votes into account (Morton and Cameron (1992) o er animplications early review) 8. contributions and votes into account (Morton and Cameron (1992) oer an groups, early review) swayed by campaign advertising. Funds for advertising, provided by interest depend 8. . contributions and votes into account (Morton and Cameron (1992) o er an early review) upon positions taken by the these take their implications less for Whilethe this work incorporates thecandidates, response ofand voters to positions overall campaign expenditures, Whilethe this work incorporates thecandidates, response ofand voters to positions overall campaign expenditures, less upon positions taken by the these take their implications for 8 While this work incorporates the response of voters toby overall campaign expenditures, less . contributions and votes into (Morton and Cameron (1992) oinuence er an early review) research investigates how theaccount sources of money raised politicians voters. This 8 research investigates how the sources of money raised by politicians inuence voters. This . contributions and votes into account (Morton and Cameron (1992) o er an early review) research investigates how sources of money raised by politicians inuence voters. This While work incorporates the response of voters to overall campaign expenditures, less occurs this partly because the the approach described above assumes that uninformed voters do occurs this partly because the approach described above to assumes uninformed votersless do While work incorporates the response of voters overallthat campaign expenditures, occurs partly because the the approach above assumes that uninformed voters do research investigates how sources of money raised by politicians voters. This not have rational expectations. As described emphasized by Coate (2004), ifinuence they did, they could not have investigates rational expectations. As emphasized by Coate (2004), ifinuence they did, they could research how the sources of money raised by politicians voters. This not have rational expectations. As described emphasized by Coate (2004), if they did, could occurs partly the approach above assumes that uninformed voters do realize that a because party involved in advertising must distort its policy platform to they obtain the realize partly that a because party involved in advertising must distort its policy platform to obtain the occurs the approach described above assumes that uninformed voters do realize that a party involved in advertising must distort its policy platform to obtain the not have rational expectations. As emphasized by Coate (2004), they did, they could necessary funds, and switch their votes against the advertised party. if Similarly, the empirical necessary funds, and switch their votes against the advertised party. if Similarly, the empirical not have rational expectations. As emphasized by Coate (2004), they did, they could necessary funds, and switch votes against the advertised party. Similarly, the empirical realize that a party involved in eadvertising must distort its policy platform to obtain the work has focused mostly on their the ect of gross campaign expenditures on electoral outcomes work has focused mostly on the eadvertising ect of gross campaign expenditures on electoral outcomes realize that a party involved in must distort its policy platform to obtain the work has focused mostly on their the or e ect of gross campaign expenditures on electoral outcomes necessary funds, and switch votes against the(see advertised party. Similarly, empirical and in altering policy positions securing favors Stratmann (2005) for a the review). and in altering policy positions or securing favors Stratmann (2005) for a the review). necessary funds, and switch their votes against the(see advertised party. Similarly, empirical and in altering policy positions or securing favors (see Stratmann (2005) fortheir a review). work has focused mostly on the in eect of (2006) gross campaign expenditures on electoral outcomes More recent work reviewed Prat assumes that voters update beliefs raMore recent work reviewed in Prat (2006) assumes that voters update their beliefs rawork has focused mostly on the e ect of gross campaign expenditures on electoral outcomes recent work reviewed in (2006) assumes that voters update their beliefs andMore in altering policy positions or Prat securing favors (see for a review). tionally. Private campaign nance creates a trade-o Stratmann between a (2005) policy distortion and raan tionally. Private campaign nance creates favors a trade-o Stratmann between a (2005) policy for distortion and an and in altering policy positions or securing (see a review). tionally. Private campaign nance creates a trade-o between a policy distortion and an More recent work reviewed in Prat (2006) assumes that voters update their beliefs rainformational benet. In equilibrium, qualied candidates receive more contributions than informational benet. In equilibrium, qualied candidates receive more contributions than More recent work reviewed in Prat (2006) assumes that voters update their beliefs rainformational benet. In equilibrium, qualied candidates receive more contributions than tionally. Private campaign nance creates amay trade-o to between atheir policy distortion and an unqualied candidates. However, candidates need distort policy choices (away unqualied candidates. However, candidates may need to distort their policy choices (away tionally. Private campaign nance creates a trade-o between a policy distortion and an unqualied However, may need to distort their policy (away informational benet. in In equilibrium, qualied candidates receive more contributions than from voterscandidates. interests) order to candidates attract private donations. When voters are choices aware of this, from voters interests) to attract private donations.receive When more voters are aware of than this, informational benet. in In order equilibrium, qualied candidates contributions from voterscandidates. interests) in order to candidates attract private When voters are aware of this, unqualied However, may donations. need may to distort their policy choices (away the amounts and patterns of campaign contributions inuence their voting decision. the amounts and patterns of campaign contributions may inuence their voting decision. unqualied candidates. However, candidates may need may to distort theirtheir policy choices (away the empirical amounts and patterns of campaign contributions inuence voting decision. from voters interests) in order to attract private is donations. When voters are aware of this, An exercise inspired by these theories conducted by Prat et al. (2006). Using An empirical exercise inspired by these theories is conducted by Prat et al. (2006). Using from voters interests) in order to attract private donations. When voters are aware of this, An empirical exercise inspired by these theories isstate conducted by Prat et al. (2006). Using the amounts and patterns of campaign contributions may inuence voting decision. a survey-based dataset about the eectiveness of legislators in their North Carolina, they a survey-based dataset about the eectiveness of statemay legislators in their Northvoting Carolina, they the amounts and patterns of campaign contributions inuence decision. a survey-based dataset about the ecandidate ectiveness of legislators in North they An empirical exercise inspired by these theories isstate conducted by Prat et al. Carolina, (2006). Using ask what voters can learn about the characteristics from the amount and pattern ask empirical what voters can learn about the candidate characteristics from the et amount and pattern An exercise inspired by these theories is conducted by Prat al. (2006). Using ask what voters can learn about the the amount and a survey-based dataset about the ecandidate ectiveness of state legislators in North Carolina, they of campaign contributions. They nd that the characteristics total amount afrom candidate receives is pattern a weak of campaign contributions. They nd that the total amount a candidate receives is a weak a survey-based dataset about the e ectiveness of state legislators in North Carolina, they of campaign contributions. They nd that the total afrom candidate receives is pattern a weak ask what voters can learn about the candidate the amount and predictor of that candidates eectiveness, andcharacteristics that amount small-sized (large) contributions from predictor of that candidates eectiveness, andcharacteristics that small-sized (large) contributions from ask what voters can learn about the candidate from the amount and pattern predictor of that candidates eectiveness, and that amount small-sized (large) contributions from of campaign contributions. They ndsignal that the total a candidate receivesby is Houser a weak organizations positively (negatively) eectiveness. Experimental results organizations positively (negatively) signal e ectiveness. Experimental results by Houser of campaign contributions. They ndsignal that the total amount a candidate results receivesby is Houser a weak organizations positively (negatively) eectiveness. Experimental predictor of that candidates eectiveness, and that small-sized (large) contributions from predictor of that candidates e ectiveness, and that small-sized (large) contributions from and Stratmann (2008) support the prediction that voters respond to advertising diHouser erently cash-based vote buying, and cause electoral gains for better performing incumbents. organizations positively (negatively) signal e ectiveness. Experimental results by cash-based vote buying, and cause electoral gains for better performing incumbents. 8 contribution is Grossman and Helpman (1996), which builds on Baron (1994). organizations positively (negatively) signal e ectiveness. Experimental results by Houser cash-based vote buying, and and cause electoral gains for better performing incumbents. 8 A much-cited between special interest publicly-nanced campaigns. 8 A much-cited contribution is Grossman and Helpman (1996), which builds on Baron (1994). A much-cited contribution is Grossman and Helpman (1996), which builds on Baron (1994). cash-based vote cause electoral gains for better performing incumbents. Only a fewbuying, other and papers have examined the impact of campaign nance composition on 8 cash-based vote buying, and cause electoral gains for 7 better performing incumbents. A much-cited contribution is Grossman and Helpman (1996), which builds on Baron (1994). 7 the US House of Representatives 8 behavior. Vanberg (2008) uses data on voter from 1990 7 A much-cited contribution is Grossman and Helpman (1996), which builds on Baron (1994). to 2002 and nds no evidence of a negative relation between a candidates reliance on large 7 7 and 9 Palda (2002) nd a negative relationship contributions and votes. Instead, Dharmapala between the concentration of contributions and vote shares for open-seat candidates and

and Stratmann (2008) support the prediction that voters respond to advertising dierently between special interest and publicly-nanced campaigns. Only a few other papers have examined the impact of campaign nance composition on voter behavior. Vanberg (2008) uses data on the US House of Representatives from 1990 and Stratmann (2008) supportof the prediction that voters respond to advertising di erently to 2002 and nds no evidence a negative relation between a candidates reliance on large between special interest and publicly-nanced campaigns. contributions and votes. Instead, Dharmapala and Palda (2002) nd a negative relationship Only the a few other papers of have examined the impact of campaign nance candidates composition on between concentration contributions and vote shares for open-seat and voter behavior. (2008) uses to data on with the US House relation of Representatives from 1990 challengers in theVanberg US House from 1980 1992, no robust for the incumbents.As topotential 2002 andexplanation, nds no evidence of a negative relation between a candidates reliance on large a they suggest, in line with the argument put forward in this paper, contributions and votes. Dharmapala (2002) nd a negative relationship that interest groups seek Instead, favors for themselves,and andPalda thus are in conict with each other and between concentration ofifcontributions and vote shares for open-seat and with the the voters. Therefore, campaign contributions are instrumental incandidates securing these challengers in the US House from 1980 to 1992, with no sources robust relation the incumbents.As favors, candidates with more dispersed contribution are lessfor susceptible to being a potential explanation, theyand suggest, in line with the argument put they forward in this paper, captured by any one group are preferred by voters. However, admit that their that interest groups seek favors for that themselves, and thus are in conict with each other and empirical strategy cannot rule out causality runs other way: candidates with a higher with the voters. Therefore, if campaign contributions are instrumental in securing these 9 . By exploiting the likelihood of winning may attract a wider pattern of contributions favors, candidates with more dispersed contribution sources are less susceptible to being dierential impact of concentrated patterns of contributions across dierent media markets, captured any one group and causality are preferred by voters. However, they admit that their I attempt by to rule out the reverse story. empirical strategy cannot rule out that causality runs other way: candidates with a higher 9 The rest the paper proceeds follows. In Section 2, I spell out basic research . the By exploiting the likelihood of of winning may attract as a wider pattern of contributions design and impact describe main datapatterns used in of the paper (a Data Appendix a more di erential of the concentrated contributions across dierentpresents media markets, complete the variables and sources I attemptdescription to rule outof the reverse causality story.of the analysis). Section 3 presents the main results for the impact of concentration of campaign contribution on incumbent vote shares The of media the paper proceeds as section, follows. IIn Section 2,aInumber spell out basic research across direst erent markets. In this also present of the robustness checks, design and describe the main data used in the paper (a Data Appendix presents a more including the placebo test designed to verify the identication hypothesis that the results are complete description of the variables and or sources of the analysis). Section 3 presents the main not driven by geographic characteristics distance from the media center of the state. The results for the impact of concentration of campaign contribution on incumbent vote shares section also shows that, unlike concentration, total share of money from interest groups does across erent media senators markets. di In section, I also present a number of robustness checks, not adi ect incumbent this erentially across media markets. The section concludes including the placebo test designed to verify the identication hypothesis the results are by showing that concentration indices are positively correlated with thethat relative frequency not driven by geographic characteristics or distance theSection media center of the The of news stories about candidates campaign money.from Next, 4 shows thatstate. the main section alsonot shows that, unlike concentration, total share money from interest groups does results are driven by partisan trends, and nds littleof evidence of information spillovers not aect incumbent senators dierentially across media The section across members of the same party, whether in the same markets. or neighboring states.concludes Section by showing that concentration indices are positively correlated with the relative frequency 5 demonstrates that the results of the paper are not sensitive to the exact classication of news stories campaign Next, Section 4 shows that the main of counties into about those candidates dominated by in-state money. or out-of-state media markets. I conclude in results are not driven by partisan trends, and nds little evidence of information spillovers 9 Section 6. Palda and Palda (1998) who suggest French voters punish candidates who raise money from see also across members of the same party, whether in the same or neighboring states. Section narrow sources. 5 demonstrates that the results of the paper are not sensitive to the exact classication 2 counties Empirical Strategy of into those dominated by in-state 8or out-of-state media markets. I conclude in
9

2.1 sources. Design and specication narrow

see also Palda and Palda (1998) who suggest French voters punish candidates who raise money from

Like Ansolabehere et al. (2006), I examine the impact of mass media on voters response 8 to politicians campaign-nance proles by exploiting the structure of media markets. Since 10 television is the primary source of political news for voters Ansolabehere et al. (1993), I focus

Section 6. Section 6. Section 6.

2 2 2
2.1 2.1 2.1

Empirical Strategy Empirical Empirical Strategy Strategy

Design and specication Design Design and and specication specication

Like Ansolabehere et al. (2006), I examine the impact of mass media on voters response Like Ansolabehere et al. (2006), I examine the impact of mass media on voters response Like Ansolabehere et al. (2006),proles I examine the impact of mass media on voters response to politicians campaign-nance by exploiting the structure of media markets. Since to politicians campaign-nance proles by exploiting the structure of media markets. Since to politicians campaign-nance proles bynews exploiting the structure of media television is the primary source of political for voters Ansolabehere et al.markets. (1993), I Since focus television is the primary source of political news for voters Ansolabehere et al. (1993), I focus television is the primary source of political newsresults for voters Ansolabehere et al. I focus on TV markets. More concretely, I compare in Senate elections for (1993), counties with on TV markets. More concretely, I compare results in Senate elections for counties with on TVmarkets markets. More concretely, I compare results inI Senate elections for counties with media centered in their own state (henceforth, will often refer to these simply as media markets centered in their own state (henceforth, I will often refer to these simply as media markets centered in their own state (henceforth, I will often refer to these simply as in-state counties), with those whose media markets are centered in a city of a neighboring in-state counties), with those whose media markets are centered in a city of a neighboring in-state counties), with those whose media markets are centered in a city of a neighboring state (out-of-state counties). state (out-of-state counties). state counties). In (out-of-state out-of-state counties, news focuses on the neighboring state politics and elections. In out-of-state counties, news focuses on the neighboring state politics and elections. out-of-state news focuses on the neighboring state politics and elections. As In a result, voters counties, receive much less television coverage of their states politics than voters As a result, voters receive much less television coverage of their states politics than voters As a result, voters receive much less television coverage of their states politics than voters covered by in-state media markets. Ansolabehere et al. (2006) present evidence indicating covered by in-state media markets. Ansolabehere et al. (2006) present evidence indicating covered by in-state in-stateversus mediaout-of-state markets. Ansolabehere et a al. (2006) of present indicating that being is more clearly function actualevidence television coverage that being in-state versus out-of-state is more clearly a function of actual television coverage that being in-state versus out-of-state is more clearly a function of actual television coverage than other measures of media market structure that have been used in the literature, such than other measures of media market structure that have been used in the literature, such than other measures of media structure that have been used the literature, such as fragmentation or number of market television stations. They searched the in on-line archives of 90 as fragmentation or number of television stations. They searched the on-line archives of 90 as fragmentation or number of television stations. They searched the on-line archives of 90 stations aliated with 51 media markets for stories that mention the governors of states. stations aliated with 51 media markets for stories that mention the governors of states. stations aliated with10 51times media for stories that governors states. News programs aired asmarkets many stories about the mention in-state the governor than of they did News programs aired 10 times as many stories about the in-state governor than they did News programs aired 10 times as many stories about the in-state governor than they did of governors from neighboring states covered by the media market. Moreover, the number of governors from neighboring states covered by the media market. Moreover, the number of governors neighboring governors states covered by the media market. Moreover, the number of stories of from the out-of-state was typically extremely small, and on the order of stories of the out-of-state governors was typically extremely small, and on the order of stories of the out-of-state governors was typically extremely small, and on the order of noise. They further report data from the National Election Studies of 1974 and 1978, of noise. They further report data from the National Election Studies of 1974 and 1978, of noise. They further report data from the National Election Studies of 1974 and 1978, which contain information to determine the type of market respondents reside in, conrming which contain information to determine the type of market respondents reside in, conrming which contain to determine type of market respondents resideAbout in, conrming that these diinformation erences on coverage havethe consequences on voter knowledge. 70% of that these dierences on coverage have consequences on voter knowledge. About 70% of that these di erences coverage have consequences on voter knowledge. About 70% of respondents of in-state on counties report that they saw a Senate candidate on television during respondents of in-state counties report that they saw a Senate candidate on television during respondents of in-state counties report they saw a Senate candidate on television the campaign, compared to only 50%that of respondents in out-of-state counties, and during this is the campaign, compared to only 50% of respondents in out-of-state counties, and this is the campaign, compared to only statistically signicant at the 0.01 50% level.of respondents in out-of-state counties, and this is statistically signicant at the 0.01 level. statistically signicant at the 0.01 level. the response of voters to the campaign proles These observations suggest comparing These observations suggest comparing the response of voters to the campaign proles These observations suggest comparing ofin-state voters to campaign proles of candidates depending on whether they the are response located in or the in out-of-state media of candidates depending on whether they are located in in-state or in out-of-state media of candidates depending on whether they are located in in-state or in out-of-state media markets. The following regression model provides the simplest comparison of this sort: (1)
I O = sP I + tP I + in inc + hI hI Vct st + hO hst

O 0 +inhI inc hI st + inhO inc hst + X Xcst + sct . 9 9 9 In (1), c indexes counties, s states, and t time. The superscript I is for incumbent and O I for his opponent. The dependent variable, Vct , is the share of the two-party vote received by the incumbent candidate running for the Senate in county c at time t; inc is a dummy variable equal to one if the county is in-state and zero if it is out-of-state; hI st is the Herndahl 11 concentration index of special-interest campaign contributions to the incumbent (hO st is the

(1)

In (1), c indexes counties, s states, and t time. The superscript I is for incumbent and O In (1), c indexes counties, s states, and t time. The superscript I is for incumbent and O In c indexes counties, s states, and t time. The superscript I is for incumbent and O for (1), his opponent. The dependent variable, VI I , is the share of the two-party vote received by for his opponent. The dependent variable, Vct I , is the share of the two-party vote received by ct for opponent. The dependent Vct , is in the share of the two-party vote received by the his incumbent candidate running variable, for the Senate county c at time t; in is a dummy varithe incumbent candidate running for the Senate in county c at time t; inc c is a dummy varithe candidate running for the Senate in if county c at time t; in c is a dummy variableincumbent equal to one if the county is in-state and zero it is out-of-state; hI st is the Herndahl able equal to one if the county is in-state and zero if it is out-of-state; hI I st is the Herndahl able equal to one if the county is in-state and zero if it is out-of-state; hst is the Herndahl concentration index of special-interest campaign contributions to the incumbent ( hO st is the concentration index of special-interest campaign contributions to the incumbent (hO st is the the concentration of special-interest campaign contributions to the incumbent (hO st is Decorresponding index measure for the opponent); and X cst is a vector of additional controls. corresponding measure for the opponent); and Xcst is a vector of additional controls. Deof additional controls. Decorresponding measure for the and X cst is a vector meaned variables are shown in opponent); bold. I demean Herndahl indices before interacting them, meaned variables are shown in bold. I demean Herndahl indices before interacting them, meaned variables are shown inmain bold.e Iect demean before interacting shows theindices impact measured at the them, mean so that the coecient on the for inHerndahl so that the coecient on the main eect for inc c shows the impact measured at the mean I coecient O I shows the impact measured at the mean so that the on the main e ect for in c and hO values of hI st . The impact of hI st on the incumbent vote share is hI for out-of-state I and h . The impact of h on the incumbent vote share is for out-of-state values of hst h O st st st O h . The impact of hI incumbent share is hI for out-of-state values ofand hI st and stin st on the (and similarlyvote for h counties I+ hI for in-state counties O st ). Throughout the paper, I I + for in-state counties (and similarly for h counties and h h inh O st ). Throughout the paper, inhI at forthe in-state counties counties hI + errors I cluster and standard county level. (and similarly for hst ). Throughout the paper, I cluster standard errors at the county level. I cluster standard errors at the a county level. This specication includes full set of year and state times incumbent party xed This specication includes a full set of year and state times incumbent party xed This specication includes a full set of year and stateinteracted times incumbent xed xed eects with the party incumbents eects (tP I and sP I respectively). Year eects (tP I and sP I respectively). Year xed eects interacted with the incumbents respectively). Year xed or eects interacted the eects are ( t P I and s P I capture party important to national political economic tides,with such as incumbents presidential party are important to capture national political or economic tides, such as presidential party are important to capture national political or economic tides, such as presidential coattails, systematic presidential punishment at the midterm, or any other general trends in coattails, systematic presidential punishment at the midterm, or any other general trends in coattails,tides systematic presidential punishment at the midterm, or any other general trends in political or economic variables (e.g. unemployment, ination) that could aect election political tides or economic variables (e.g. unemployment, ination) that could aect election political tides or economic variables unemployment, ination) that could aects ect election results in all counties. Including a (e.g. full set of state-incumbent party xed e P I is results in all counties. Including a full set of state-incumbent party xed eects s sP I is results in all counties. Including a full set of state-incumbent party xed e ects sP I is important for several reasons. First, the state xed eects focus the comparison between important for several reasons. First, the state xed eects focus the comparison between important for several reasons. First, thethe state xed eects focus thein comparison between in-state and out-of-state counties within same state. Thus, since all counties within in-state and out-of-state counties within the same state. Thus, since in all counties within in-state and out-of-state counties within the same state. Thus, since in all counties within a state the same two candidates are running for Senate at each time period, I can hold the a state the same two candidates are running for Senate at each time period, I can hold the a state the running, same twothe candidates are for Senate at each time period, I canconstant. hold the candidates closeness of running the election, and other features of the race candidates running, the closeness of the election, and other features of the race constant. candidates running, the closeness election, hard and other features the race constant. This comparison is very useful, asof itthe is typically to control for of issues like candidate This comparison is very useful, as it is typically hard to control for issues like candidate This comparison is capture very useful, as it is partisanship typically hard control issues like candidate quality. Finally, to the average of to each state for s, P I also includes a quality. Finally, to capture the average partisanship of each state s, s sP I also includes a includes quality. variable Finally, for to capture the partisanship ofinteraction each state with s, s P I also dummy the party of average the incumbent and its state xed eects.a dummy variable for the party of the incumbent and its interaction with state xed eects. dummy variable the party of the are incumbent and its interaction with state xed eects. The main coefor cients of interest hI and inhO . I expect inhI to be negative, The main coecients of interest are in inhI and inhO . I expect inhI to be negative, The main cients of interest are inhI and in inhI tofew be economic negative, hO . I expect implying that coe when the incumbents campaign money comes from relatively implying that when the incumbents campaign money comes from relatively few economic implying the incumbents campaign money comes from relatively few economic sectors (a that high when hI st ), voters residing in in-state counties are more likely to nd out and punish sectors (a high hI I st ), voters residing in in-state counties are more likely to nd out and punish residing in in-state counties are more likely to nd out and punish sectors (a high relative hst ), voters the incumbent to voters in out-of-state counties. With a similar logic, I expect hO the incumbent relative to voters in out-of-state counties. With a similar logic, I expect in inhO the incumbent to be positive. relative to voters in out-of-state counties. With a similar logic, I expect inhO to be positive. to be positive. However, there are some threats to the 10 correct identication of inhI and inhO in 10 10 (1). One potential issue is that the counties in out-of-state dominated media markets are not an adequate control group for in-state counties. As will be shown below, out-ofstate counties are indeed dierent along a number of dimensions (e.g., they are smaller, less urban, poorer, and exhibit dierent age and racial composition) compared to in-state counties. These dierences are concerning if these characteristics inuence voters preferences in ways that lead them to vote more for or against candidates with concentrated campaign contributions. If this is the case, the eect attributed to media exposure may actually be driven by these other dierences between counties. I implement the simplest solution to this problem by controlling for these county characteristics in the regression. However, there may 12 persist additional unobserved factors that are not controlled for and may be generating the
10

hst Vct = sP I + tP I + in inc + hI hst + hOO 0 +inhI inc hI I 0 Xcst + sct . st + inhO inc hO st + X O + in + h +inhI inc hI c O 0 Xcst + sct . in h st st X Xcst + sct . +inhI inc hst + inhO inc hst + X

state counties are indeed dierent along age a number of dimensions (e.g., they are less urban, poorer, and exhibit dierent and racial composition) compared to smaller, in-state less urban, poorer, and exhibit dierent age and racial composition) compared to in-state less urban, poorer, and exhibit dierent age andcharacteristics racial composition) compared to in-state counties. These dierences are concerning if these inuence voters preferences counties. These dierences are concerning if these characteristics inuence voters preferences counties. These di erences are concerning if these characteristics inuence voters preferences in ways that lead them to vote more for or against candidates with concentrated campaign in ways that lead them to vote more for or against candidates with concentrated campaign in ways that lead them vote more or against candidates with concentrated campaign contributions. If this is to the case, thefor eect attributed to media exposure may actually be contributions. If this is the case, the eect attributed to media exposure may actually be contributions. If this is the case, the e ect attributed to media exposure may actually be driven by these other dierences between counties. I implement the simplest solution to this driven by these other dierences between counties. I implement the simplest solution to this driven by these other di erences between counties. I implement the simplest solution to may this problem by controlling for these county characteristics in the regression. However, there problem by controlling for these county characteristics in the regression. However, there may However, there are for some threats to the identication ofmay in and there inhO in problem by controlling these county characteristics in the regression. However, may hI generating persist additional unobserved factors that arecorrect not controlled for and be the persist additional unobserved factors that are not controlled for and may be generating the 10 (1). One issue is that the counties out-of-state dominated media markets the are persist additional unobserved factors that are in not controlled for and may be generating . potential results . results10 not an10 adequate control group for in-state counties. As will First, be shown below,(1) out-of. results To address this problem I adopt three additional approaches. I estimate with To address this problem I adopt three additional approaches. First, I estimate (1) with state counties are indeed di erent along a number of dimensions (e.g., they are smaller, To address this problem I adopt three additional approaches. First, I estimate (1) with the inclusion of county (times party of the incumbent) xed eects. In particular, I run the the inclusion of county (times party of the incumbent) xed eects. In particular, I run the less urban, poorer, and(times exhibit dierent age and racial composition) to the inclusion of county party of the incumbent) xed eects. Incompared particular, I in-state run the following regression: following regression: counties. These dierences are concerning if these characteristics inuence voters preferences following regression:
I them to vote more for or I against candidates O in ways that lead with concentrated campaign (2) Vct I = cP I + tP I + hI hI st + hO hO st O (2) + h Vct I = cP I + tP I + hI hI O h st st I + (2) = tP the hst attributed + hO hst to media I O 0 c PI + hI contributions.Vct If this is the case, e ect exposure may actually be + 0 Xcst + sct . hI inc hI st + inhO inc hO st + X I O in + in + h h X + +in c c cst sct . I O 0 inh h st st driven by these other di between implement simplest solution to this + in X . + erences c hst counties. c hst + the cst + sct inhI in inhO Iin XX problem controlling county characteristicsxed in the regression. However, there may denotes a for fullthese set of county-times-party e ects. Again, the motivation to In (2), by P I In (2), c cP I denotes a full set of county-times-party xed eects. Again, the motivation to I denotes a full set of county-times-party xed e ects. Again, the motivation to In (2), c P partisanship persist additional unobserved factors that are not controlled for and may be generating the interact the of the incumbent with the county xed eects comes from having, interact the partisanship of the incumbent with the county xed e ects comes from having, 10 the partisanship of the incumbent with the county xed eects comes from having, interact . pro-incumbent counties, heavily democratic counties and heavily republiresults more than more than pro-incumbent counties, heavily democratic counties and heavily republimore than pro-incumbent heavily democratic counties and heavily republiTo address this problem Icounties, adopt three additional approaches. First, Ie estimate (1) with can counties. In other words, the incumbent party times county xed ects capture the can counties. In other words, the incumbent party times county xed eects capture the can counties. In other words, the incumbent party times county xed e ects capture the the inclusion of county (times party of in the incumbent) underlying partisanship (normal vote) each county. xed eects. In particular, I run the underlying partisanship (normal vote) in each county. underlying partisanship (normal vote) in each county. following While regression: estimation of (2) comes at the cost of not estimating the direct eect of in-state While estimation of (2) comes at the cost of not estimating the direct eect of in-state While estimation of (2) comes at the cost of not estimating the direct eect of in-state status on the incumbent vote share, any time-invariant characteristic of the counties that status on the incumbent vote share, any time-invariant characteristic of the counties that I I O I + (2) = generating cP hI h status on the V incumbent vote share, time-invariant characteristic the counties that t spurious P I +any ct be st + hO hst is controlled could potentially a relationship for. of Indeed, such specicould potentially be generating a spurious relationship is controlled for. Indeed, such speciI 0 for. Indeed, such specicould be generating a spurious relationship controlled incis X h hO X . same county +inhI in cationpotentially relies solely on variation Herndahl indices across time within the cthe cst + sct st + inhO st + cation relies solely on variation in the Herndahl indices across time within the same county cation relies the solely on variation in the Herndahl indices across time within counties the sametend county to estimate coe cient of interest. This estimate asks whether in-state to to estimate the coecient of interest. This estimate asks whether in-state counties tend to I denotes a full set of county-times-party xed e ects. Again, the motivation In (2), to estimate the coe cient of interest. This estimate asks whether in-state counties tend to cP punish increases in the Herndahl index of the candidate more than out-of-state counties. punish increases in the Herndahl index of the candidate more than out-of-state counties. interactincreases the partisanship of the incumbent the county xedthan eects comes from having, punish in the Herndahl index ofwith the candidate more out-of-state counties. 10 10 In the context of their examination of the incumbency advantage, in order to assuage these concerns the context theirmatch examination of the incumbency advantage, in orderwith to assuage these concerns more pro-incumbent counties, heavily democratic and heavily republi10 Inthan Ansolabehere et al. of (2006) the counties out-of-state media counties markets counties with in-state In the context of theirmatch examination of thewith incumbency advantage, in orderwith to assuage these concerns Ansolabehere et al. (2006) the counties with out-of-state media markets counties with in-state Equation (2) is therefore a much more demanding specication. However, one may media on several dimensions and estimate the size of the incumbency advantage using only the matched can counties. In(2006) othermatch words, the incumbent party times county xed e ects capture the Ansolabehere et al. the counties with out-of-state media markets with counties with in-state media on several dimensions and estimate the size of the incumbency advantage using only the matched counties. This solution may the be preferable inthe that it relaxes somedistinct of the assumptions of the basic regression still be concerned that two sets of counties have trends. If so, the response media on several dimensions and estimate size of the incumbency advantage using only the matched counties. This solution may (normal be preferable in that it relaxes some of the assumptions of the basic regression underlying partisanship vote) in that each county. alternative with added controls, yet the concern there are unaccounted factors which are not controlled counties. This solution may be preferable in that it relaxes some of the assumptions of the basic regression alternative with added controls, yet the concern that there are unaccounted factors which are not controlled of an increase in the Herndahl index in-state relative to out-of-state may reect such a for remains. While estimation of (2) comes at the cost of not estimating the direct e ect of in-state alternative with added controls, yet the concern that there are unaccounted factors which are not controlled for remains. diremains. erential trend rather than the eect of media. Thus, as a second potential solution I for status on the incumbent vote share, any time-invariant characteristic of the counties that 11 comparable to out-of-state counties. To do look for a set of in-state counties that are more 11 could potentially be generating a spurious relationship is controlled for. Indeed, such speci11 this, I focus on in-state counties that are geographically contiguous to out-of-state counties. cation relies solely on variation in the Herndahl indices across time within the same county After demonstrating that these counties are indeed more alike out-of-state counties along a to estimate the coecient of interest. This estimate asks whether in-state counties tend to number of dimensions, I estimate (2) using solely these observations. punish increases in the Herndahl index of the candidate more than out-of-state counties. A third relevant robustness check comes from a placebo test designed to verify whether the 10 In the context of their examination of the incumbency advantage, in order to assuage these concerns movement from in-state toward out-of-state captures a general geographic trend in response Ansolabehere et al. (2006) match the counties with out-of-state media markets with counties with in-state to concentrated contributions that is the unrelated with media exposure. particular, dene media on several dimensions and estimate size of the incumbency advantageIn using only the I matched counties. This solution may be preferable in that it relaxes some of the assumptions of the basic regression the contiguous, in-state counties to be the (placebo) out-of-state counties, and compare them alternative with added controls, yet the concern that there are unaccounted factors which are not controlled to adjacent for remains. in-state counties. Since these sets of counties are both in-state, there should be no correlation between increases in concentration and moving across these categories of counties be close to zero). (i.e., the estimated inhI and inhO should11 13 These robustness checks serve to rule out several potential sources of bias. Note that

di erential trend rather counties than thethat eect ofmore media. Thus, asto a out-of-state second potential solution I look for a set of in-state are comparable counties. To do look for a set of in-state counties that are more comparable to out-of-state counties. To do look a set ofin-state in-statecounties countiesthat that are more comparable to out-of-state counties. To do this, for I focus on are geographically contiguous to out-of-state counties. this, I focus on in-state counties that are geographically contiguous to out-of-state counties. this, focus on in-state counties that are are geographically contiguous to out-of-state counties. AfterI demonstrating that these counties indeed more alike out-of-state counties along a After demonstrating that these counties are indeed more alike out-of-state counties along a After demonstrating that these counties aresolely indeed more alike out-of-state counties along a number of dimensions, I estimate (2) using these observations. number of dimensions, I estimate (2) using solely these observations. number of dimensions, I estimate (2) usingfrom solely these observations. A third relevant robustness check comes a placebo test designed to verify whether the A third relevant robustness check comes from a placebo test designed to verify whether the A third from relevant robustness check comes from a placebo test designed to verify the movement in-state toward out-of-state captures a general geographic trend whether in response movement from in-state toward out-of-state captures a general geographic trend in response movement from in-state toward out-of-state captures a general geographic trend in response to concentrated contributions that is unrelated with media exposure. In particular, I dene to concentrated contributions that is unrelated with media exposure. In particular, I dene to concentrated contributions that is unrelated without-of-state media exposure. In particular, I dene the contiguous, in-state counties to be the (placebo) counties, and compare them the contiguous, in-state counties to be the (placebo) out-of-state counties, and compare them the contiguous, in-state counties to be the (placebo) out-of-state counties, and compare them to adjacent in-state counties. Since these sets of counties are both in-state, there should be no to adjacent in-state counties. Since these sets of counties are both in-state, there should be no to adjacent between in-state counties. these sets of counties are both in-state, there should be no correlation increases Since in concentration and moving across these categories of counties correlation between increases in concentration and moving across these categories of counties correlation between increases in concentration and moving across these categories of counties (i.e., the estimated hI and inhO should be close to zero). (i.e., the estimated in inhI and inhO should be close to zero). into should be close to zero). sources of bias. Note that (i.e., the estimated in hI and hO rule These robustness checks serve out several potential These robustness checks serve to rule out several potential sources of bias. Note that robustness checks out of several potential sources of bias. Note that the These variation to estimate the serve main to coerule cients interest comes from changes across time in the variation to estimate the main coecients of interest comes from changes across time in the variation toindex estimate the main coe cients interest comes time in the Herndahl of the incumbent and his of opponent. Thus, from whilechanges it is stillacross possible that the Herndahl index of the incumbent and his opponent. Thus, while it is still possible that the Herndahl index of the incumbent his have opponent. Thus, it is still changes possible(e.g., that omitted variables bias the results, theseand would to come fromwhile time-varying omitted variables bias the results, these would have to come from time-varying changes (e.g., omitted bias the results, these time would to come from time-varying changes (e.g., a changevariables in candidate quality through ashave opposed to a relatively xed quality trait of a change in candidate quality through time as opposed to a relatively xed quality trait of a change in candidate through time as opposed to a relatively xed quality trait of the candidate) that arequality correlated with changes in the campaign concentration indices. Also, the candidate) that are correlated with changes in the campaign concentration indices. Also, the candidate) that are correlated changes in main the campaign concentration indices. some of the additional robustnesswith checks on the results to be presented below, Also, such some of the additional robustness checks on the main results to be presented below, such some of the additional robustness checks on the main results to be presented such as controlling for a dierential impact of total campaign funds across categories below, of counties, as controlling for a dierential impact of total campaign funds across categories of counties, as controlling for a this dierential of total campaign funds across categories of counties, may help alleviate kind of impact concern. may help alleviate this kind of concern. may help alleviate this kind of concern.

The sources and denitions of all variables are presented in the Data Appendix. Here, I The sources and denitions of all variables are presented in the Data Appendix. Here, I The sources and denitions all variables are presented in the Data Appendix. Here, I discuss details concerning the of information on media exposure and on campaign contributions, discuss details concerning the information on media exposure and on campaign contributions, discuss concerning the information on media exposure and on campaign contributions, the twodetails most important aspects of the analysis. the two most important aspects of the analysis. the Itwo most aspects of the1980-2002. analysis. The choice of this period is based partly focus myimportant analysis in the period I focus my analysis in the period 1980-2002. The choice of this period is based partly I focus my analysis inpartly the period 1980-2002. Campaign The choice nance of this data period is based partly on data availability and on convenience. from the Federal on data availability and partly on convenience. Campaign nance data from the Federal on data availability and partly while on convenience. Campaign nance the Federal Elections Commission (FEC), available somewhat earlier, is data most from comparable and Elections Commission (FEC), while available somewhat earlier, is most comparable and Elections Commission (FEC), while available earlier, is most comparable and believed to be of best quality since around 1980. somewhat The year 2002 is a natural time to stop, since believed to be of best quality since around 1980. The year 2002 is a natural time to stop, since believed to be of best quality since around 1980. The year 2002 is a natural time to stop, since after 2002 changes in Federal regulation are believed to have impacted the way campaign 12 funds are raised. 12 12 Media exposure Following Ansolabehere et al. (2006), I use the Areas of Dominance Inuence (ADIs), constructed by Arbitron, to dene media markets. ADIs dene each television market exclusive of the others, based on the geographic distribution of television viewers. Importantly, every county is allocated exclusively to one ADI, and each markets ADI is the set of counties in which the home market stations receive a preponderance of viewing11 . A simple example with the Denver-CO and Lincoln-NE ADIs is depicted in Figure 1. As seen in the gure, a few counties in Wyoming and Nebraska are in the Denver-CO ADI, but since most counties of the Denver-CO ADI lie in Colorado, this14 is a Colorado-centered media market. Hence, such counties in Wyoming and Nebraska likely receive information about Colorado politics,

2.2 2.2 2.2

Data Data Data

Following Ansolabehere et al. (2006), I use the Areas of Dominance Inuence (ADIs), conFollowing Ansolabehere et al. (2006), I use the Areas of Dominance Inuence (ADIs), conFollowing Ansolabehere al. (2006), I use the ADIs Areas dene of Dominance Inuence (ADIs), constructed by Arbitron, to et dene media markets. each television market exclusive structed by Arbitron, to dene media markets. ADIs dene each television market exclusive structed by Arbitron, tothe dene media markets. ADIs eachviewers. television market exclusive of the others, based on geographic distribution of dene television Importantly, every of the others, based on the geographic distribution of television viewers. Importantly, every of the others, based on the geographic distribution of television viewers. Importantly, every county is allocated exclusively to one ADI, and each markets ADI is the set of counties in county is allocated exclusively to one ADI, and each markets ADI is the set of counties in county is allocated exclusively to receive one ADI, and each markets ADI is11 the set of counties in which the home market stations a preponderance of viewing . A simple example which the home market stations receive a preponderance of viewing11 11 . A simple example A simple which the home market receive a preponderance of viewing with the Denver-CO and stations Lincoln-NE ADIs is depicted in Figure 1. As .seen in the example gure, a with the Denver-CO and Lincoln-NE ADIs is depicted in Figure 1. As seen in the gure, a after 2002 changes in Federal regulation are believed to have impacted the way campaign with the Denver-CO and Lincoln-NE ADIs is the depicted in Figure 1. but As seen the gure, a few counties in Wyoming and Nebraska are in Denver-CO ADI, sincein most counties few counties in Wyoming and Nebraska are in the Denver-CO ADI, but since most counties funds are raised. few counties in Wyoming and Nebraska are in Denver-CO ADI, but since most counties of the Denver-CO ADI lie in Colorado, this isthe a Colorado-centered media market. Hence, of the Denver-CO ADI lie in Colorado, this is a Colorado-centered media market. Hence, of the Denver-CO ADI lie and in Colorado, is receive a Colorado-centered media market. politics, Hence, such counties in Wyoming Nebraska this likely information about Colorado such counties in Wyoming and Nebraska likely receive information about Colorado politics, Media exposure such counties inown Wyoming receive information Colorado politics, instead of their states. and For Nebraska example, likely a Nebraska county of this about type will have less politinstead of their own states. For example, a Nebraska county of this type will have less politinstead of Ansolabehere their own states. For example, a Nebraska county of this will have less politFollowing et al. (2006), use the than Areas of counties Dominance Inuence (ADIs), conical information about Nebraska stateI politics the in type the Area of Dominance ical information about Nebraska state politics than the counties in the Area of Dominance ical information about Nebraska state politics than the counties in the Area of Dominance structed by to dene media markets. ADIs dene each television market exclusive Inuence of Arbitron, the Lincoln-NE market shown in the gure, a media market largely comprised Inuence of the Lincoln-NE market shown in the gure, a media market largely comprised Inuence of the Lincoln-NE market shown in the gure, a media marketImportantly, largely comprised of the others, based on the geographic of television viewers. every Nebraska counties and thus centered in Nebraska. 12 same state throughout the period . distribution of Nebraska counties and thus centered in Nebraska. of Nebraska counties and thus centered in Nebraska. county is follow allocated exclusively to one and each state markets ADIconcerns ismarket the setto ofbe counties in I also thelter authors in dening the dominant a media the state A nal sample considered byADI, Ansolabehere et al.of (2006) I also follow the authors in dening the dominant state of a media marketoverwhelmed to be the state 11 I has also follow authors in of dening the state of media market to be the state . in Aa simple example which the home market stations receive a dominant preponderance of aviewing that at leastthe x % (x 50) the population of market. Likewise, a county is states. only small percentage ofthat the media population lives media market that hasIn atthese least states, x% (x 50)aof the population of that media market. Likewise, a county is that has at least x% ( x 50) of the population of if that media market. Likewise, a county is with the Denver-CO and Lincoln-NE ADIs is depicted in Figure 1. As seen in the gure, a in a media market that is out-of-state dominated the state the county is in has less than that is dominated state. Politicians and news broadcasts probably do not neglect a large in a media market in that is out-of-state dominated if the state the county is in has less than in a media market that is out-of-state dominated if the state the county is in has less than few in the Wyoming and Nebraska are in the Denver-CO ADI, but in since most counties (100counties x) % of population of the media market. population into account in this percentage in this case. For this reason, ITaking drop all counties states where less (100 x) %of of voters the population of the media market. Taking population into account in this (100 of x)course % of the population of the media market. Takingthat population intomarket. account Hence, in and this of the Denver-CO ADI lie in Colorado, this is a Colorado-centered media way is crucial as these are presumably the viewers advertisers, politicians, than two-thirds of the population live in in-state dominated media markets. politicians, and way is of course crucial as these are presumably the viewers that advertisers, way is of course as these are presumably the that advertisers, politicians, and such counties in crucial Wyoming and likely receive information about Colorado politics, broadcasters are attempting to Nebraska reach. When x= 50viewers , every county can be classied as being broadcasters are attempting to reach. When x = 50, every county can be classied as being broadcasters are attempting to reach. When x= 50,inclusion every county can be classied as being instead of their own states. For example, a Nebraska county ofof this type will have less politinor out-of-state, but such choice may imply the counties in media markets Campaign contributions in- or out-of-state, but such choice may imply the inclusion of counties in media markets inor out-of-state, but such choice may imply the inclusion of counties in media markets ical information about by Nebraska state than the other counties in the Area of Dominance not clearly dominated any state. Apolitics large x, on the hand, may exclude too many not clearly contributions dominated byby any state.interest A largegroups x, on the other hand, may at exclude too many Campaign special have been regulated least since cornot clearly dominated by any state. A large on the other hand, may exclude too many Inuence of the market in the gure, a media largely comprised counties. As in Lincoln-NE Ansolabehere et al. shown (2006) I x, use two-thirds (x = market 66) as as the benchmark counties. As in Ansolabehere et and al. (2006) I use were two-thirds (xby = 66) as as the benchmark porate contributions to parties candidates banned Congress in 1911. This counties. As in Ansolabehere al. of (2006) I use two-thirds (x = of 66) of Nebraska counties and thus et centered in Nebraska. threshold, and verify the robustness my results to other values x.as as the benchmark threshold, and verify robustness of my resultsand to other of x. prohibition was later the extended to labor unions tradevalues associations. These rules were threshold, and verify the robustness of the my results to other values of x. I also follow the authors in dominant of a media market toover be the state A second important issue is dening that the boundaries of state television markets change time, as A second important is that the boundaries of and television markets had change overPolitical time, as weakly enforced and by issue the early 1970s many unions organizations formed A second important issue is that the boundaries of television markets change over time, that media has at markets least x%are (x created, 50) of and the old population of that media market. Likewise, county as is new ones cease to exist by splintering or slow a absorption new media markets are created, and old ones cease to exist by splintering or slow absorption Action Committees (PACs) to collect contributions for candidates without violating the law. new media markets are created, and old ones cease to by the splintering slow absorption in a other mediamedia market that is Moreover, out-of-state if the state county or isof in has less than into markets. as dominated noted above, I exist only have denitions media markets into other media markets. Moreover, as noted above, I only have denitions of media markets In 1974, the Federal Election Campaign Act (FECA) sanctioned this practice, mandating into other media markets. Moreover, as noted above, I only have denitions of media markets (100 x)decade. % of the population the analysis, media market. Taking population into account in once per Hence, in the of basic I include counties that were dominated by this the once per decade. Hence, in the to basic analysis, I include counties that were dominated by the the disclosure of contributions the FEC. once per Hence, in the basic analysis, I include counties were dominated by and the way is of decade. course crucial as these are presumably the viewers that that advertisers, politicians, 12 Ansolabehere et al. (2003) summarize the main features of this campaign nance regulasame state throughout the period 12 . broadcasters are attempting to reach. When x = 50, every county can be classied as being same 11 state throughout the period . These market denitions are from Broadcast and Cable (1980, 1990, 2000) and were kindly provided tion. The FECA recognizes two main types of campaign organizations: candidate campaign 11 A nal sample lter considered by Ansolabehere et al. (2006) concerns overwhelmed These market denitions are choice from Broadcast and Cable (1980, 1990, 2000) and were kindly provided in-11 or out-of-state, but such may imply the inclusion of counties in media markets by James Snyder. A nal sample lter considered by Ansolabehere et al. (2006) concerns overwhelmed These market denitions are from Broadcast and Cable (1980, 1990, 2000) and were kindly provided by James Snyder. committees and party committees; and two main sources of funds: individuals and interest states. In these states,by only a state. small percentage of the a media market by James Snyder. not clearly dominated any A large x, on the population other hand,lives mayin exclude too many states. In these states, only a small percentage of the population lives in a media market 13 . groups that is dominated in state. Politicians and news neglect a large counties. As in Ansolabehere et al. (2006) I13 usebroadcasts two-thirdsprobably (x = 66) do as not as the benchmark 13 that is dominated in state. Politicians and news broadcasts probably do not neglect a large 13 Under the Act, organizations wishing to contribute to federal candidates and parties must percentage of voters in this case. For this reason, I drop all counties in states where threshold, and verify in the robustness of my toI other of x. in states where less percentage of voters this case. For thisresults reason, drop values all counties less create PACs (the refers to live them as separate and segregated funds14 ). Moreover, than two-thirds of FECA the population in in-state dominated media markets. A second important issue is that the boundaries of television markets change over time, as than two-thirds of the population live in in-state dominated media markets. organizations may not give money directly to the PAC for the purpose of contributing to a new media markets are created, and old ones cease to exist by splintering or slow absorption federal campaign. The organization, however, as the sponsor of the PAC (i.e., its connected Campaign contributions into other media markets. Moreover, as noted above, I only have denitions of media markets Campaign contributions organization), may absorb all the costs of establishing and operating the committee and once per decade. Hence, in basicinterest analysis, I include counties that wereat dominated bycorthe Campaign contributions bythe special groups have been regulated least since Campaign contributions interest groups have been regulated at least since corsoliciting contributions toby it special from individuals. porate contributions to parties and candidates were banned by Congress in 1911. This porate contributions to parties candidates were in 1911. Also, individuals, PACs, and and party committees arebanned subject by to Congress contribution limits. This The prohibition was later extended to labor unions and trade associations. These rules were 11 These market denitions are from Broadcast and not Cable (1980, 2000) and were kindly provided prohibition was later extended toare labor unions and trade1990, associations. These rules were constraints on PAC contributions typically binding. Ansolabehere et al. (2003) note by James Snyder. and by the early 1970s many unions and organizations had formed Political weakly enforced weakly enforced and by the early 1970s of many unions and organizations had formed Political 12 I veried the robustness to the inclusion all counties and found similar results. Action Committees (PACs) to collect contributions for candidates without violating the law. 13 Other sources of funds and to campaigning are also allowed are relatively unimportant in the practice. Action Committees (PACs) collect contributions for but candidates without violating law. 13 Parties and candidates also give to each other, individuals and groups can raise theirmandating own money In 1974, the Federalmay Election Campaign Actand (FECA) sanctioned this practice, 15 In theown Federal Election Campaign Act (FECA) practice, mandating and1974, run their advocacy or independent campaigns on behalfsanctioned of or againstthis individual candidates. the disclosure of contributions to the FEC. 14 The press and refer to all noncandidate committees as PACs or political action comthe disclosure of public contributions tononparty, the FEC.

states. In these states, only a small percentage of the population lives in a media market that is dominated in state. Politicians and news broadcasts probably do in not a large states. In these states, only a small percentage of the population lives a neglect media market that that is is dominated dominated in in state. state. Politicians Politicians and and news news broadcasts broadcasts probably probably do do not not neglect neglect a a large large percentage of voters in this case. For this reason, I drop all counties in states where less that is dominated in state. Politicians and news broadcasts probably do not neglect a large percentage of voters in this case. For this reason, I drop all counties in states where percentage of voters in this case. For this reason, I drop all counties in states where less less than of the live in-state dominated media markets. percentage of voters inpopulation this case.12 For this reason, I drop all counties in states where less same two-thirds state throughout the period . in than two-thirds of the population live in in-state dominated media markets. than of the population live in-state dominated media markets. same two-thirds state throughout the period12 . in than of the population live in-state dominated media concerns markets. overwhelmed A two-thirds nal sample lter considered by in Ansolabehere et al. (2006) A nal sample lter considered by Ansolabehere et al. (2006) concerns overwhelmed Campaign contributions Campaign contributions states. In these states, only a small percentage of the population lives in a media market Campaign contributions states. In these states, only a small percentage of the population lives in a media market Campaign contributions that is dominated in state. and news broadcasts probably do not neglect a large Campaign contributions byPoliticians special interest interest groups have been been regulated at least least since corCampaign contributions by special groups have regulated at since corthat is dominated in state. Politicians and news broadcasts probably do not neglect a large Campaign contributions by special interest groups have been regulated at least since corpercentage of voters in case. For this reason, I drop all counties inat states where less porate contributions contributions to this parties and candidates were banned by Congress Congress in 1911. 1911. This Campaign contributions by special interest groups have been regulated least since This corporate to parties and candidates were banned by in percentage of voters in this case. For this reason, I drop all counties in states where less porate contributions to parties and candidates were banned by Congress in 1911. This that only 4 was percent of all PACto contributions to House and Senate candidates are at or than two-thirds of the population live in in-state dominated media markets. prohibition later extended labor unions and trade associations. These rules were porate contributions to parties and candidates were banned by Congress in 1911. This prohibition was later extended to labor unions and trade associations. These rules were 15 than two-thirds the extended population live in in-state dominated media prohibition was of later labor unions and trade is associations. These rules were ,the and theto average PAC contribution muchmarkets. smaller ($1,700 in their near the $10,000 limit weakly enforced and by early 1970s many unions and organizations had formed Political prohibition was and laterby extended to labor unions and and trade associations. These rules were weakly enforced the early 1970s many unions organizations had formed Political weakly enforced and by the earlycandidate 1970s many unions and organizations had formed Political sample). The Act also limited and group campaign expenditures, but in 1976 Campaign contributions Action Committees Committees (PACs) to collect contributions for candidates without violating the law. weakly enforced and(PACs) by the early 1970s many unions and organizations had formed Political Action to contributions for candidates without violating the Campaign contributions Action Committees (PACs) to collect collect contributions for candidates without violating the law. law. the Supreme Court struck down spending limits as a violation of free speech in Buckley v. In 1974, 1974, the Federal Election Campaign Act (FECA) sanctioned this practice, practice, mandating Action Committees (PACs) to collect contributions for candidates without violating the law. Campaign contributions by special interest groups have been regulated at least since corIn the Federal Election Campaign Act (FECA) sanctioned this mandating 16 the Federal Election Campaign Act (FECA) sanctioned this practice, mandating In 1974, . contributions Valeo Campaign by special interest groups have been regulated at leastmandating since corthe disclosure of contributions contributions toand the FEC. In 1974, the Federal Election Campaign Act (FECA) sanctioned practice, porate contributions to parties candidates were banned by this Congress in 1911. This the disclosure of to the FEC. the disclosure of contributions toand the candidates FEC. As Ansolabehere et al. (2003) note, two loopholes inbanned the FECA constraints have received porate contributions to parties were by campaign Congress in 1911. This Ansolabehere et al. (2003) summarize the main features this nance regulathe disclosure of later contributions to the FEC. prohibition was extended to labor unions and trade of associations. These rules were Ansolabehere et al. (2003) summarize the main features of this campaign nance regulaAnsolabehere et al. (2003) summarize the main features of this campaign nance regulacriticism. The created by main the Buckley v. Valeo court decision, candidate is These that individuals, prohibition was rst, later extended to labor unions and trade associations. rules were tion. The FECA recognizes two types of campaign campaign Ansolabehere et al. summarize the main features of this campaign nance regulaweakly enforced and by(2003) the early 1970s many unions and organizations: organizations had formed Political tion. The FECA recognizes two main types of campaign organizations: candidate campaign tion. The FECA recognizes two main types campaign candidate campaign groups, and corporations are permitted to of spend unlimited amounts of independent exweakly enforced and by the early 1970s many unions and organizations: organizations had formed Political committees and party committees; and two main sources of funds: individuals and interest tion. The FECA recognizes two main types of campaign organizations: candidate campaign Action Committees (PACs) to collect contributions for candidates without violating the law. committees and party committees; and two main sources of funds: individuals and interest committees and party committees; and two main as sources of funds: individuals and the interest penditures on behalf of or to against acontributions candidate, long as they without are not violating coordinated with 13 Action Committees (PACs) collect for candidates law. . groups 13 committees and party committees; and two main sources of funds: individuals and interest In 1974, the Federal Election Campaign Act (FECA) sanctioned this practice, mandating . groups 13 . groups the1974, candidate or party campaigns. The to second, created by a series of FEC rulings in must 1978 13 the In Election Campaign Act (FECA) sanctioned this practice, mandating Under theFederal Act, organizations wishing contribute to federal candidates and parties . the groups the Under disclosure of contributions towishing the FEC. Act, organizations to contribute to federal candidates and parties must the Act, organizations wishing to contribute toand federal candidates and must andUnder 1979, created a distinction between hard money soft money, making spending 14 parties the disclosure of contributions to the FEC. ). Moreover, create PACs (the FECA refers to them as separate and segregated funds 14 Under the(the Act, organizations wishing to contribute to federal candidates and parties must Ansolabehere et al. (2003) summarize the main features of this campaign nance regula). Moreover, create PACs FECA refers to them as separate and segregated funds 14 ). Moreover, create PACs (the FECA refers to them as separate and segregated funds limits applicable to the former. Soft money, raised through national party organizations 14 Ansolabehere et al. give (2003) summarize the mainPAC features of this campaign nance regulaorganizations may not money directly to for the purpose of contributing to a ). Moreover, create PACs (the FECA refers them as separate and segregated funds tion. The FECA recognizes two to main types of the campaign organizations: candidate campaign organizations may not give money directly to the PAC for the purpose of contributing to a organizations may not give money directly to the PAC for the purpose of contributing to a for non-federal accounts, may be spent on of non-federal election activities. Individuals and tion. The FECA recognizes two main types campaign organizations: candidate campaign federal campaign. The organization, however, as the sponsor of the PAC (i.e., its connected organizations may not give money directly to the PAC for the purpose of contributing to a committees and party committees; and two main sources of of funds: individuals and interest federal campaign. The organization, however, as the sponsor the PAC (i.e., its connected federal campaign. The organization, however, as the sponsor of the PAC (i.e., its connected groups may give unlimited amounts to non-federal party funds for the purpose of party committees and may party committees; and two main sources of of funds: individuals interest organization), absorb all the costs of establishing and operating the committee and 13 federal campaign. The organization, however, as the sponsor the PAC (i.e., itsand connected . groups organization), may absorb all the costs of establishing and operating the committee and 13 organization), may absorb all the costs of establishing and operating the committee and building .activities. Such funds, conceived to strengthen party organizations in the individgroups soliciting contributions to it from individuals. organization), may absorb all the costs of establishing and operating the committee and Under contributions the Act, organizations wishing to contribute to federal candidates and parties must soliciting to it from individuals. soliciting contributions to it from individuals. ual Also, states, have in practice been used by the national party organizations to raise money. Underindividuals, the Act, organizations wishing to contribute tosubject federal to candidates and parties must PACs, and party are contribution The soliciting contributions torefers it from individuals. Moreover, create PACs (the FECA to themcommittees as separate and segregated funds14 ).limits. Also, individuals, PACs, and party committees are subject to contribution contribution limits. The 14 limits. must Also, individuals, PACs, and party committees are subject to The Although unlimited in amount, independent expenditures and soft party donations ).limits. Moreover, create PACs (the FECA refers to them as separate and segregated funds constraints on PAC contributions are typically not binding. Ansolabehere et al. (2003) note Also, individuals, PACs, and party committees are subject to contribution The organizations give directly to the for the purpose of et contributing to a constraints onmay PACnot contributions are typically notPAC binding. Ansolabehere al. note 17 money constraints on PAC contributions are typically not binding. Ansolabehere et al. (2003) (2003) note . money A third loophole allows legislators to set upof leadership PACs still be publicly disclosed organizations may not give directly to the PAC for the purpose contributing to a 12 constraints on contributions arehowever, typically not binding. et al. note thePAC robustness to the inclusion of all counties and found Ansolabehere similar results. federal campaign. The organization, as the sponsor of the PAC (i.e., its (2003) connected 12 I veried veried the robustness to up the to inclusion of all counties and found similar results. 12 13 I which allow donors to give $10,000 to a candidate, but such funds cannot be used on I veried robustness to the inclusion of all also counties and found similar results. Other sources of The funds and campaigning are allowed but areof relatively unimportant in practice. federal campaign. however, as the sponsor the PAC (i.e., its connected 13 12 that only 4 the percent oforganization, all PAC contributions to House and Senate candidates are at or sources of funds and campaigning are also allowed but aresimilar relatively unimportant in practice. 13 Other organization), may absorb all the costs of establishing and operating the committee and I veried the robustness to the inclusion of all counties and found results. Other of funds and give campaigning are also allowed but are unimportant practice. Parties and sources candidates may also to each other, and individuals andrelatively groups can raise their in own money that candidates campaigns. 13 Parties and candidates may also give to each other, and individuals and groups can raise their own money 15 and campaigning organization), may absorb all the costs ofalso establishing and operating the committee and Other sources of funds are allowed but are relatively unimportant practice. , also and the average PAC contribution is much smaller in their near the $10,000 limit Parties and candidates may give to each other, and individuals and groups can raise($1,700 their in own money and run their own advocacy or independent campaigns on behalf of or against individual candidates. soliciting contributions to from individuals. and runand their own advocacy or it independent campaigns on behalf of or against individual candidates. Parties candidates may also give to each other, and individuals and groups can raise their own money 14 The data on PAC contributions from the FEC provides itemized contributions containing and own or it independent campaigns behalf of or against individual candidates. Thetheir press andadvocacy public refer tofrom all nonparty, noncandidate committees as PACs or political action comsoliciting contributions to individuals. 14run sample). The Act also limited candidate and on group campaign expenditures, but in 1976 Thetheir press and public refer toand all nonparty, noncandidate committees as PACs or political action com14 Also, individuals, PACs, party committees are subject to contribution limits. The and run own advocacy or independent campaigns on behalf of or against individual candidates. The press and public refer to all nonparty, noncandidate committees as PACs or political action committees. However, the Act and Commission Regulations distinguish between two types of PACs: separate each contribution or independent expenditure made by a PAC, party committee, candidate 14 mittees. However, the Act and Commission Regulations distinguish between two types of PACs: separate Also, individuals, PACs, and party committees are subject to contribution limits. The The press and public refer to all nonparty, noncandidate committees as PACs or political action comthe Supreme Court struck spending limits asbinding. a violation of nonconnected free speech in Buckley v. mittees. However, the Act and down Commission Regulations distinguish between two types PACs: separate segregated funds described above and nonconnected political committees (or PACs). A note nonconstraints on PAC contributions are typically not Ansolabehere etofal. (2003) segregated funds described above and nonconnected political committees (or nonconnected PACs). A nonmittees. However, the Act and Commission Regulations distinguish between two types of PACs: separate committee, or other federal committee to a candidate during each two-year election cycle. segregated funds described above andpay nonconnected political committees (or nonconnected PACs). A note nonconnected committee must for its own administrative expenses, using et theal. contributions it 16 political constraints on PAC contributions are typically not binding. Ansolabehere (2003) . political Valeo connected committee pay for of itsall own administrative using the contributions it 12 segregated funds above and nonconnected political committees (or nonconnected PACs). A nonI veried thedescribed robustness to must the inclusion counties and foundexpenses, similar results. connected political committee must pay for its own administrative expenses, using the contributions it raises. If an organization spends funds to establish or support a nonconnected PAC, these expenditures are Types of expenditures range from independent expenditures for and against candidates, coraises. If an organization spends funds to establish or support a nonconnected PAC, these expenditures are 12 13 I veried connected political committee must pay for subject its own administrative expenses, using the expenditures contributions it the robustness to the inclusion of all counties and found similar results. Other sources of funds and campaigning are also allowed are relatively unimportant in practice. As Ansolabehere et al. (2003) note, two loopholes in the FECA constraints have received raises. If an organization spends funds to are establish or support a but nonconnected PAC, these are considered contributions to the PAC and to the dollar limits and other requirements of the Act. considered contributions to the PAC and are subject to the dollar limits and other requirements of practice. the Act. 13 raises. If an organization spends funds to establish or support a nonconnected PAC, these expenditures are Other sources of funds and campaigning are also allowed but are relatively unimportant in ordinated expenditures, nancial contributions to candidates, and in-kind contributions. For Parties and candidates may also give to each other, and individuals and groups can raise their own money considered contributions to the PAC and are subject to the dollar limits and other requirements of the Act. criticism. The rst,may created by the Buckley v. Valeo court decision, is that individuals, considered contributions to the PAC and are subject to the dollar of limits and other requirements of the Act. Parties candidates also give to each other, and individuals and groups can raise their own money and runand their own advocacy or independent campaigns on behalf or against individual candidates. each interest-group PAC, I aggregate all nancial contributions (24K), and in-kind contribu14 14run their own advocacy or independent campaigns on behalf of or against individual candidates. and The press and public refer to all nonparty, noncandidate committees as PACs or political action comgroups, and corporations are permitted to spend unlimited amounts of independent ex14 14 distinguish 14 The press public to allThe nonparty, noncandidate committees as PACs or political action comtion made to and candidates (24Z). most important component are 24Ks. In analysis, I mittees. However, the Actrefer and Commission Regulations between two types of my PACs: separate 14 penditures on behalf orCommission against a Regulations candidate, as committees long as they are coordinated with mittees. However, the Actof and distinguish between two not types of PACs: segregated funds described above and nonconnected political (or nonconnected PACs).separate A nonexclude independent expenditures for candidates (24E), because(or a candidate can more A easily segregated funds or described andpay nonconnected political committees nonconnected PACs). nonconnected political committee must for its own administrative using the contributions it the candidate party above campaigns. The second, created by expenses, a series of FEC rulings in 1978 connected political committee must pay for its own administrative expenses, using the expenditures contributions it distance himself from such contributions (as noted, byalaw these contributions should not are be raises. If an organization spends funds to establish or support nonconnected PAC, these and 1979, created a distinction between hard money and soft money, making spending raises. If an organization spends funds to are establish orto support a nonconnected PAC, these expenditures are considered contributions to the PAC and subject the dollar limits and other requirements of the Act. 15 considered contributions to the PAC and are subject to raised the dollar limits and other requirements of the Act. limits applicable to the former. Soft money, through national party organizations This refers to the pre-2002 limit, relevant to my analysis. On Nov 2002, the Bipartisan Campaign Reform Act increased the contribution for an 14 for non-federal accounts, may be limits. spent See on http://www.opensecrets.org/overview/limits.php non-federal election activities. Individuals and overview of the changes. 14 16 groups may give unlimited spending amounts to non-federal party funds purpose party An exception is presidential limits, which survived because theyfor are the voluntary: any of candidate who wishes to receive federal must abide by limits. building activities. Such funding funds, conceived to the strengthen party organizations in the individ17 In 2002, amendments to the FECA by the Bipartisan Campaign Reform Act placed restrictions on soft ual states, havethe in limits practice been used by the national party organizations to raise money. money but raised on hard money party contributions. Although unlimited in amount, independent 16 expenditures and soft party donations must 17 allows legislators to set up leadership PACs still be publicly disclosed . A third loophole15

near the $10,000 limit15 , and the average PAC contribution is much smaller ($1,700 in their sample). The Act also limited candidate and group campaign expenditures, but in 1976 the Supreme Court struck down spending limits as a violation of free speech in Buckley v. Valeo16 . As Ansolabehere et al. (2003) note, two loopholes in the FECA constraints have received criticism. The rst, created by the Buckley v. Valeo court decision, is that individuals, that only 4 percent of allare PAC contributions to House andamounts Senate candidates are at exor groups, and corporations permitted to spend unlimited of independent 15 , and the average PAC contribution much smaller ($1,700 in their near the $10,000 limit of penditures on behalf or against a candidate, as long asis they are not coordinated with sample). The or Act alsocampaigns. limited candidate and group campaign expenditures, but in 1976 the candidate party The second, created by a series of FEC rulings in 1978 the Supreme Court down spending limits as a violation of money, free speech in Buckley v. and a struck distinction between hard money and soft making spending that 1979, only created 4 percent of all PAC contributions to House and Senate candidates are at or . Valeo16 15 former. Soft money, raised through national party organizations limits applicable to the , and the average PAC contribution is much smaller ($1,700 in their near the $10,000 limit As Ansolabehere et al. (2003) note, two loopholes in the FECA constraints have received for non-federal accounts, may be spent on and non-federal election activities. Individuals and sample). The Act also limited candidate group campaign expenditures, but in 1976 criticism. The rst, created by the Buckley v. Valeo court decision, is that individuals, groups may give unlimited amounts to non-federal funds purpose of party the Supreme Court struck down spending limits as aparty violation of for free the speech in Buckley v. groups, and corporations are permitted to spend unlimited amounts of independent exbuilding Valeo16 . activities. Such funds, conceived to strengthen party organizations in the individpenditures on behalf of or been against a candidate, as long as they are not coordinated with ual states, have in practice used by the national organizations to have raisereceived money. As Ansolabehere et al. (2003) note, two loopholes in party the FECA constraints the candidate or party campaigns. The second, created by and a series FEC donations rulings in must 1978 Although amount, expenditures softof party criticism. unlimited The rst, in created by independent the Buckley v. Valeo court decision, is that individuals, and be 1979, created a distinction between hard money and soft making spending 17 . A permitted third loophole allows legislators to money, set up PACs still disclosed groups, publicly and corporations are to spend unlimited amounts ofleadership independent exlimits applicable to the former. Soft money, raised through national party organizations which allow donors to give to $10,000 to a candidate, suchare funds be used on penditures on behalf of orup against a candidate, as long but as they notcannot coordinated with for non-federal accounts, may be spent on non-federal election activities. Individuals and that candidates the candidate or campaigns. party campaigns. The second, created by a series of FEC rulings in 1978 groups may give unlimited amounts to the non-federal partyitemized funds for the purpose of party data on PAC contributions from FEC provides contributions containing andThe 1979, created a distinction between hard money and soft money, making spending building activities.orSuch funds, conceived to strengthen in the individeach independent expenditure by aparty PAC,organizations party committee, candidate limitscontribution applicable to the former. Soft money,made raised through national party organizations ual states, have in practice been used by the national during party organizations to raise money. committee, or other federal committee candidate each two-year election cycle. for non-federal accounts, may be spent to onanon-federal election activities. Individuals and Although unlimited in amount, independent expenditures and soft party donations must Types expenditures range amounts from independent expenditures for and candidates, cogroups of may give unlimited to non-federal party funds foragainst the purpose of party 17 . A third loophole allows legislators to set up leadership PACs still be publicly disclosed ordinated expenditures, contributions to candidates, and in-kind contributions. For building activities. Suchnancial funds, conceived to strengthen party organizations in the individwhich allow donors to give up to $10,000 to a candidate, but such funds cannot be used on each interest-group PAC, I aggregate allby nancial contributions (24K), and in-kind contribuual states, have in practice been used the national party organizations to raise money. that candidates campaigns. tion made to candidates (24Z). The most important component 24Ks. my analysis, I Although unlimited in amount, independent expenditures and are soft party In donations must The data on PAC contributions from the FEC provides itemized contributions containing 17 exclude independent expenditures forloophole candidates (24E), becauseto a candidate can more easily . A third allows legislators set up leadership PACs still be publicly disclosed each contribution or independent expenditure made by a PAC, party committee, candidate distance himself from by lawbut these contributions should not on be which allow donors tosuch give contributions up to $10,000(as to noted, a candidate, such funds cannot be used committee, or other federal committee to a candidate during each two-year election cycle. coordinated withcampaigns. the candidate). that candidates 15 This to the pre-2002 to my expenditures analysis. On Nov the Bipartisan Campaign Types of refers expenditures range limit, from relevant independent for 2002, and against candidates, coTheAct data on PAC contributions from the FEC provides itemized contributions containing Reform increased the contribution limits. See http://www.opensecrets.org/overview/limits.php for an ordinated nancial contributions to candidates, and in-kind contributions. For overview of expenditures, the contributions changes. Classifying each contribution or independent expenditure made by a PAC, party committee, candidate 16 exception is presidential spending limits, which survived because they are voluntary: anycontribucandidate eachAn interest-group PAC, I aggregate all nancial contributions (24K), and in-kind committee, or other federal committee to a candidate during each two-year election cycle. who wishes to receive federal funding must abide by the limits. The FEC provides a very rough ve-fold categorization of PAC contribution data by industry: 17 made to candidates (24Z). The most important component are 24Ks. In my analysis, I tion In 2002, amendments to the FECA by the Bipartisan Campaign Reform Actagainst placed restrictions on soft Types of expenditures range frommembership independent expenditures for and candidates, cocorporation, labor organization, organization, trade association, cooperative, money but raised the limits on hard money party contributions. exclude independent expenditures for candidates (24E), because a candidate can more easily ordinated expenditures, contributions to categories candidates, contributions. For and corporation withoutnancial capital stock. These are ofand the in-kind sponsoring (or connected) distance himself from such contributions (as noted, by law these contributions should not be each interest-group PAC, I aggregate all nancial and in-kind contribu18 15 . Since these organization for the committee, provided on thecontributions statement of (24K), organization 15 made to candidates (24Z). The most important component are 24Ks. In my analysis, I tion This refers the pre-2002 limit, relevant to my for analysis. On NovPolitics 2002, the Bipartisan Campaign categories are to too broad, I rely on the Center Responsive (CRP) classication Reform Act increased the contribution limits. See http://www.opensecrets.org/overview/limits.php for an exclude to independent expenditures for of candidates (24E), because a candidate can more easily system identify the interest group each PAC. overview of the changes. 16 distance himself from such contributions (as noted, by law these contributions should not be An exception is presidential spending limits, which survived because they are voluntary: any candidate The CRP coding system is hierarchical, with ve super-categories (Business, Labor, who wishes to receive federal funding must abide by the limits. 17 15 In 2002, amendments to Other Ideological/Single-Issue, and Unknown) into 13 about 100 industhe FECA by the Bipartisan Campaign Act the placed restrictions on soft This refers to the pre-2002 limit, relevant to my divided analysis. On Reform Novsectors, 2002, Bipartisan Campaign money but raised the limits on hard money party contributions. Reform Act increased the contribution limits. See http://www.opensecrets.org/overview/limits.php for an tries, and about 400 categories. The CRP has been classifying PACs since 1990, whereas I overview of the changes. 16 study contributions and election results since 1980. To classify PAC contribution Ancampaign exception is presidential spending limits, which 15 survived because they are voluntary: any candidate who wishes to receive federal funding by First, the limits. money before 1990, I proceed inmust two abide steps. 17 I check whether the pre-1990 contribut17 In 2002, amendments to the FECA by the Bipartisan Campaign Reform Act placed restrictions on soft ing PAC Ifmoney so, I party assign the sector allocated by CRP to the respective money but survives raised the after limits 1990. on hard contributions.

coordinated with the candidate). coordinated with the candidate). coordinated with the candidate). Classifying contributions Classifying contributions Classifying contributions The FEC provides a very rough ve-fold categorization of PAC contribution data by industry: coordinated with the candidate). The FEC provides a very rough ve-fold categorization of PAC contribution data by industry: The FEC provides a very rough ve-fold categorization of PAC contribution data cooperative, by industry: corporation, labor organization, membership organization, trade association, corporation, labor organization, membership organization, trade association, cooperative, corporation, labor organization, membership organization, trade association, cooperative, and corporation without capital stock. These are categories of the sponsoring (or connected) Classifying contributions and corporation without capital stock. These are categories of the sponsoring (or connected) 18 and corporation without capital stock. These are categories of the sponsoring (or connected) these organization for the committee, provided on the statement of organization18 . Since . Since these organization for the committee, provided on the statement of organization 18 by industry: The FEC provides a very rough ve-fold categorization of PAC contribution data Since these organization the committee, provided on the ofPolitics organization categories arefor too broad, I rely on the Center for statement Responsive (CRP). classication categories are too broad, I rely on the Centerorganization, for Responsive Politics (CRP) classication corporation, labor organization, membership trade association, cooperative, categories are too the broad, I rely on the for Responsive Politics (CRP) classication system to identify interest group of Center each PAC. system to identify the interest group ofThese each PAC. and corporation without capital stock. are categories of the sponsoring (or connected) system identify the system interest group of each PAC. The to CRP coding is hierarchical, with ve super-categories (Business, Labor, 18 The CRP coding system is hierarchical, withstatement ve super-categories (Business, Labor, . Since these organization for the committee, provided on the of organization The CRP coding system is hierarchical, with ve super-categories (Business, Labor, Ideological/Single-Issue, Other and Unknown) divided into 13 sectors, about 100 indusIdeological/Single-Issue, Unknown) into 13 sectors, aboutclassication 100 induscategories are too broad,Other I relyand on the Center divided for Responsive Politics (CRP) Ideological/Single-Issue, Other and Unknown) into 13 sectors, about 100 industries, and about 400 categories. The CRP hasdivided been classifying PACs since 1990, whereas I tries, and about 400 categories. Theof CRP has been classifying PACs since 1990, whereas I system to identify the interest group each PAC. tries, and about 400 categories. The CRP has been classifying PACs since 1990, whereas I study campaign contributions and election results since 1980. To classify PAC contribution study campaign contributions and election results since 1980. To classify PAC contribution The CRP coding system is hierarchical, with ve 1980. super-categories (Business, Labor, study campaign contributions election To classify PAC contribution money before 1990, I proceed and in two steps.results First, since I check whether the pre-1990 contributmoney before 1990, I proceed in two steps. First, I check whether the pre-1990 contributIdeological/Single-Issue, Otherin and Unknown) divided into whether 13 sectors, about 100 indusmoney before 1990, I proceed two I check pre-1990 ing PAC survives after 1990. If so, I steps. assign First, the sector allocated bythe CRP to the contributrespective ing PAC survives after 1990. If so, I assign thebeen sector allocated by CRP to the respective tries, and about 400 categories. The CRP has classifying PACs since 1990, whereas I ing PAC survives 1990. If so, assign the sector allocated CRP to the respective PAC to the earlierafter period. Next, for Ithe set of unmatched PACs, by I follow the guidelines for PAC to the earlier period. Next, for the setresults of unmatched PACs, I follow PAC the guidelines for study campaign contributions and election since 1980. To classify contribution PAC to the earlier from period. for the set of the unmatched PACs, I follow the guidelines for to identify committees interest group. As shown in PAC classication theNext, CRP19 19 to steps. identify the committees interest group. Ascontributshown in PAC classication from the CRP 19 money before 1990, I proceed in two First, I check whether the pre-1990 to identify the committees interest group. As shown in PAC classication from the CRP greater detail in the Data Appendix, the rst step allows me to classify more than 90% of the greater detail in the Data Appendix, the rst step allows me to classify more than 90% of the ing PAC survives after 1990. If so, I assign the sector by CRP to the respective greater detail in the Data Appendix, rst allowsallocated me to money). classify more than of the pre-1990 contributions (and a larger the share of step the contribution With the 90% additional pre-1990 contributions (and a larger share ofof the contribution money). With the additional PAC to the earlier period. Next, for the set unmatched PACs, I follow the guidelines for pre-1990 contributions a contributions larger share ofwere the contribution money). With the additional classication, over 95% (and of the typically matched. 19 classication, over from 95% of the contributions were matched. to identify thetypically committees interest group. As shown in PAC classication the CRP classication, over 95% of the contributions were typically matched. The CRP classication system is likely to match the publics perception of the interest The detail CRP in classication system isthe likely to match theme publics perception of the interest greater the Data Appendix, rst step allows to classify more than 90% of the The CRP classication system is likely to match the publics perception the interest group behind contributions, not just because it is based on the PAC namesof and sponsors group behind contributions, not justshare because it is based on the PAC names and sponsors pre-1990 contributions (and a larger of the contribution money). With the additional group behind contributions, just because it because is basedthe on general the PAC names and sponsors (the information easily visiblenot to voters) but also objective of the system (the information easily visible to voters) but were also because the general objective of the system classication, over 95% of the contributions typically matched. (the information easily visible to voters) but also because the general objective of the system and the Center is precisely to monitor money in politics and inform citizens about this. andThe the CRP Center is precisely to monitor money in politics and inform citizens about this. classication is likely to in match the and publics perception of the this. interest and the Center is precisely system to monitor money politics inform citizens about group behind contributions, not just because it is based on the PAC names and sponsors (the information easily visible to voters) but also because the general objective of the system and the Center is precisely to monitor money in politics and inform citizens about this.
18 18 These categories apply for the two kinds of apply for the two kinds of 18 These categories committees and qualied non-party committees.

committees studied in this paper: non-party, non-qualied committees studied in this paper: funds non-party, non-qualied The rst are separate segregated and nonconnected These categories apply for the two kinds of committees studied in this paper: funds non-party, non-qualied committees and qualied non-party committees. The rst are separate segregated and nonconnected committees that have not qualied ascommittees. multi-candidate committees, and may currently contribute up to $1,000 committees and qualied non-party The rst are separate segregated funds and nonconnected committees that have not qualied as multi-candidate committees, and may currently contribute up to $1,000 per candidate per election. Qualied non-party committess instead qualify as multi-candidate committees committees that have not qualied as multi-candidate committees, and may currently contribute up to $1,000 per candidate per election. Qualied non-party committess instead qualify as multi-candidate committees (to this end, they must be registered for 6 months, have received contributions from more thancommittees 50 people, per candidate per election. Qualied non-party committess instead qualify as multi-candidate (to this end, they must be 6 months, have received contributions from more than 50 people, and made contributions to registered at 5for federal candidates) and may currently contribute up to $5,000 per 18 (to this end, they must be registered for 6 months, have received contributions from more than 50 people, These categories apply for least the two kinds of committees studied in this paper: non-party, non-qualied and made contributions to at least 5 federal candidates) and may currently contribute up to $5,000 per candidate per election. and made contributions to at least 5 federal candidates) and may currently contribute up to $5,000 per committees and qualied non-party committees. The rst are separate segregated funds and nonconnected candidate per election. 19 http://www.opensecrets.org/action/ftm/index.php candidate per election. 19 committees that have not qualied as multi-candidate committees, and may currently contribute up to $1,000 19 http://www.opensecrets.org/action/ftm/index.php http://www.opensecrets.org/action/ftm/index.php per candidate per election. Qualied non-party committess instead qualify as multi-candidate committees (to this end, they must be registered for 6 months, have 16 received contributions from more than 50 people, 16 and may currently contribute up to $5,000 per and made contributions to at least 5 federal candidates) 16 candidate per election. 19 http://www.opensecrets.org/action/ftm/index.php

16
18

Table 1 presents descriptive statistics for the main variables of the analysis. The incumbent Table 1 presents descriptive statistics for the main variables of the analysis. The incumbent Table 1 presents descriptive foraverage the main variables of elections the analysis. incumbent share of the two-party vote statistics is 58% on (uncontested and The open seats are share of the two-party vote is 58% on average (uncontested elections and open seats are share of the two-party vote is for 58% on variable average is (uncontested elections open seats are not included). The maximum this 92%. In a few cases and incumbents receive not included). The maximum for this variable is 92%. In a few cases incumbents receive not included). maximum for this is 92%. In a few incumbents receive a very low vote The share (the minimum in variable the sample is 12.6%), but cases in most of the cases, the a very low vote share (the minimum in the sample is 12.6%), but in most of the cases, the a very low vote share (the minimum the (the sample is 12.6%), but in in most the cases, the incumbent receives a majority of thein vote median, not shown the of table, is 57.8%, incumbent receives a majority of the vote (the median, not shown in the table, is 57.8%, incumbent a majority the vote (the median, not shown in the table, 57.8%, and in just receives about one-fourth ofof the county-years the incumbent receives less thanishalf the and in just about one-fourth of the county-years the incumbent receives less than half the and in just about one-fourth of the county-years the incumbent receives less than half the votes). votes). votes). There are a total of 180 Senate races in the base sample, in which almost exactly half There are a total of 180 Senate races in the base sample, in which almost exactly half There are a total of are 180Democrats. Senate races inthe thePAC base level, sample, which Herndahl almost exactly half (89) of the incumbents At thein average index of (89) of the incumbents are Democrats. At the PAC level, the average Herndahl index of (89) of the incumbents are Democrats. At standard the PAC deviation level, the of average index of campaign contributions is about 0.02 (with 0.089)Herndahl for the incumbents campaign contributions is about 0.02 (with standard deviation of 0.089) for the incumbents campaign contributions is larger about 0 .02 (with standard deviation of for 0.089) for the incumbents in these races, and much (0 .15, standard deviation 0.24) challengers. A similar in these races, and much larger (0.15, standard deviation 0.24) for challengers. A similar in these is races, and much (0.15, standard deviation 0.24) for challengers. A of similar pattern observed at the larger industry level, where the contrast is between an average 0.06 pattern is observed at the industry level, where the contrast is between an average of 0.06 pattern is observed at the industry level, where the contrast is between an average of 0.06 for incumbents (with a standard deviation of 0.11) and 0.23 for challengers (with a standard for incumbents (with a standard deviation of 0.11) and 0.23 for challengers (with a standard for incumbents (with a standard raise deviation ofmore 0.11)money and 0.23 for challengers (with standard deviation of 0.23). Incumbents much (on average around $5 a million, in deviation of 0.23). Incumbents raise much more money (on average around $5 million, in deviation of 0 . 23). Incumbents raise much more money (on average around $5 million, in 2, 000 constant prices) than their challengers (around $3 million on average). Of this money, 2, 000 constant prices) than their challengers (around $3 million on average). Of this money, 2 000 constant prices) than their challengers (around $3 million on average). OfPAC this money, a, relatively small percentage comes from special interests. Over the period, money a relatively small percentage comes from special interests. Over the period, PAC money a relatively small percentage comes from special interests. Over the period, PAC money added on average $1 million for incumbents and about $300, 000 for challengers per election added on average $1 million for incumbents and about $300, 000 for challengers per election added There on average $1 million for incumbents about $300 , 000 for challengers per election cycle. is substantial variation both for and incumbents and challengers in the money raised cycle. There is substantial variation both for incumbents and challengers in the money raised cycle. There is substantial variation both for incumbents and challengers in the money raised and the extent of its concentration. and the extent of its concentration. andThe the base extent of itsincludes concentration. sample 1, 759 US counties, 86% of which are in-state. When restricting The base sample includes 1, 759 US counties, 86% of which are in-state. When restricting The base sample includes , 759 US counties, 86% of which are When restricting attention to out-of-state and 1 in-state and contiguous counties, thein-state. sample is reduced to 542 attention to out-of-state and in-state and contiguous counties, the sample is reduced to 542 attention with to out-of-state and in-state andas contiguous counties, the sample is reduced to 542 counties, roughly one-half classied in-state (56%). counties, with roughly one-half classied as in-state (56%). counties, with roughly one-half classied as in-state (56%).

2.3 2.3 2.3

Descriptive Statistics Descriptive Descriptive Statistics Statistics

3 3 3
3.1 3.1 3.1

Main results Main Main results results

Main results for the full sample of counties Main Main results results for for the the full full sample sample of of counties counties

I begin by estimating (1) for the sample of US Senate candidates in the period 1980-2002. I begin by estimating (1) for the sample of US Senate candidates in the period 1980-2002. I begin by estimating the 2. sample of USbefore, Senateall candidates the period The results are shown(1) in for Table As noted standardin errors in the 1980-2002. paper are The results are shown in Table 2. As noted before, all standard errors in the paper are The shown in Table 2. As and noted before, all standard inlevel. the paper fully results robust are against heteroskedasticity serial correlation at theerrors county Also, are all fully robust against heteroskedasticity and serial correlation at the county level. Also, all fully robust against heteroskedasticity and serial correlation at the county level. Also, all regressions include a full set of year (times party of the incumbent) xed eects, which are regressions include a full set of year (times party of the incumbent) xed eects, which are regressions include a tables full set of year (times party of the incumbent) xed eects, which are not presented in the to save space. not presented in the tables to save space. not In presented the tables to save indices space. are constructed at the PAC level. This is a useful columnsin 1-5, the Herndahl In columns 1-5, the Herndahl indices are constructed at the PAC level. This is a useful In columns 1-5, the Herndahl indices are constructed at the PAC level. This is a useful reference point. These indices rely solely on the 17 extent to which contributions come mostly 17 17 few PACs, and thus do not depend on the in the form of large contributions from relatively exact choice for classifying PACs by industry20 . Column 1 estimates (1) without any additional controls. The coecient (0.171) on hI 19 st is positive and signicant. This result indicates that in out-of-state counties, an increase in

reference point. These indices rely solely on the extent to which contributions come mostly reference point. These indices rely solely on the extent to which contributions come mostly reference point. These indices relyfrom solely on the extent to which contributions come mostly in the form of large contributions relatively few PACs, and thus do not depend on the in the form of large contributions from relatively few PACs, and thus do not depend on the 20 in the choice form of large contributions from relatively few PACs, and thus do not depend on the exact for classifying PACs by industry 20 . . exact choice for classifying PACs by industry 20 reference point. These indices rely solely on the to which contributions come . extent exact choice for classifying by industry Column 1 estimates (1)PACs without any additional controls. The coecient (0 .171) mostly on hI st Column 1 estimates (1) without any additional controls. The coe cient (0 . 171) on the hI I st in the form of large contributions from relatively few PACs, and thus do not depend on Column 1 estimates (1) without any additional controls. The coe cient (0 . 171) on h st is positive and signicant. This result indicates that in out-of-state counties, an increase in is positive and signicant. This result indicates that in out-of-state counties, an increase in 20 .with exact choice forsignicant. classifying PACs by industry is positive and result indicates that a inhigher out-of-state counties, an incumbent. increase in the incumbents HerndahlThis index is correlated vote share for the the Column incumbents Herndahl index is correlated with a higher vote share for the incumbent. 1 estimates (1) without any The coe cient (0.171) on hI the Herndahl index is correlated with controls. awith higher vote share fordeviation the incumbent. st Theincumbents estimated coe cient implies that anadditional incumbent a one-standard higher The estimated coe cient implies that an incumbent with a one-standard deviation higher is positive and signicant. This result indicates that in out-of-state counties, an increase in The estimated coe cient that an incumbent with a one-standard higher Herndahl index than the implies average in the sample (a rise in the index in 0.09, deviation see Table 1) will Herndahl index than the average in the sample (a rise in the index in 0 . 09 , see Table 1) will the incumbents index in is correlated a higher vote share for the incumbent. Herndahl index Herndahl than the average the samplewith (a rise in the index in 0.09 , see Table 1) will have an advantage of 1 .5 additional percentage points over his opponent in these counties. have an advantage of 1 . 5 additional percentage points over his opponent in these counties. The coe estimated coeh cient implies that an incumbent with one-standard deviation higher O have an advantage of 1.5 additional percentage points overahis opponent in these counties. The cient for O st , though also positive and signicant, is much smaller (0.078) and is not , though also positive and signicant, is much smaller (0 . 078) and is will not The coe cient for h O st Herndahl index than the average in the remaining sample (a rise in the indexsmaller in 0.09, see Table 1) also in positive and signicant, is much (0 .078) andresults is not The coe cient for hst , though robust to the inclusion of controls the columns. More important are the robust toadvantage the inclusion of controls in the remaining columns. More important are the results have an of 1 .5cients, additional percentage points over his opponent in these counties. robust to the inclusion of controls in the remaining columns. More important are the results for the interaction coe in hI and inhO . Column 1 estimates inhI = 0.171, for the interaction coe O cients, inhI and inhO . Column 1 estimates inhI = 0.171, , though also positive and signicant, is much smaller (0 . 078) and is not The coe cient hcoe in . Column 1 estimates 0 .171 , for the interaction cients, in st hI and hO in hI = indicating thatfor the positive correlation between the Herndahl for the incumbent and the indicating that the positive correlation between the Herndahl for the incumbent and the robust to the inclusion of controls in the between remaining columns. More important are results indicating that the positive correlation the Herndahl the incumbent and the incumbent vote share vanishes for in-state counties. The degree offor concentration ofthe the oppoincumbent vote share vanishes for in-state counties. The degree of concentration of the oppoinshare Column 1 of estimates inhIof = 0 .171 , for the interaction coe cients, in incumbent vote share vanishes for in-state counties. degree concentration the oppohI and hO . The nent does not seem to a ect the incumbent vote di erentially across media markets, as nent does not seem to a ect the incumbent vote share di erentially across media markets, as indicating that the for positive correlation between the Herndahl foracross the incumbent and the nent does estimate not seem to a ect the incumbent vote share dierentially media markets, as the point inhO , though positive, is small and statistically insignicant. That O , though positive, is small and statistically insignicant. That the point estimate for in h incumbent vote share in-state counties. The and degree of concentration of the oppoO , for though positive, is small statistically insignicant. That the point forvanishes for inh results areestimate present only the incumbents is reasonable given the well-established fact that results are present only for the incumbents is reasonable given the well-established fact that nent does not seem to a ect the incumbent vote share di erentially across media markets, as results are present for the incumbents is reasonable given the well-established fact that incumbents receiveonly more media coverage than their challengers (see, e.g. Kahn (1993)). incumbents receive for more coverage than their challengers (see, e.g. insignicant. Kahn (1993)). though positive, is small and statistically That the Columns point estimate inmedia incumbents receive more media coverage than their challengers (see, e.g.inclusion Kahn (1993)). hO , robustness 2-5 examine the of the previous results to the of several Columns 2-5 examine the robustness of the previous results to the inclusion of several results are present only forthe therobustness incumbents reasonable given the well-established fact that Columns 2-5 examine ofisthe previous results to the inclusion several additional controls. Concerns include whether the patterns of concentration of of campaign additional controls. Concerns include whether the challengers patterns of (see, concentration of campaign incumbents receive more media coverage than their e.g. Kahnin (1993)). additional controls. Concerns include whether the patterns of concentration of campaign contributions merely mirror the pattern of concentration of economic activity each state contributions merely mirror the pattern of concentration of economic activity in each state Columns 2-5 examine the robustness of the previous results to the inclusion of several contributions mirror the pattern concentration of economic in incumbent each state at time t, andmerely whether a change in such of degree of concentration may activity aect the at time t, and whether a change in such degree of concentration may a ect the incumbent additional controls. Concerns include whether patterns of to concentration ofincumbent campaign at time t, and whether a change in such degree the ofpotential concentration may aect the vote share for reasons unrelated to the candidates link interest groups. For such vote share for reasons unrelated to the candidates potential link to interest groups. For state such contributions merely mirror the pattern of concentration of economic activity in each vote share for reasons unrelated to thewould candidates potential link to interest groups. For such a story to drive the results the e ect need to dier across the two types of counties, a story to drive the results the e ect would need to dier across the types counties, at time t, and whether a change in such degree of concentration maytwo a theof incumbent a story drive the results the e ect would need to dier across the two types of which isto not obvious. Still, to guard against this possibility, in column 2 I ect include a counties, measure which is not obvious. Still, to guard against this possibility, in column 2 I include a For measure vote share for reasons unrelated to the candidates potential link to interest groups. such which is not obvious. Still, to guard against this possibility, 2 I include a measure of the degree of concentration of economic activity in state sin atcolumn time t (the Herndahl index of the degree of concentration of economic activity in state s at time t (the Herndahl index a story toState drive the results ect would need as er across thet two of counties, of the degree of Product concentration ofesectoral economic activity indi state s atinteraction time (thetypes Herndahl index of Gross at the the level) as to well its with the in-state of Gross State Product at the sectoral level) as well as its interaction with the in-state which is not obvious. Still, to guard against this possibility, in column 2 I include a measure of Gross State Product at the sectoral level) as well coe as its interaction with the in-state dummy. These variables are signicant, but the main cient of interest (in hI ) is still I ) is still dummy. These variables are signicant, but the main coe cient of interest ( in h of the degree of concentration of economic activity in state s cient at time t (the Herndahl index still dummy. These variables are signicant, the main coe of interest (inhI ) is negative and statistically signicant and,but moreover, the point estimate is remarkably stable negative and statistically signicant and, moreover, the point estimate is remarkably stable of Gross State Product at the sectoral as well its estimate interaction with the in-state negative statistically signicant and, level) moreover, the as point is remarkably stable ( 0 .168).and ( 0 . 168). still dummy. These variables are signicant, the main cient interest (inhI ) is of 0A .168). ( more serious concern may be that but the results arecoe driven byof other characteristics a A more serious concern may be that the results are driven by other characteristics of a negative and statistically signicant and, moreover, the point estimate is remarkably stable A more serious concern may be that the results are driven by other characteristics of a 20 second reason why this benchmark is useful is the fact that, prior to 1990, the sector of a few PACs (20 0.A 168). second could reason why benchmark useful the fact Appendix, that, prior this to 1990, the sector problem of a few as PACs 20 Adataset in the not bethis established. Asis noted inis the Data is not a major the A second could reason why this benchmark is useful is the fact Appendix, that, prior this to 1990, the sector problem of a few as PACs in the dataset not be established. As noted in the Data is not a major the A dataset more serious concern may be that the results byHerndahl other of a success in classifying PAC was quite high. Still,are the driven PAC-level veries that results in the could not becontributions established. As noted in the Data Appendix, this is not a characteristics major problem as the
success classifying PAC contributions was quite high. Still, PAC-level Herndahl veries that results are not in driven from di erential success in classifying PACs for the dierent candidates or years. success classifying PAC contributions was quite high. Still, PAC-level Herndahl veries that results are20 not in driven from di erential success in classifying PACs for the dierent candidates or years. A second this benchmark is useful isPACs the fact prior to 1990, the sector of a few PACs are not driven reason from diwhy erential success in classifying for that, dierent candidates or years. in the dataset could not be established. As noted in the Data Appendix, this is not a major problem as the 18 success in classifying PAC contributions was quite high. 18 Still, the PAC-level Herndahl veries that results are not driven from dierential success in classifying18 PACs for dierent candidates or years.

18
20

negative and statistically signicant and, moreover, the point estimate is remarkably stable (0.168). A more serious concern may be that the results are driven by other characteristics of a candidates campaign contribution patterns. For instance, candidates with dierent Hernd20 candidates patterns. For candidates dierent HerndA secondcampaign reason whycontribution this benchmark is useful is theinstance, fact that, prior to 1990,with the sector of a few PACs candidates campaign contribution patterns. For instance, candidates with di erent Herndahl indices could also be candidates who receive and spend dithis erent sums of money in as their in the dataset could not be established. As noted in the Data Appendix, is not a major problem the candidates campaign contribution patterns. For instance, candidates with di erent Herndahl indices could also be candidates who receive and spend dierent sums of money in their ahl indices could also be candidates who receive and spend di erent sums of money in their success in classifying PAC contributions was quite high. Still, the PAC-level Herndahl veries that results campaigns. Moreover, since candidates subject contribution limits, may be a ahl indices could also be candidates who are receive and to spend dierent sums ofthere money in their campaigns. since candidates are subject contribution limits, there may be a are not driven Moreover, from dierential success in classifying PACs forto di erent candidates or years. campaigns. Moreover, since candidates are subject to contribution limits, there may be a mechanical negative relationship between the amount of money raised andthere the Herndahl campaigns. Moreover, since candidates are subject to contribution limits, may be a mechanical negative relationship between the amount of money raised and the Herndahl mechanical negative relationship between the amount of money raised and the Herndahl index, as raising more money may necessarily imply raising money from and a larger number of 18 mechanical negative relationship between the amount of money raised the Herndahl index, as raising more money may necessarily imply raising money from a larger number of index, as raising more money may necessarily imply raising money from a larger number of PACs (see, however, Ansolabehere et al. (2003) who show money that individual donors grow in index, as raising more money may necessarily imply raising from a larger number of PACs (see, however, Ansolabehere et al. (2003) who show that individual donors grow in PACs (see, however, Ansolabehere et al. (2003) who show that individual donors grow in importance relative to PAC money et as demand for campaign cashindividual increases). Thus, grow in ColPACs (see, however, Ansolabehere (2003) who show that donors in importance relative to PAC money as al. demand for campaign cash increases). Thus, in Colimportance relative to PAC money as demand for campaign cash increases). Thus, in Column 3 I include the total campaign receipts by the incumbent and his opponent, as well as importance relative PAC money as demand cash Thus, Column 3 I include the to total campaign receipts by for thecampaign incumbent andincreases). his opponent, as in well as umn 3 I include the total campaign receipts by the incumbent and his opponent, as well as the total campaign receipts from interest-group PACs, as additional controls. The estimated umn 3 I include thereceipts total campaign receipts by PACs, the incumbent and his opponent, as well as the total campaign from interest-group as additional controls. The estimated the total campaign receipts from interest-group PACs, as additional controls. The estimated slightly in absolute value (0 .150), remaining signicant at more value of inhI decreases the total campaign receipts from interest-group PACs, as additional controls. The estimated I decreases slightly in absolute value ( 0 . 150), remaining signicant at more value of inh slightly in absolute value (0 .150), remaining signicant at more value of in hI decreases than 99% condence level. Also, once other campaign contributions patterns are taken into slightly in absolute value (0 .150), remaining signicant at more value of in hI decreases than 99% condence level. Also, once other campaign contributions patterns are taken into than 99% condence level. Also, once other campaign contributions patterns are taken into to be signicantly positive. This lines up with my hypothesis, account, condence inhO is estimated than 99% level. Also, once other campaign contributions patterns are taken into account, inhO is estimated to be signicantly positive. This lines up with my hypothesis, O is estimated to be signicantly positive. This lines up with my hypothesis, account, in h though the coecient is still quite small, 0.033.This lines up with my hypothesis, to be signicantly positive. account, inestimated hO is estimated though the estimated coecient is still quite small, 0.033. though the estimated coe cient is still quite small, 0.033. characteristics. These include Column 4 controls for a number of time-varying though the estimated cient is still small, 0.county 033. characteristics. These include Column 4 controlscoe for a number of quite time-varying county Column 4 controls for a number of time-varying county characteristics. These include realColumn per capita incomefor (in 2000 US$ population, density, percent of 4 controls number ofdollars), time-varying county population characteristics. These include real per capita income (ina 2000 US$ dollars), population, population density, percent of real per capita income (in 2000 US$ dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, real per capita under income (in 2000 US$ dollars), population, density, percent of the population 18, percent of the population over 65, population percent female, percent black, the population under 18, percent of the population over 65, percent female, percent black, percent urban, under and percent with 12 or more years of education (among people with black, 25 or the population 18, percent of the population over 65, percent female, percent percent urban, and percent with 12 or more years of education (among people with 25 or percent urban, and percent with 12 or more years of education (among people with 25 or more years of age). column includes state income growth as a control, percent urban, and This percent withalso 12 or more years of education (among peoplesince with voters 25 or more years of age). This column also includes state income growth as a control, since voters more years of age). This column also includes state growth asSnyder a control, since Again, voters tend to punish incumbent senators when growth is income poor Levitt and (1997). more years of age). This column also includes state growth asSnyder a control, since Again, voters tend to punish incumbent senators when growth is income poor Levitt and (1997). tend to punish incumbent senators when growth is poor Levitt and Snyder (1997). Again, is remarkably (0.154 with a standard of the estimated cient for inhI when tend to punishcoe incumbent senators growth isstable poor Levitt and Snyder (1997). error Again, stable (0.154 with a standard error of the estimated coe cient for inhI is remarkably remarkably stable ( 0erences .154 with a standard error of the estimated coe cient for in hI is are 0.02), assuaging concerns that results driven by other di between the two types I is remarkably stable ( 0 . 154 with a standard error of the estimated coe cient for h 0.02), assuaging concerns that in results are driven by other dierences between the two types 0 . 02), assuaging concerns that results are driven by other di erences between the two types of counties. 0 .02), assuaging concerns that results are driven by other dierences between the two types of counties. of counties. While column 4 includes county characteristics, interacting each of these controls by the of counties. While column 4 includes county characteristics, interacting each of these controls by the While column 4 includes county interacting these controls by the party of the incumbent allows us to characteristics, account for potentially dieach erentof e ects of demographic While column 4 includes county interacting these controls by the party of the incumbent allows us to characteristics, account for potentially dieach erentof e ects of demographic party of the incumbent allows us to account for potentially dierent einteract ects of the demographic and social characteristics on either party. Also, it may be important to campaign party of the incumbent allows us to account for dierent ects of the demographic and social characteristics on either party. Also, it potentially may be important toeinteract campaign and social characteristics on either party. Also, it may be important to interact the campaign nance controls with the in-state dummy to make sure that the signicant interaction of the and social characteristics on either party. Also, it may important to interact the campaign nance controls with the in-state dummy to make surebe that the signicant interaction of the nance controls with the in-state dummy di tomake sure the signicant interaction of the Herndahl is not capturing a potentially erential ethat ect of such variables. nance controls with the in-state dummy di tomake sure the signicant interaction of the Herndahl is not capturing a potentially erential ethat ect of such variables. Herndahl is not capturing a potentially di erential e ect of such variables. This last exercise is especially useful to rule out a number of plausible alternative mechHerndahl is exercise not capturing a potentially di erential ect of such variables. This last is especially useful to rule out aenumber of plausible alternative mechThis last exercise is especially useful to rule out a number of plausible alternative mechanisms driving the results. For instance, candidates with concentrated patterns of campaign Thisdriving last exercise is especially useful to rule out a number of plausible alternative mechanisms the results. For instance, candidates with concentrated patterns of campaign anisms driving the results. For instance, candidates with concentrated patterns of campaign contributions could have di erent abilities to raise campaign funds (overall, or from conanisms drivingcould the results. For instance, candidates concentrated patternsor of from campaign contributions have di erent abilities to raise with campaign funds (overall, concontributions could have di erent abilities to raise campaign funds (overall, or from connected PACs). Sincehave campaign funds are used for advertising, this could aect candidates contributions dierent abilities to raise campaign funds (overall, orcandidates from connected PACs).could Since campaign funds are used for advertising, this could aect nected PACs). Since campaign funds are used for advertising, this could a ect candidates supportPACs). in-state relative to out-of-state. On the one hand, more advertising funds could nected Since campaign funds are used for one advertising, this advertising could aect funds candidates support in-state relative to out-of-state. On the hand, more could support in-state relative to out-of-state. On the one hand, more advertising funds could imply more exposure in in-state TV markets, and hence a greater dierence in candidate support in-state relative to out-of-state. On the hand, more advertising funds could imply more exposure in in-state TV markets, andone hence a greater dierence in candidate imply more exposure in in-state TV markets, and hence a greater di erence in candidate support across the two types of counties. Or, alternatively, fundsdi could allow imply more exposure in-state TV markets, and hence more a greater erence incandidates candidate support across the two in types of counties. Or, alternatively, more funds could allow candidates support across the two types of counties. Or, alternatively, more funds could allow candidates support across the two types of counties. Or, alternatively, more funds could allow candidates to reach to out-of-state voters, reducing such19 dierence. 19 19 Whatever the correlation, failure to account 1921 for it may bias the results. For similar reasons, changes in candidate quality (be it objective performance characteristics of the

of Gross State Product at the sectoral level) as well as its interaction with the in-state dummy. These variables are signicant, but the main coecient of interest (inhI ) is still

to reach to out-of-state voters, reducing such dierence. to reach to out-of-state voters, reducing such dierence. to reach to out-of-state voters, failure reducing dierence. Whatever the correlation, to such account for it may bias the results. For similar to reach to out-of-state voters, failure reducing dierence. Whatever the correlation, to such account for it may bias the results. For similar Whatever thein correlation, failure to (be account for it may bias the characteristics results. For similar reasons, changes candidate quality it objective performance of the Whatever the correlation, failure to account for it may bias the characteristics results. For similar reasons, changes in candidate quality (be it objective performance of the reasons, changes in candidate quality (be it objective performance characteristics of the candidate or other traits such as charisma) may a ect the candidates electoral perforreasons, changes in traits candidate (be it objective performance characteristics of the candidate or other such quality as charisma) may aect the candidates electoral perforcandidate or other traits such as charisma) may a ect the candidates electoral performance across dierent counties. such a variable is hard to measure,electoral total campaign candidate or other traits such asWhile charisma) may aect the candidates performance across dierent counties. While such a variable is hard to measure, total campaign mance across di erent counties. While such a variable is hard to measure, total campaign receipts may be thought of as a crude proxy. mance di erent counties. While such a variable is hard to measure, total campaign receiptsacross may be thought of as a crude proxy. receipts may be thought of as a crude proxy. For these reasons, column additional controls, the interaction of each county receipts may be thought of as 5 a adds, crude as proxy. For these reasons, column 5 adds, as additional controls, the interaction of each county For these reasons, column 5 adds, as additional controls, the interaction of each control with a dummy for the party of the incumbent as well as the interaction of eachcounty of the For with theseareasons, 5 adds, as additional controls, the interaction of each control dummy column for the party of the incumbent as well as the interaction of eachcounty of the control with a dummy for the party ofin-state the incumbent as While well asthe thepoint interaction of each of the campaign nance variables with the dummy. estimate for inthe hI control with a dummy for the party of the incumbent as well as the interaction of each of campaign nance variables with the in-state dummy. While the point estimate for inhI campaign nance variables with the in-state dummy. While the point estimate for hI decreases in magnitude (from 0the .154 to 0.129), it is still highly signicant (thefor standard campaign nance variables with dummy. While the point estimate in inhI decreases in magnitude (from 0.154in-state to 0.129), it is still highly signicant (the standard decreases in magnitude (from 0.154 to 0.129), it is still highly signicant (the standard error is 0.021). decreases in magnitude (from 0.154 to 0.129), it is still highly signicant (the standard error is 0.021). error is 0 . 021). In columns 6-10, I run similar specications as in columns 1-5, but with the Herndahl error 0.021). 6-10, Inis columns I run similar specications as in columns 1-5, but with the Herndahl In calculated columns 6-10, I run similar specications as in 1-5 columns 1-5,advantage but with the Herndahl index at the industry level. While columns have the of not dependIn columns 6-10, I run similar specications as in 1-5 columns 1-5,advantage but with the Herndahl index calculated at the industry level. While columns have the of not dependindex at classication, the industry level. While columns 1-5 have the advantage of not depending oncalculated any sort of this exercise is also informative. In principle, a candidate index at classication, the industry level. While columns 1-5 have the advantage of not depending oncalculated any sort of this exercise is also informative. In principle, a candidate ing on anya sort of classication, this of exercise is also informative. In principle, a candidate may have highly dispersed pattern contributions across PACs within a single industry, ing on anya sort of classication, this of exercise is also informative. principle, a candidate may have highly dispersed pattern contributions across PACs In within a single industry, may have a highly dispersed pattern of contributions across PACs within a single industry, but in this case it would be desirable to interpret his pattern of contributions as suggestive mayin have highly dispersed pattern of across PACs within a single industry, but thisacase it would be desirable to contributions interpret his pattern of contributions as suggestive but in this case it would be desirable to interpret his pattern of contributions as suggestive of a close link with the respective industry. As it turns out, results in columns 6-10 are very but this case it would be desirable to interpret his pattern of contributions as suggestive of a in close link with the respective industry. As it turns out, results in columns 6-10 are very of a close with the respective industry. As it turns results in columns 6-10negative, are very similar to link those in columns 1-5. Most importantly, the out, coe cient on inhI is still of a close with the respective industry. As it turns results in columns 6-10 are very similar to link those in columns 1-5. Most importantly, the out, coe cient on inhI is still negative, negative, similar to those in columns Most importantly, the coecient on estimated inhI is still highly and the1-5. magnitude of the eect implied by the coe cient is negative, similar signicative, to those in columns Most importantly, the coecient on estimated inhI is still highly signicative, and the1-5. magnitude of the eect implied by the coe cient is highly signicative, and the magnitude of the eect implied by the estimated coecient is similar. highly signicative, and the magnitude of the eect implied by the estimated coecient is similar. similar. In sum, the results in Table 2 are largely in line with the hypothesis that voters with similar. In sum, the results in Table 2 are largely in line with the hypothesis that voters with In sum, the results in Table 2 are largely in line with the hypothesis that voters with better access to results information punish a more concentrated pattern of campaign contributions In sum, the in Table 2 are largely in line with the hypothesis that voters with better access to information punish a more concentrated pattern of campaign contributions better access to information punish a more concentrated pattern of campaign contributions is punished. as discussed Section 2.1, a morepattern convincing strategy to control for better accessHowever, to information punish in a more concentrated of campaign contributions is punished. However, as discussed in Section 2.1, a more convincing strategy to control for is punished. However, as discussed in Section 2.1, aacross more convincing strategy to control for potential di erences in unobservable characteristics counties and capture the underlyis punished. However, discussed in Section 2.1, aacross more convincing strategy control for potential dierences in as unobservable characteristics counties and captureto the underlypotential dierences in unobservable characteristics across counties and capture the underlying partisanship (normal vote) in each county, includes a full set of county-party xed eects potential dierences in unobservable characteristics across counties and capture the underlying partisanship (normal vote) in each county, includes a full set of county-party xed eects ingin partisanship (normal vote) in eachin county, includes a has full the set of county-party xed eects as (2). The results are presented Table 3, which same structure as Table 2. ing partisanship (normal vote) in eachin county, full the set of county-party eects as in (2). The results are presented Table includes 3, whicha has same structure xed as Table 2. as in (2). The results are presented in Table 3, which has the same structure as Table 2. negative and signicant at more than 99% Again, the The main interaction coecient inhI is as in (2). results are presented in Table 3, which has the same structure as Table 2. Again, the main interaction coecient inhI is negative and signicant at more than 99% and signicant at more 99% Again, the level. main interaction coe cient indenition hI is negative condence result is robust to the of the Herndahl, and to thethan inclusion and signicant at more 99% Again, the level. main The interaction coe cient indenition hI is negative condence The result is robust to the of the Herndahl, and to thethan inclusion condence level. The result is robust to the denition of the Herndahl, and to the inclusion of controls in columns 2-5 and 6-10. The coe cient is again remarkably stable across specicondence The result is robust to the denition the Herndahl, to the inclusion of controls level. in columns 2-5 and 6-10. The coe cient isof again remarkably and stable across speciof controls in columns 2-5 and 6-10. The coe cient is again remarkably stable across specications, though the estimated e ect is smaller than with state times incumbent-party xed of controls in columns 2-5 and 6-10. coecient again remarkably stable across specications, though the estimated eectThe is smaller than is with state times incumbent-party xed cations, though the estimated e ect is smaller than with state times incumbent-party xed e ects alone (the point estimate ranges from about 0 . 05 to 0 . 10 in Table 2, compared to cations, though estimated ect is smaller than with times xed eects alone (the the point estimatee ranges from about 0.05 state to 0 .10 inincumbent-party Table 2, compared to e ects alone estimate ranges aboutvalue 0.05 to 0.10 in Table 2, compared to I indicates that an increase in about 0.14 (the to point 0.17 in Table 3). This from estimated in h e ects alone (the point estimate ranges from about 0 . 05 to 0 . 10 in Table 2, compared in about 0.14 to 0.17 in Table 3). This estimated value inhI indicates that an increase to that anthe increase in about 0.14 to 0.17 in Table 3). This estimated value inout-of-state, hI indicates though concentration leads to lower vote shares in-state relative to e ect is that anthe increase about 0.14 to 0.17 Table 3). This estimated value inout-of-state, hI indicates though concentration leads to in lower vote shares in-state relative to eect in is concentration lower a vote shares in-state relative out-of-state, though the e ect is quantitatively leads smallto (about half of a percentage point to for a one-standard increase in the concentration leads to lower vote shares in-state relative to out-of-state, though the e ect is quantitatively small (about a half of a percentage point for a one-standard increase in the quantitatively small (about a half of a percentage point for a one-standard increase in the quantitatively small (about a half of a percentage point for a one-standard increase in the concentration indices). The coecient on the20 incumbents Herndahl in the most demanding 20 20 both relative to columns without controls specications with full controls is also reduced, 20 and to Table 2.
22

3.2

Examining neighboring counties

concentration indices). The coecient on the incumbents Herndahl in the most demanding concentration indices). The coecient on the incumbents Herndahl in the most demanding concentration indices). coe cient the incumbents Herndahl in the without most demanding specications with full The controls is alsoon reduced, both relative to columns controls concentration indices). coe cient the incumbents Herndahl in the without most demanding specications with full The controls is alsoon reduced, both relative to columns controls specications with full controls is also reduced, both relative to columns without controls and to Table 2. specications with full controls is also reduced, both relative to columns without controls and to Table 2. and to Table 2. and to Table 2.

within the same counties county to estimate theincreases coecient Those specications ask whether in-state tend to punish in of theinterest. Herndahl index of the candidate within the same counties county to estimate theincreases coecient Those specications ask whether in-state tend to punish in of theinterest. Herndahl index of the candidate whether in-state counties tend to This punish increases in the Herndahl which index of the candidate more than out-of-state counties. is a demanding specication controls for any whether in-state counties tend to This punish in the Herndahl which index of the candidate more than out-of-state counties. is increases a demanding specication controls for any more than out-of-state counties. This is a demanding specication which controls for any time-invariant characteristics of counties, but the response of an increase in the Herndahl more than out-of-state counties. This is a demanding specication whichin controls for any time-invariant characteristics of counties, but the response of an increase the Herndahl time-invariant characteristics of counties, butreect the response of an increase in counties the Herndahl index in-state relative to out-of-state may still di erential trends across rather time-invariant characteristics of counties, butreect the response of an increase in counties the Herndahl index in-state relative to out-of-state may still dierential trends across rather index in-state relative to out-of-state may still reect dierential trends across counties rather than the e ect of media. index in-state relative to out-of-state may still reect dierential trends across counties rather than the eect of media. than the subsection eect of media. This investigates the possibility. First, I focus on in-state counties that are than the subsection eect of media. This investigates the possibility. First, I focus on in-state counties that are This subsection investigates the possibility. First, I focus on in-state counties are geographically contiguous to out-of-state counties. Table 4 demonstrates that these that counties This subsection investigates the possibility. First, I focus on in-state counties are geographically contiguous to out-of-state counties. Table 4 demonstrates that these that counties geographically contiguous counties to out-of-state counties. Table 4 demonstrates that these counties are similar to out-of-state along a number of dimensions. This table runs regressions geographically contiguous counties to out-of-state Table 4 demonstrates that these counties are similar to out-of-state along acounties. number of dimensions. This table runs regressions are similar to out-of-state counties along a number of dimensions. This table runs regressions of the following form: are similar to out-of-state counties along a number of dimensions. This table runs regressions of the following form: of the following form: ycst = s + t + in inc + cst , of the following form: y = s + t + in inc + cst , = s + t + in inc + cst , ycst s + e + in in , ycst cst = t ects, c+ xed and ycstvaries per column and is one of where s and t are state and year where s and t are state and year xed eects, and ycst cst varies per column and is one of and are state and year xed e ects, and y varies per column and is one of where s t several county namely: real per capita income (in 2, 000 US $ state characteristics, and year xed e ects, and ycst varies per column and is one of where observable s and t are cst several observable county characteristics, namely: real per capita income (in 2, 000 US $ several observable county characteristics, namely: realpopulation per capitaunder income (in 2, 000of Uthe S$ dollars), population, population density, percent of the 18, percent several county characteristics, namely: realpopulation per capitaunder income 2, 000of Uthe S$ dollars),observable population, population density, percent of the 18,(in percent dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, percent urban, and percent with 12 or dollars), population, population density, percent of the population 18, percent population over 65, percent female, percent black, percent urban, under and percent with of 12the or population over 65, percent female, percent black, percent urban, and percent with 12 or more years of education (among people with 25 or more years of age). population over 65, percent female, percent black, percent urban, and percent with 12 or more years of education (among people with 25 or more years of age). more years of education (among people with 25 or more years of age). is simply the in-state dummy dened before (equals 1 if the state the In Panel A, in more education (among people with 25 or more years of age). In years Panelof A, inc c is simply the in-state dummy dened before (equals 1 if the state the In Panel A, in simply the in-state dummy dened before (equals 1 if the the c is than county is in has more two-thirds of the population of that media market, and state zero if it simply the in-state dummy dened before (equals 1 if the the In Panel A, in c is than county is in has more two-thirds of the population of that media market, and state zero if it county is in has more than two-thirds of the population of that media market, and zero if it measures the average di erence in observable characterhas less than one-third). Thus, county in has more than two-thirds of the population of that media market, and zero if it has lessis than one-third). Thus, in in measures the average dierence in observable charactermeasures the controlling average difor erence instate observable characterhas less one-third). Thus, incounties istics forthan in-state and out-of-state after xed and overall time measures the controlling average difor erence instate observable characterhas less one-third). Thus, incounties istics forthan in-state and out-of-state after xed and overall time istics for in-state and out-of-state counties after controlling for xed state and overall time trends. In-state counties have: about $1 , 700 higher per capita income than out of state istics forIn-state in-state counties and out-of-state counties after controlling for xed statethan and out overall time trends. have: about $1, 700 higher per capita income of state trends. In-state counties have: about $1 , 700 higher per capita income than out of state counties (column counties 1); about twice about as many and about 98 more individuals perof square trends. In-state have: $1,people 700 higher per capita income than out state counties (column 1); about twice as many people and about 98 more individuals per square counties (column 1); about twice as many people and about 98 more individuals per square mile (columns 2 and 3); a higher of young (under98 18), black, and urban counties (column 1); about twice proportion as many people and about more individuals perpopulasquare mile (columns 2 and 3); a higher proportion of young (under 18), black, and urban populamile (columns 2 and 3); a higher proportion of young (under 18), black, and urban population, (columns and a smaller proportion of proportion people over 65 (columns 4-8); and a higher proportion of mile 2 and 3); a higher of young (under 18), black, and urban population, and a smaller proportion of people over 65 (columns 4-8); and a higher proportion of tion, and a smaller proportion of people over 65 (columns 4-8); and a higher proportion of people with 12 or more years of schooling (column 9). tion, and a smaller proportion people over 65 (columns 4-8); and a higher proportion of people with 12 or more years ofof schooling (column 9). people with 12 or schooling (column 9). equals 1 if it satises the same conditions as in Panel A and In contrast, in more Panelyears B incof people with 12 or schooling (column 9). equals 1 if it satises the same conditions as in Panel A and In contrast, in more Panelyears B incof In contrast, in Panel B in equals 1 if it satises the same conditions as in Panel A and c out-of-state county in the In in-state county borders an the same state. These contiguous incontrast, in Panel B in 1 if it county satises same conditions as in Panel A and c equals the in-state county borders an out-of-state inthe the same state. These contiguous inthe in-state county borders an out-of-state county in the same state. These contiguous inthe in-state county borders an out-of-state county in the same state. These contiguous in21 set of out-of-state counties. In all cases, the state counties are much more comparable to the 21 21 estimated coecients that were previously signicant fall in absolute terms. Moreover, the 21 two sets of counties are not found to be signicantly dierent in terms of population density, or any of the basic dimensions of population structure (age, gender, race, or urban-rural). However, three estimated coecients are still signicant. In-state and contiguous counties are signicantly richer, more populous, and slightly better educated than neighboring instate counties. Yet even in these cases, it is apparent that the dierences are smaller than 23 in Panel A.

3.2 Examining neighboring counties 3.2 Examining neighboring counties 3.2 Examining neighboring counties 3.2 Examining neighboring The regressions from Table 3 rely solelycounties on variation in the Herndahl indices across time The regressions from Table 3 rely solely on variation in the Herndahl indices across time The regressions Table rely solely variation in interest. the Herndahl across time within the samefrom county to 3 estimate theon coe cient of Thoseindices specications ask The regressions Table rely solely variation in interest. the Herndahl across time within the samefrom county to 3 estimate theon coe cient of Thoseindices specications ask

state counties are much more comparable tosignicant the set of out-of-state counties. all cases, the estimated coe cients that were previously fall in absolute terms.In Moreover, the estimated coecients that were previously signicant fall in absolute terms. Moreover, the estimated cients that previously signicant fall in absolute terms. Moreover, the two sets ofcoe counties are not were found to be signicantly di erent in terms of population density, two sets of counties are not found to be signicantly dierent in terms of population density, two sets ofthe counties not found be signicantly dierent terms of population density, or any of basic are dimensions of to population structure (age, in gender, race, or urban-rural). or any of the basic dimensions of population structure (age, gender, race, or urban-rural). or any of three the basic dimensions of population (age, gender, race, or urban-rural). However, estimated coecients are still structure signicant. In-state and contiguous counties However, three estimated coecients are still signicant. In-state and contiguous counties However, three estimated coe cients are still signicant. In-state and contiguous counties are signicantly richer, more populous, and slightly better educated than neighboring inare signicantly richer, more populous, and slightly better educated than neighboring inare signicantly richer, populous, and slightly better educated thanare neighboring instate counties. Yet evenmore in these cases, it is apparent that the dierences smaller than state counties. Yet even in these cases, it is apparent that the dierences are smaller than state counties. Yet even in these cases, it is apparent that the dierences are smaller than in Panel A. in Panel A. in Panel TableA. 5 focuses on this set of more comparable counties, running specications as in Table 5 focuses on this set of more comparable counties, running specications as in Table 5 focuses this set of comparable running specications as in Table 3. The resultson are similar to more the ones obtainedcounties, for the full sample of counties. With Table 3. The results are similar to the ones obtained for the full sample of counties. With Table 3. The results are similar to the ones obtained for and the full sample of counties. With either Herndahl measure (at the PAC or industry level) with or without controls, the either Herndahl measure (at the PAC or industry level) and with or without controls, the either Herndahl measure (at PAC or industry level) and with or without controls, the estimated absolute value of the hI decreases slightly. Without controls, it is still close to estimated absolute value of in inhI decreases slightly. Without controls, it is still close to I decreases slightly. Without controls, it is still close to estimated absolute valueerror of in h 0.08 with a standard of about 0.03. The inclusion of the set of controls further 0.08 with a standard error of about 0.03. The inclusion of the set of controls further 0.08 with a point standard error to of around about 0 .03. The statistical inclusion of the set ofalso controls further decreases the estimate 0.5. The signicance falls, though decreases the point estimate to around 0.5. The statistical signicance also falls, though decreases the point estimate to around 0.5 . The statistical signicance also controls, falls, though even in columns 4 and 9, with a full set of county and campaign nance the even in columns 4 and 9, with a full set of county and campaign nance controls, the even in columns 4 and 9, withterm a full and nance controls, the hypothesis that the interaction is set zeroofiscounty rejected at campaign conventional signicance levels. hypothesis that the interaction term is zero is rejected at conventional signicance levels. hypothesis that the 10, interaction term is zero is rejected at conventional Only columns 5 and where (i) county controls are interacted with the signicance incumbentslevels. party Only columns 5 and 10, where (i) county controls are interacted with the incumbents party Only columns 5 and 10, where (i) county controls are interacted the incumbents dummy and (ii) campaign nance controls are interacted with with the in-state dummy, party have dummy and (ii) campaign nance controls are interacted with the in-state dummy, have dummy and (ii) campaign nance controls are interacted with the in-state dummy, marginally signicant coecients on the interaction term. Note, however, that even inhave this marginally signicant coecients on the interaction term. Note, however, that even in this marginally signicant coeis cients on interaction term. Note, even in this case, the point estimate similar inthe magnitude ( 0 .04) to the however, previous that estimates. Also, case, the point estimate is similar in magnitude (0.04) to the previous estimates. Also, case, the point estimate errors is similar in surprising magnitudegiven (0.04) to the previous estimates. Also, the increase in standard is not the simultaneous inclusion of various the increase in standard errors is not surprising given the simultaneous inclusion of various the increase in standard errors is not surprising given the simultaneous inclusion of various campaign nance variables in the regression. campaign nance variables in the regression. campaign nance variables in the regression.

Table 5 shows that the negative impact of increased campaign concentration in counties with Table 5 shows that the negative impact of increased campaign concentration in counties with Table showsto that the negative impact increased campaign concentration with better 5 access information relative to of those with less information holds forin a counties restricted set better access to information relative to those with less information holds for a restricted set better access to information to those with less information holds for a restricted set of more comparable in-staterelative and out-of-state counties. Such evidence suggests that the of more comparable in-state and out-of-state counties. Such evidence suggests that the of more comparable and counties. Such evidence suggests that previous results were in-state not driven byout-of-state dierential trends across counties rather than the ethe ect previous results were not driven by dierential trends across counties rather than the eect previous were driven by erential trends across counties rather than the eect of media.results However, itnot is possible to di further investigate this possibility by running a simple of media. However, it is possible to further investigate this possibility by running a simple of media. However, it is possible to further investigate this possibility by running a simple placebo test. placebo test. placebo test. 6, I dene the contiguous, in-state counties to be the (placebo) out-of-state In Table In Table 6, I dene the contiguous, in-state counties to be the (placebo) out-of-state In Table 6, I dene the in-state counties to same be the (placebo) out-of-state counties, comparing them to contiguous, adjacent in-state counties in the state. Since these sets of counties, comparing them to adjacent in-state counties in the same state. Since these sets of counties, comparing to adjacent in-state counties in the samebe state. Since these sets of counties are in-state them with good access to information, there should no correlation between counties are in-state with good access to information, there should be no correlation between counties are in-state with good access to information, there should be no correlation between increases in concentration and moving across county categories (i.e., the estimated inhI 22 and inhO should be close to zero). However, if the previous results were driven by a general 22 22 geographic trend as one moves from in-state towards out-of-state captures, these coecients could be signicant, raising concerns about the validity of the preceding results. Results from Table 6 reassure previous results. Unlike Tables 2, 3 and 5 the interaction coecient inhI (and inhO ) is not statistically signicant in any of the specications (the structure of the table is the same as before). Moreover, the point estimate is very small in all columns suggesting this is a precisely measured non-eect rather than an uncertain potential eect.
24

3.3 3.3 3.3

A placebo test A A placebo placebo test test

3.4

Share of connected-PACs money

and inhO should be close to zero). if the previous results were driven by a general geographic trend as one moves from However, in-state towards out-of-state captures, these coe cients geographic trend as one moves from in-state towards out-of-state captures, these coecients geographic trend as one moves from in-state towards out-of-state captures, these coe cients could be signicant, raising concerns about the validity of the preceding results. Results could be signicant, raising concerns about the validity of the preceding results. Results could be signicant, concerns about the validity ofand the 5preceding results. Results from Table 6 reassureraising previous results. Unlike Tables 2, 3 the interaction coe cient from Table 6 reassure previous results. Unlike Tables 2, 3 and 5 the interaction coecient from Table 6 reassure previous results. Unlike Tables 2, 3 and 5 the interaction coe cient inhI (and inhO ) is not statistically signicant in any of the specications (the structure inhI (and inhO ) is not statistically signicant in any of the specications (the structure (and is in ) is not statistically signicant in any of the specications inthe hI table hOsame of the as before). Moreover, the point estimate is very small (the in allstructure columns of the table is the same as before). Moreover, the point estimate is very small in all columns of the table is the same as before). Moreover, the point estimate is very small in all columns suggesting this is a precisely measured non-eect rather than an uncertain potential eect. suggesting this is a precisely measured non-eect rather than an uncertain potential eect. suggesting this is a precisely measured non-eect rather than an uncertain potential eect.

The evidence presented thus far suggests that voters dislike concentrated special interest The evidence presented thus far suggests that voters dislike concentrated special interest The evidence suggests that voters dislike concentrated special interest money. But is presented it the casethus that far voters dislike special interest money in general, concentrated money. But is it the case that voters dislike special interest money in general, concentrated money. But is it the case that voters interest similar money in general, concentrated or otherwise? Table 7 suggests not.dislike This special table presents regressions to the ones or otherwise? Table 7 suggests not. This table presents similar regressions to the ones or otherwise? Table 7 suggests This table presents the ones presented before, where the sharenot. of connected PAC moneysimilar is used regressions (in place ofto Herndahl presented before, where the share of connected PAC money is used (in place of Herndahl presented before, where the of connected PAC is used (in place of Herndahl indices of concentration), as share a proxy for capture of money politicians by interest groups. indices of concentration), as a proxy for capture of politicians by interest groups. indices of concentration), as a proxy for of politicians by interest In the table, we see that the share ofcapture connected PAC money of the groups. incumbent and In the table, we see that the share of connected PAC money of the incumbent and In the table, we see that the share of connected PAC money of the incumbent and the opponent are not robustly correlated with dierent vote share patterns across media the opponent are not robustly correlated with dierent vote share patterns across media the opponent are notof robustly correlated with dierent vote share patterns across media markets. In the case the incumbent, the estimated interaction term is close to zero and markets. In the case of the incumbent, the estimated interaction term is close to zero and markets. In the case of the incumbent, the estimated interaction term is close to zero and not statistically signicant with either the full sample of counties (and including state times not statistically signicant with either the full sample of counties (and including state times not statistically signicant with either full sample of counties (and including times party xed eects as in columns 1 and the 2 or county times party xed e ects as instate columns 3 party xed eects as in columns 1 and 2 or county times party xed eects as in columns 3 party xed e ects as in columns 1 and 2 or county times party xed e ects as in columns 3 and 4), or with the set of contiguous counties only (and county times party xed eects as and 4), or with the set of contiguous counties only (and county times party xed eects as and 4), or with the set of contiguous counties only (and county times party xed eects as in columns 5 and 6). in columns 5 and 6). in columns 5 and are 6). consistent with idea that voters view PAC money as buying inuence. These results These results are consistent with idea that voters view PAC money as buying inuence. These results are with money as buying inuence. More specically, it isconsistent in line with the idea viewthat that voters a high view sharePAC of PAC-money is not necessarily More specically, it is in line with the view that a high share of PAC-money is not necessarily More specically, it is in line with the view that a high shareis ofdispersed PAC-money is not sectors, necessarily bad for the incumbents vote share because, if this money in many his bad for the incumbents vote share because, if this money is dispersed in many sectors, his bad for cannot the incumbents vote because, if this money dispersed many sectors, his agenda be captured byshare any one of them. Results in is the previousin subsections suggest agenda cannot be captured by any one of them. Results in the previous subsections suggest agenda cannot be captured by any one them. Results the indicate previouscapture. subsections suggest that it is PAC money concentrated in aof few sectors whatin may that it is PAC money concentrated in a few sectors what may indicate capture. that it is PAC money concentrated in a few sectors what may indicate capture.

3.4 3.4 3.4

Share of connected-PACs money Share of connected-PACs Share of connected-PACs money money

A valid concern with results presented thus far is whether the suggested mechanism demands A valid concern with results presented thus far is whether the suggested mechanism demands A concern with results presented is whether the suggested mechanism demands anvalid unrealistic degree of knowledge onthus the far part of voters. However, even if individuals do an unrealistic degree of knowledge on the part of voters. However, even if individuals do an unrealistic degree of knowledge on the part of voters. However, even if individuals do not know the approximate pattern of campaign contributions, a candidate with a more not know the approximate pattern of campaign contributions, a candidate with a more not know thepattern approximate pattern of is campaign a candidate with a or more concentrated of contributions prone tocontributions, being pinpointed by journalists his concentrated pattern of contributions is prone to being pinpointed by journalists or his concentrated pattern of contributions is prone to being pinpointed by journalists or his opponent as potentially captured. The presumption is that such a candidate is likely 23 opponent as potentially captured. The presumption is that such a candidate is likely 23 23 to gure in ads and news as linked to the economic sectors that heavily contribute to his to gure in ads and news as linked to the economic sectors that heavily contribute to his campaign. campaign. An additional important issue is whether campaign nance concentration captures some An additional important issue is whether campaign nance concentration captures some other omitted candidate characteristics which dierentially aect their performance across other omitted candidate characteristics which dierentially aect their performance across counties (e.g. whether candidates with higher Herndahls are more popular or chariscounties (e.g. whether candidates with higher Herndahls are more popular or charismatic). As noted above, other campaign nance controls such as total campaign receipts matic). As noted above, other campaign nance controls such as total campaign receipts may be thought of as a crude proxy for this, and I have demonstrated the robustness of the may be thought of as a crude proxy for this, and I have demonstrated the robustness of the results to the inclusion of such controls. However, in this section, I present an additional results to the inclusion of such controls. However, in this section, I present an additional 25 simple exercise to help address these two issues. simple exercise to help address these two issues. In particular, I conduct automated news searches for a sample of US newspapers verifying

3.5 3.5 3.5

Automated news searches Automated Automated news news searches searches

campaign. An additional important issue is whether campaign nance concentration captures some campaign. An additional important issue is whether campaign nance concentration captures some Anomitted additional important issue is whether campaign nance concentration capturesacross some other candidate characteristics which dierentially aect their performance Anomitted additional important issue is whether campaign nance concentration capturesacross some other candidate characteristics which dierentially aect their performance other omitted candidate characteristics which diHerndahls erentially a ect their performance across counties (e.g. whether candidates with higher are more popular or charisother omitted characteristics erentially a ect their performance across counties (e.g. candidate whether candidates with which higherdiHerndahls are more popular or chariscounties (e.g. whether candidates with higher Herndahls are more popular or charismatic). noted above, other campaign nance controls such total campaign counties As (e.g. whether candidates with higher Herndahls are as more popular orreceipts charismatic). As noted above, other campaign nance controls such as total campaign receipts matic). As noted above, other campaign nance controls such as total campaign receipts may be thought of as a crude proxy for this, and I have demonstrated the robustness of the matic). As noted other campaign nance such as total receipts may be thought of above, as a crude proxy for this, and I controls have demonstrated thecampaign robustness of the may be to thought of as a crude proxy for this, and I have demonstrated the robustness of the results the inclusion of such controls. However, in this section, I present an additional may be thought of as a crude proxy for this, and I have demonstrated the robustness of the results to as thepotentially inclusion of such controls. However, in this section, I present an additional opponent captured. The presumption is that such a candidate is likely results to the inclusion of such controls. However, in this section, I present an additional simple exercise to help address these two issues. results to the inclusion of such controls. However, in this section, I present an additional simple exercise help address these to two issues. to gure in adsto and news as linked the economic sectors that heavily contribute to his simple exercise to address these two issues. In particular, I help conduct automated news searches for a sample of US newspapers verifying simple exercise to help address these two issues. In particular, I conduct automated news searches for a sample of US newspapers verifying campaign. In particular, I conduct automated news searches for To a sample of US newspapers verifying how often are Senate candidates featured in the news. this end, I search for back issues In particular, I conduct automated news searches for To a sample of US newspapers verifying how often are Senate candidates featured in the news. this end, I search for back issues An additional important issue is whether campaign nance concentration captures some how often are candidates featured in the news. To thisOverall, end, I search for back issues of a sample of Senate US newspapers available on NewsLibrary.com. the search is for 139 how often are Senate candidates featured in the news. To this end, I search for back issues of a sample of US newspapers available on NewsLibrary.com. Overall, the search is for 139 other omitted candidate characteristics which dierentially aOverall, ect theirthe performance across of a sample of US newspapers available on NewsLibrary.com. search is for 139 newspapers across the US. The major limitation of this exercise is that only a few newspapers of a sample across of US the newspapers available on NewsLibrary.com. Overall, theasearch is for 139 newspapers US. The major limitation of this exercise is that only few newspapers counties (e.g. whether candidates with higher of Herndahls are that moreonly popular or charisnewspapers across the US. The this exercise a few newspapers have online archives going backmajor to thelimitation 1980s (in of fact, only one, is The Boston goes back newspapers across the US. The major limitation this exercise is that only aGlobe few newspapers have online archives going back to the 1980s (in fact, only one, The Boston Globe goes back matic). As archives noted above, other campaign nance controls such as total campaign receipts have online going back to the 1980s (in fact, only one, The Boston Globe goes back to 1982). Appendix Table A.3 shows the number of states in the sample, per year, for which have online archives Table going A.3 back to the 1980s (in fact, only in one, The Boston Globe goes back to 1982). Appendix shows the number of states the sample, per year, for which may be thought of as a crude for this, and of I have demonstrated the robustness of the tocan 1982). Appendix Table A.3 proxy shows the number states in the sample, per year, for which I nd one or more newspapers on which to conduct news searches. tocan 1982). Appendix Table A.3 shows the number of states in the sample, per year, for which I nd or more newspapers on which to conduct news searches. results to one the inclusion of such controls. However, in this section, I present an additional I can nd one or more newspapers on which to conduct news searches. For each candidate, I conduct two searches. In the rst one, I look for stories about the I can nd one or more newspapers on which to conduct news searches. For exercise each candidate, I conduct twotwo searches. rst one, I look for stories about the simple to help address these issues. In the For each candidate, I conduct two searches. the rst one, lookmonths for stories about candidate during the campaign period (which I In dene to be theI two leading upthe to For each candidate, I conduct two searches. In the rst one, lookmonths for stories about candidate during Ithe campaign periodnews (which I dene to be theI two leading upthe to In particular, conduct automated searches for a sample of US newspapers verifying candidate during the campaign period (which I dene to be the two months leading up to the election). The search is limited to newspapers within the candidates state, and allow candidate during the campaign period (which I dene to be the two months leading up to the election). The search is limited to newspapers within the candidates state, and allow howelection). often are Senate candidates featured in the news. To this end, I search for back issues the The search is limited to newspapers within the rst candidates state, and allow for various possible variations in the usage of the candidates name, middle name (or the various election). The search is limited newspapers within the rst candidates state, and allow for variations in theto usage of the candidates name, name (or of avarious sample possible of US newspapers available on NewsLibrary.com. Overall, the middle search is for 139 for possible variations in the usage of the candidates rst name, middle name (or initial), last names, and nicknames. Any news story containing the candidates name and the for various possible variations in the usage ofstory the candidates rst name, middle name (or initial), last names, and nicknames. Any news containing the candidates name and the newspapers the US. The major limitation of this exercise the is that only a few newspapers initial), last across names, and nicknames. Any news story containing candidates name and the words senate or senator is counted as one hit that adds to the candidates total stories. initial), last names, and nicknames. Any story containing the candidates namestories. and the words senate or senator is counted asnews one(in hit that adds one, to the candidates total have online archives going back to the 1980s fact, only The Boston Globe goes back words senate or senator is counted as one hit that adds to the candidates total stories. The second search demands stories to meet the prior criteria and, in addition, total that the words words senate or senator is counted as one hit that adds to the candidates stories. The second search demands stories to the meet the prior criteria in addition, that the to 1982). Appendix Table A.3 shows number of states inand, the sample, per year, for words which 21 The second demands stories meet the prior criteria and, in addition, that the words This subset of stories are the candidates PAC PAC and search money feature in theto story 21 .the The second search demands stories to meet prior criteria and, in addition, that the words . This subset of stories are the candidates PAC PAC and money feature in the story 21 I can nd one or more newspapers on which to conduct news searches. This subset of stories are candidates PAC PAC and money feature into the story 21 .whether stories. The main interest is examine the proportion of the PAC stories for each . This subset of stories are the candidates PAC PAC and money feature in the story stories. The main interest is to examine whether the proportion of PAC stories for each For each candidate, I conduct two searches. In the rst one, I look for stories about the 22 stories. The main interest is to examine whether the proportion of PAC stories for each . candidate is positively correlated with the candidates concentration indices 22 stories. The main interest is to examine whether the proportion of PAC stories for each . candidate is positively correlated with the candidates concentration indices 22 leading up to candidate is during the campaign period (which I dene to be the two indices months candidate positively with the 22 . for each canTable 8 reports thecorrelated results. Column 1 candidates reports the concentration results of a regression . for each cancandidate is positively correlated with the candidates concentration indices Table 8 reports the results. Column 1 reports the results of a regression the Table election). The search is limited to newspapers within the of candidates state, and allow 8 reports the results. Column 1 reports the results a regression for each candidates Herndahl index (at the PAC level) on the ratio of PAC stories to total stories, TableHerndahl 8 reports the results. Column 1 reports theratio results of a regression for each candidates index (at the PAC level) on the of PAC stories to total stories, for various possible index variations in the usage of on thethe candidates rst name, name (or didates Herndahl (at the PAC level) ratio of PAC storiesmiddle to total stories, didates Herndahl index (at the PAC level) on the ratio of PAC stories to total stories, initial), last names, and nicknames. Any news story containing the candidates name and the 21 with other keywords toincumbent nd the number of stories concerning campaign money, including controlling for year xed e ects, an dummy, and a Democratic dummy. The ra21 I experimented other keywords to nd number of stories concerning campaign money, stories. including words senate orwith senator is counted asthe one hit that addsothers. to theHowever, candidates total 21 I experimented contributions, campaign nance, and interest groups, among an examination of the I experimented with other keywords to nd the number of stories concerning campaign money, including tio21 of PAC stories is positively and very signicantly correlated with the concentration index contributions, campaign nance, and interest among others. However, an examination of the I experimented with other keywords to nd the groups, number of stories concerning campaign money, including resulting stories suggested that the selected alternative both included the major stories about interest group contributions, campaign nance, and interest groups, among others. However, an examination of the The second search demands stories to alternative meet the prior criteriathe and, in addition, that the words resulting stories suggested that the selected both included major stories about interest group contributions, campaign nance, and interest groups, among others. However, an examination of the money and (unlike other0keywords) did not alternative include as many unrelated stories. (the point estimate, .993, has a robust standard error of 0 . 161). In column 2, instead, the resulting stories suggested that the selected both included the major stories about interest group 21 as many unrelated stories. money and (unlike other keywords) did not include 22 . This subset of stories are the candidates PAC PAC and money feature in the story resulting stories suggested that the selected alternative both included the major stories about interest group Table A.3 keywords) gives summary statistics for number of total stories, stories about PACs, and money and (unlike other did not include as the many unrelated stories. 22 Appendix total number of stories is used as the main independent variable of interest (again, xed Appendix Table A.3 keywords) gives summary statistics for number of total stories, stories about year PACs, and money and (unlike other did not include as the many unrelated stories. the22 ratio. Appendix Table A.3 gives summary statistics for the number of total stories, storiesstories about PACs, and stories. The main interest is to examine whether the proportion of PAC for each the22 ratio. Appendix Table A.3 gives summary statistics for the number of total stories, stories about PACs, and eects and partisan and incumbency dummies are included). Also interesting, this regression the ratio. candidate the ratio. is positively correlated with the candidates concentration indices22 . shows no signicant correlation between the24 total number of stories and the concentration 24 Table 8 reports the results. Column 1 reports the results of a regression for each can24 3, where index. The same conclusion emerges in column the proportion of PAC stories and 24on the ratio didates Herndahl index (at the PAC level) of PAC stories to total stories, the total number of stories are included simultaneously in the regression. A similar picture emerges in columns 4-6, which repeat the specications in columns 1-3 but use the concen21 I experimented with other keywords to nd the number of stories concerning campaign money, including tration index at the industry level as the dependent variable. only di erence is of that contributions, campaign nance, and interest groups, among others. The However, an examination the resulting that marginally the selected alternative both included correlation the major stories about total interest group there is stories now asuggested small (and signicant) negative between stories money and (unlike other keywords) did not include as many unrelated stories. and 22 the concentration index. Appendix Table A.3 gives summary statistics for the number of total stories, stories about PACs, and the ratio. These results are very reassuring about the validity of the approach. In particular, the positive and signicant correlation between the share of PAC stories for a candidate and the 24 candidates Herndahl concentration index suggests that the concentration indices do proxy 26 for media exposure of campaign contribution proles. And the weak correlation between the

total number of stories used as the dummies main independent variable ofinteresting, interest (again, year xed eects and partisan andis incumbency are included). Also this regression eects and partisan and incumbency dummies are included). Also interesting, this regression e ects no and partisan and incumbency dummies are included). Also interesting, regression shows signicant correlation between the total number of stories and the this concentration shows no signicant correlation between the total number of stories and the concentration shows signicant correlation between the total number ofproportion stories andof the concentration index. no The same conclusion emerges in column 3, where the PAC stories and index. The same conclusion emerges in column 3, where the proportion of PAC stories and index. The same conclusion emerges in column 3, wherein the proportion ofA PAC stories and the total number of stories are included simultaneously the regression. similar picture the total number of stories are included simultaneously in the regression. A similar picture the totalin number of 4-6, stories are repeat included simultaneously regression. Ause similar picture emerges columns which the specications in in the columns 1-3 but the concenemerges in columns 4-6, which repeat the specications in columns 1-3 but use the concenemergesindex in columns whichlevel repeat in columns but the concentration at the4-6, industry as the the specications dependent variable. The1-3 only diuse erence is that tration index at the industry level as the dependent variable. The only di erence is that controlling for year e ects, an as incumbent dummy, and a Democratic dummy. ratration thexed industry level the dependent variable. The only di erence is that there is index now aat small (and marginally signicant) negative correlation between totalThe stories there is now a small (and marginally signicant) negative correlation between total stories tio PAC stories is positively and very signicantly correlated with the concentration index there is now a small (and marginally signicant) negative correlation between total stories andof the concentration index. and the concentration index. (the point estimate, .993, has a robust standard error of .161). In column 2, instead, the the and the concentration index. These results are 0very reassuring about the validity of 0 the approach. In particular, These results are very reassuring about the validity of the approach. In particular, the total number of stories is used as the main independent of interest (again, year xed These results are very reassuring about the share validity of the approach. In particular, the positive and signicant correlation between the ofvariable PAC stories for a candidate and the positive and signicant correlation between the share of PAC stories for a candidate and the e ects and partisan andconcentration incumbency dummies are included). interesting, this regression positive and signicant correlation between the share of PAC stories for a candidate and the candidates Herndahl index suggests that the Also concentration indices do proxy candidates Herndahl concentration index suggests that the concentration indices do proxy shows no signicant between the total number ofconcentration stories and the concentration candidates Herndahl concentration index suggests that the indices do proxy for media exposure ofcorrelation campaign contribution proles. And the weak correlation between the for media exposure of campaign contribution proles. And the weak correlation between the index. The same conclusion emerges in column 3, where the proportion of PAC stories and for media exposure of campaign contribution proles. And the weak correlation between the candidates Herndahl and his total number of stories suggests the former is not capturing candidates Herndahl and his total number of stories suggests the former is not capturing the total number of stories are included simultaneously in the lines, regression. A picture candidates Herndahl and his total of stories the former issimilar notin capturing di erences in candidate charisma or number popularity. Alongsuggests these columns 7-9 Table 8 di erences in candidate charisma or popularity. Along these lines, columns 7-9 in Table 8 emerges columns which repeat the in specications inbut columns use concendierences in candidate charisma orthose popularity. Along these lines, columns 7-9the in variable Table 8 present ain nal set of 4-6, regressions like columns 1-3, where1-3 thebut dependent present a nal set of regressions like those in columns 1-3, but where the dependent variable tration at industry as the in dependent variable. The the only dierence is total that present a nalcontributions setthe of regressions like those columns 1-3, but where dependent variable is total index PAC to level the candidate. While this is positively correlated with is total PAC contributions to the candidate. While this is positively correlated with total there isabout now a (and it marginally signicant) negative between total stories is total PAC contributions to candidate. While this iscorrelation positively correlated with total stories a small candidate, is the uncorrelated with the share of PAC stories. stories about a candidate, it is uncorrelated with the share of PAC stories. andIn the concentration index. stories about a candidate, it is uncorrelated with the share of PAC short, these results lend further credibility to the validity of stories. using the concentration In short, these results lend further credibility to the validity of using the concentration These are reassuring about the validity the In particular, the In short, these results lend further credibility to the of validity of main usingconclusions, the concentration indices as results a proxy forvery capture. To provide further support for approach. the I now indices as a proxy for capture. To provide further support for the main conclusions, I now positiveas and signicant correlation the share of stories for a conclusions, candidate and the indices a proxy for capture. To between provide further support for the main I now present additional results that highlight the key role ofPAC information and the robustness of present additional results that highlight the key role of information and the robustness of candidates Herndahl concentration index suggests that the concentration indices do proxy present additional results that highlight the key role of information and the robustness of the main results. the main results. for media exposure of campaign contribution proles. And the weak correlation between the the main results. candidates Herndahl and his total number of stories suggests the former is not capturing

4 Other results: turnout and information spillovers di erences in candidate charisma or popularity. Along these lines, columns 7-9 in Table 8 4 Other results: turnout and information spillovers 4 Other results: turnout and information spillovers present a nal set of regressions like those in columns 1-3, but where the dependent variable
4.1 is totalTurnout PAC contributions to the candidate. While this is positively correlated with total 4.1 4.1 Turnout Turnout 23

stories a candidate, it is that uncorrelated with the share of PAC stories.. Since the sugTheoryabout and evidence suggest better information increases turnout Theory and evidence suggest that better information increases turnout23 . Since the sugTheory and evidence suggest that better information increases turnout . Since the sugIn short, these results lend further credibility the validity of using23 the concentration gested mechanism, whereby concentration of PAC to money aects incumbents vote shares, is gested mechanism, whereby concentration of PAC money aects incumbents vote shares, is gested mechanism, whereby concentration PAC money a ects incumbents vote shares, is indices as a proxy for capture. To provideof further support for the main conclusions, I now 23 23 For a review of some theoretical arguments, see Feddersen (2004) and in particular Feddersen and review of some theoretical arguments, see Feddersen (2004) and in and particular Feddersen information, a useful question to ask is whether turnout is higher in-state than out-of-state. 23 For a present additional results that highlight the key role information the robustness of Pesendorfer (1996); Fedderson and Pesendorfer (1999). Some of of the empirical was discussed in and the For a review of some theoretical arguments, see Feddersen (2004) and evidence in particular Feddersen Pesendorfer (1996); Fedderson and Pesendorfer (1999). Some of the empirical evidence was discussed in and the Introduction (see also Prat and Str o mbergs (2010) review). Pesendorfer (1996); Fedderson and Pesendorfer (1999). Some empirical discussed the Imain explore this question in Table 9. (2010) Columns 1 to of 3 the compare all evidence in-statewas counties to in outthe results. Introduction (see also Prat and Str o mbergs review). Introduction (see also Prat and Str ombergs (2010) review). of-state counties. The columns present regressions of the ratio of total votes to population 25 25 that equals 1 for in-state counties. Perhaps in each county on the baseline dummy variable 25 4 Other results: turnout and information spillovers surprisingly, columns 1 (with state and year xed eects) and 2 (with state-times-party and year-times-party xed eects) nd no signicant eect of being in-state on turnout. 4.1 Turnout However, recall that in-state counties are much more populous, which mechanically reduces Theory and evidence suggest that better increases turnout23 . for Since the sugthe dependent variable by increasing the information denominator. After controlling population gested whereby concentration PAC money aects incumbents vote shares, on the mechanism, right hand side (as in column 3), of in-state counties do have a turnout around 1.is 5 percentage points than out-of-state counties. This (2004) suggests population grows faster 23 For a review of higher some theoretical arguments, see Feddersen and in particular Feddersen and Pesendorfer (1996); and Pesendorfer (1999). Some of the empirical evidence was discussed in the than turnout andFedderson is suggestive of the role of the better information in in-state counties. Introduction (see also Prat and Str ombergs (2010) review). Perhaps more convincingly, columns 4 to 6, which replicate the regressions of columns 1 to 3 but for contiguous counties only, show that 25 irrespective of controlling for total population, 27 in-state counties do have larger turnout than their otherwise relatively similar out-of-state neighbors. The coecient is quite stable and implies a 1.7 to 2 percentage point higher

in each county on the baseline dummy variable that e equals for in-state Perhaps surprisingly, columns 1 (with state and year xed ects)1 and 2 (with counties. state-times-party in each county on the baseline dummy variable that e equals for in-state Perhaps surprisingly, columns 1 (with state and year xed ects)1and 2 (with counties. state-times-party surprisingly, columns 1 (with state and year xed e ects) and 2 (with state-times-party and year-times-party eects) signicant eect and of being in-state on turnout. surprisingly, columns xed 1 (with state nd and no year xed eects) 2 (with state-times-party and year-times-party xed eects) nd no signicant eect of being in-state on turnout. and year-times-party xed e ects) nd no signicant e ect of being in-state on turnout. However, recall that in-state counties are much more populous, which mechanically reduces and year-times-party xed e ects) nd signicant eect of which being mechanically in-state on turnout. However, recall that in-state counties areno much more populous, reduces However, recall variable that in-state counties are much more populous, mechanically reduces the dependent by increasing the denominator. Afterwhich controlling for population However, recall that in-state counties are much more populous, which mechanically reduces the dependent variable by increasing the denominator. After controlling for population the dependent variable by increasing the denominator. After controlling for population on the right hand side (as in column the 3), in-state counties do have a turnout 1.5 the dependent variable by increasing denominator. After controlling for around population on the right hand side (as in column 3), in-state counties do have a turnout around 1.5 on the right handhigher side (as in out-of-state column 3), counties. in-state counties do have a turnout around 1.5 percentage points than This suggests population grows faster on the right hand side (as in column 3), in-state counties do have a turnout around 1.5 percentage points higher than out-of-state counties. This suggests population grows faster percentage points higher than out-of-state counties. This suggests population grows faster than turnout and higher is suggestive of the role of the better information in in-state counties. percentage points than out-of-state counties. This suggests population grows faster than turnout and is suggestive of the role of the better information in in-state counties. than turnout and is suggestive of the role of the better information in in-state counties. Perhaps more 4 to which replicate the regressions of counties. columns 1 to than turnout andconvincingly, is suggestive columns of the role of6, the better information in in-state Perhaps more convincingly, columns 4 to 6, which replicate the regressions of columns 1 to Perhaps more convincingly, columns 4 to 6, which replicate the regressions of columns 1 to 3 but for contiguous counties only, show4that of controlling for total population, Perhaps more convincingly, columns to 6,irrespective which replicate the regressions of columns 1 to 3 but for contiguous counties only, show that irrespective of controlling for total population, 3 but for contiguous counties only, show that irrespective of controlling for total population, in-state do have larger turnout thanirrespective their otherwise relativelyfor similar 3 but forcounties contiguous counties only, show that of controlling total out-of-state population, in-state counties do have larger turnout than their otherwise relatively similar out-of-state in-state counties do have larger turnout than their otherwise relatively similar out-of-state neighbors. The coe cient is quite stable and implies a 1.7 to 2 percentage higher in-state counties do have larger turnout than their otherwise relatively similarpoint out-of-state neighbors. The coe cient is quite stable and implies a 1.7 to 2 percentage point higher neighbors. The coe cient isWhile quite we stable and implies a 1.7sets to 2 higher turnout in in-state counties. have seen that these of percentage contiguous point counties are neighbors. The coe cient is quite stable and implies a 1 . 7 to 2 percentage point higher turnout in in-state counties. While we have seen that these sets of contiguous counties are turnout in in-state counties. While we haveexplain seen that these sets of behavior contiguous counties and are quite similar, di erences remain that could this di erential of turnout turnout in in-state counties. While we haveexplain seen that sets of behavior contiguous counties and are quite similar, dierences remain that could thisthese dierential of turnout quite similar, with dierences remain For thatthis could explain this di erential behavior of turnout and be unrelated information. reason, columns 7 to 9 present a regression of the quite similar, with dierences remain For thatthis could explain this di behavior of turnout be unrelated information. reason, columns 7 erential to 9 present a regression of and the be unrelated with information. For this reason, columns 7 to 9 present a regression of the dependent variable on the placebo out-of-state counties versus their neighbors as before. be unrelated with information. For out-of-state this reason, columns to 9 present a regression of the dependent variable on the placebo counties 7 versus their neighbors as before. dependent variable on signicant the placebo out-of-state counties versus their neighbors as before. In this case, I nd no e ects, conrming the hypothesis that the di erences in dependent variable on signicant the placebo counties versus their neighbors as before. In this case, I nd no eout-of-state ects, conrming the hypothesis that the dierences in In this case, I in nd no signicant e ects, by conrming the and hypothesis that the di erences in turnout found columns 3-6 are driven information not by other di erences from In this case, I in nd no signicant ects, by conrming the and hypothesis the di erences in turnout found columns 3-6 are e driven information not by that other di erences from turnout towards found inout-of-state columns 3-6 are driven by information and not by other dierences from moving counties. turnout towards found inout-of-state columns 3-6 are driven by information and not by other dierences from moving counties. moving towards out-of-state counties. At rst glance, these result may seem to contradict some studies (e.g. Gentzkow (2006)) moving towards out-of-state counties. At rst glance, these result may seem to contradict some studies (e.g. Gentzkow (2006)) At rst glance, these result seem on to contradict some the studies (e.g. impact Gentzkow (2006)) that nd a glance, negative impact of may television turnout, unlike positive of radio or At rst these result seem on to contradict some the studies (e.g. impact Gentzkow (2006)) that nd a negative impact of may television turnout, unlike positive of radio or that nd a negative impact of television on turnout, unlike the positive impact of radioout or newspapers discussed in the Introduction. The erosion of social capital and the crowding that nd a negative of television on turnout, the positive impact of radioout or newspapers discussedimpact in the Introduction. The erosionunlike of social capital and the crowding newspapers discussed in the Introduction. The erosion of social capital and the crowding out of other media with better political coverage have been suggested as possible mechanisms for newspapers discussed in the Introduction. The erosion social capital and the crowding out of other media with better political coverage have been of suggested as possible mechanisms for of other media with better political coverage have been suggested as possible mechanisms for this e ect. Notice, however, that the results presented above are not necessarily inconsistent of other media withhowever, better political coverage have beenabove suggested as possible mechanisms for this eect. Notice, that the results presented are not necessarily inconsistent this esuch ect.nding. Notice, The however, thatbetween the results presented above are not necessarily inconsistent with variation in-state and out-of-state counties is not determined this esuch ect.nding. Notice, The however, thatbetween the results presented above are not necessarily with variation in-state and out-of-state counties is notinconsistent determined with such nding. The variation between in-state and out-of-state counties is not determined by the availability of TV, but the relevance of its content for voters. On the contrary, with such nding. The variation between in-state out-of-state is not determined by the availability of TV, but the relevance of and its content for counties voters. On the contrary, by the availability of TV, but the relevance of its content for voters. On the contrary, given the well-established connection between information andfor turnout, these results further by the availability of TV, but the relevance of its content voters. On the contrary, given the well-established connection between information and turnout, these results further given the well-established connectioncounties betweenainformation and turnout, these results further conrm that moving to out-of-state ects information relevant for voters. given the well-established connectioncounties betweenainformation and turnout, these results further conrm that moving to out-of-state ects information relevant for voters. conrm that moving to out-of-state counties aects information relevant forvote voters. But is the main channel for the earlier results concerning incumbent share that conrm that moving to out-of-state aects information relevant forvote voters. But is the main channel for thecounties earlier results concerning incumbent share that But is the main channel for the earlier results concerning incumbent vote share that news about captured politicians discourages voters in-state relative to out-of-state? We But is the main channel for the earlier results concerning incumbent vote share that news about captured politicians discourages voters in-state relative to out-of-state? We news about captured politicians discourages voters in-state relative to out-of-state? We can investigate this question by running the main specication with turnout, rather than news about captured politicians discourages voters in-state relative to out-of-state? We can investigate this question by running the main specication with turnout, rather than can investigate this question by running the main specication with turnout, rather than can investigate this question by running the main specication with turnout, rather than 26 variable. This is done in Table 10. As is vote share for the incumbent, as the dependent 26 26 for the incumbent does not seem to reduce clear from the table, an increased Herndahl 26 turnout dierentially in-state relative to out-of-state. This is suggestive that turnout is not the main or sole channel. Moreover, by looking at the non-interacted eect of the incumbent concentration index, we see that high Herndahl years do not seem systematically correlated with turnout, suggesting that eects presented for the vote share of the incumbent do not mechanically follow from dierences in turnout when the Herndahl is high.

4.2

Spillovers

The results presented thus far, by revealing the electoral cost that captured politicians 28 discussion about whether the concentration face with better-informed voters, open up the of campaign nance money for Senate candidates has negative spillovers for fellow party

the main or sole channel. Moreover, by looking at the non-interacted esystematically ect of the incumbent concentration index, we see that high Herndahl years do not seem correconcentration index, we see that high Herndahl years do not seem systematically correconcentration index, we see that high Herndahl do vote not seem correlated with turnout, suggesting that eects presentedyears for the share systematically of the incumbent do lated with turnout, suggesting that eects presented for the vote share of the incumbent do lated with turnout, suggesting that e ects presented for the vote share of the incumbent do not mechanically follow from dierences in turnout when the Herndahl is high. not mechanically follow from dierences in turnout when the Herndahl is high. not mechanically follow from dierences in turnout when the Herndahl is high.

The results presented thus far,as by revealing the variable. electoral cost captured vote share for the incumbent, the dependent This that is done in Table politicians 10. As is The results presented thus far, by revealing the electoral cost that captured politicians The results presented thus far, by revealing the electoral cost that captured politicians face with voters, open up the for discussion about whether the concentration clear frombetter-informed the table, an increased Herndahl the incumbent does not seem to reduce face with better-informed voters, open up the discussion about whether the concentration face with better-informed voters, open up the discussion about whether the concentration of campaign nance money Senate has negative spillovers for fellow is party turnout dierentially in-statefor relative to candidates out-of-state. This is suggestive that turnout not of campaign nance money for Senate candidates has negative spillovers for fellow party 24 of campaign nance money for Senate candidates has negative spillovers for fellow party . the In columns 1 members running for other oces. by I examine this in Table 1124 the main or sole channel. Moreover, looking at thequestion non-interacted eect of . In incumbent columns 1 members running for other oces. I examine this question in Table 1124 . Inpresidential columns 1 members running for other o ces. I examine this question in Table 11 and 2, I present results for the main specication but with vote share for the concentration index, we for see the thatmain high Herndahl but years do vote not seem systematically correand 2, I present results specication with share for the presidential and 2, I present results for the main specication but with vote share for the presidential candidate the same party as the incumbent senatorfor functioning as theof dependent variable. lated with of turnout, suggesting that eects presented the vote share the incumbent do candidate of the same party as the incumbent senator functioning as the dependent variable. candidate of the same party as the incumbent senator functioning as the dependent variable. The mechanically standard concentration (and campaign nance however, for the not follow frommeasures dierences in turnout when the controls), Herndahl is high. are The standard concentration measures (and campaign nance controls), however, are for the The standard concentration (and campaign controls), however, are for the Senate candidates. If there measures are signicant spillovers,nance the coe cient on the interaction of Senate candidates. If there are signicant spillovers, the coecient on the interaction of Senate candidates. there are signicant spillovers, coecient on should the interaction of the Herndahl indexIf for the incumbent senator and thethe in-state dummy be negative 4.2 Spillovers the Herndahl index for the incumbent senator and the in-state dummy should be negative the Herndahl index for the of incumbent and the in-state bepositive). negative (and that on the interaction the indexsenator for the opponent senatordummy and theshould dummy, (and that on the interaction of the index for the opponent senator and the dummy, positive). The results presented thus far, byindex revealing the electoral cost that captured (and that onwould the interaction of the for the opponent senator and the dummy,politicians positive). Indeed, this be an indication that in years where, for example, the Democratic senator Indeed, this would be an indication that in years where, for example, the Democratic senator face with better-informed voters, open up the discussion about whether the concentration Indeed, this would be an indication that in years where, for example, the Democratic senator has a relatively high Herndahl, the Democratic presidential candidate does worse in in-state has a relatively high Herndahl, the Democratic presidential candidate does worse in in-state of campaign nance money for the Senate candidates has negative for fellow party has a relatively high Herndahl, Democratic presidential candidate does worse in in-state counties relative to out-of-state counties. Column 1 presents thespillovers results for the PAC-level counties relative to out-of-state counties. Column 1 presents the results for 24 the PAC-level . zero. In columns 1 members running for other o ces. I examine this Table 11for counties relative out-of-state counties. Column 1 question presents in the results the PAC-level Herndahl index. to None of the interaction terms is signicantly di erent from A similar Herndahl index. None of the interaction terms is signicantly di erent from zero. A similar and 2,is I obtained present for the main specication but with vote share from for presidential Herndahl index.results None the interaction terms is signicantly dicolumn erent zero. A similar result in theof case of the industry-level Herndahl in 2. the result is obtained in the case of the industry-level Herndahl in column 2. candidate of the same party as the incumbent senator functioning as the the dependent variable. result is obtained in the case of the industry-level Herndahl in column 2. A similar exercise can be accomplished with the vote share for House candidate of A similar exercise can be accomplished with the vote share for the House candidate of The standard concentration (and campaign nance controls), are for the 25 A similar exercise bemeasures accomplished with the variable vote share for the however, House candidate of the same party as thecan incumbent as the dependent 25 . The results are reported in . The results are reported in the same party as the incumbent as the dependent variable 25 Senate candidates. If there are signicant spillovers, the coe.The cient on the interaction of results are reported in the same the incumbent as theindustry-level dependent variable columns 3party and 4as for the PAC-level and concentration measures, respectively. columns 3 and 4 for the PAC-level and industry-level concentration measures, respectively. the Herndahl for the incumbent and in-state dummy should be negative columns 3 and index 4 forestimated the PAC-level and senator industry-level concentration measures, respectively. Though imprecisely (the standard error is 0the .316), the negative coe cient of column Though imprecisely estimated (the standard error is 0 . 316), the negative coe cient of column (and that on the interaction of the index for the opponent senator and the dummy, positive). Though imprecisely estimated (the standard error is 0 . 316), the negative coe cient of column 24 All regressions in this table focus on the set of contiguous counties, include a full set of county times 24 regressions in this table focus on the set of contiguous counties, include a senator full set of county times Indeed, this would be an indication that in years where, for example, the Democratic senator 3 (24 All 0 . 594) suggests the presence of spillover e ects. That is, a tainted a ects fellow party and year times party xed e ects, as well as the campaign nance controls, county controls, and state All regressions in this xed table e focus on the set of contiguous counties, include county a full set of county times party and year times party ects, as well as the campaign nance controls, controls, and state Herndahl controls described above. party year times party xed efurther ects, asDemocratic well as the campaign controls, controls, state has a and relatively high Herndahl, the presidential does in and in-state congressmen. However, upon examination, this nance doescandidate not seemcounty to beworse the case. First, Herndahl controls described above. 25 case, since voting and other data is at the county-level, a decision must be taken regarding Herndahl controls described above. 25 In this In this case, since voting other data is Column at the county-level, a the decision must be taken regarding counties to out-of-state counties. 1 an presents results for column 4relative for the industry-level Herndahl estimates almost identical the cient, yet it is 25 the treatment of counties with and multiple congressional districts. Results presented incoe Table 11 PAC-level simply add In this case, since voting and other data is at the county-level, a decision must be taken regarding the treatment of counties with multiple congressional districts. Results presented in Table 11 simply add the votes for all the Democratic and Republican candidates running for Congress in a county to nd the the of the counties with multiple congressional districts. Results in 11 simply Herndahl index. None of the interaction terms is signicantly di erent from zero. A similar nottreatment statistically signicant. Second, and more importantly, columns 5 and 6 control for add the the votes forand all Democratic Republican candidates running for presented Congress in Table awhen county to nd the Democratic Republican vote and share per county. I veried that results are similar restricting to the votes for all the Democratic and Republican candidates running for Congress in a county to nd the Democratic and Republican vote share per county. I veried that results are similar when restricting to result is obtained in the case of the industry-level Herndahl in column 2. House candidates campaign contributions concentration measures, and their interaction with counties with no more than one congressional district. Democratic and vote share per county. I veried that results are similar when restricting to counties with no Republican more than one congressional district. counties with no more than one congressional district. similar exercise can be accomplished with vote share for the House candidate of the A in-state dummy. Indeed, it could be that thethe signicant negative interaction in column 25 27 House . Thecandidates results arecontribution reported in the party as the the dependent variable 3 is same in fact driven by incumbent correlation as between the and Senate 27 27 columns and 4the forsame the PAC-level and industry-level concentration measures, respectively. patterns 3 within county and year. After all, Senate and Congress candidates of the Though imprecisely estimated (the standard error isnance 0.316), proles. the negative cient of column same party and state may have similar campaign Thecoe results suggest this
24 because the coein cients on the interaction with the Senate candidates Herndahls are All regressions this table focus on the setterms of contiguous counties, include a full set of county times party and year and timesstatistically party xed e ects, as well as the campaign nance controls, county controls, and state now smaller indistinguishable from zero. Herndahl controls described above. 25 While of theand absence of important spillover e results Table 11 at In this indicative case, since voting other data is at the county-level, aects, decision must from be taken regarding the of counties with multiple congressional districts. Results presented Table 11 simply add the treatment same time reassure that the results presented before are not driveninby partisan trends. the votes for all the Democratic and Republican candidates running for Congress in a county to nd the Indeed, a potentially serious to the models thus far is that Democratic and Republican vote objection share per county. I veriedestimated that results are similar whenpartisanship restricting to counties with more directions than one congressional district. moves in dino erent in dierent counties across dierent years. Unfortunately, I cannot control for a full set of county-year xed eects, since the variation I use to estimate 27 the key interaction terms is at this level. However, 29 the President and House regressions just examined can be considered placebo regressions to rule out this interpretation. If important

4.2 4.2 4.2

Spillovers Spillovers Spillovers

not statistically signicant. and more importantly, columns 5 their and 6 control for the House candidates campaign Second, contributions concentration measures, and interaction with House candidates campaign contributions concentration measures, and their interaction with House candidates campaign concentration measures, andinteraction their interaction with the in-state dummy. Indeed,contributions it could be that the signicant negative in column the in-state dummy. Indeed, it could be that the signicant negative interaction in column the Indeed, it could be that the signicant negative interaction in column 3 is in-state in fact dummy. driven by correlation between the House and Senate candidates contribution 3 is in fact driven by correlation between the House and Senate candidates contribution 3 is in fact driven correlation between the House and Senate candidates contribution patterns within the by same county and year. After all, Senate and Congress candidates of the patterns within the same county and year. After all, Senate and Congress candidates of the patterns within same county and year. After all, Senate and Congress candidates of this the same party and the state may have similar campaign nance proles. The results suggest same party and state may have similar campaign nance proles. The results suggest this same party state may have similar campaign nance proles. The results suggest this because the and coe cients on the interaction terms with the Senate candidates Herndahls are because the coecients on the interaction terms with the Senate candidates Herndahls are because the coe statistically cients on the interaction terms withzero. the Senate candidates Herndahls are now smaller and indistinguishable from now smaller and statistically indistinguishable from zero. now smaller and statistically indistinguishable from zero. eects, results from Table 11 at While indicative of the absence of important spillover While indicative of the absence of important spillover eects, results from Table 11 at indicative of the of important enot ects, results from Table 11 at the While same time reassure thatabsence the results presented spillover before are driven by partisan trends. the same time reassure that the results presented before are not driven by partisan trends. the same reassure that objection the results presented before are not driven by partisan trends. Indeed, a time potentially serious to the models estimated thus far is that partisanship Indeed, a potentially serious objection to the models estimated thus far is that partisanship Indeed,in a potentially serious objection to the models estimated thusyears. far is that partisanship moves dierent directions in dierent counties across dierent Unfortunately, I moves in dierent directions in dierent counties across dierent years. Unfortunately, I moves in di erent directions in di erent counties across di erent years. Unfortunately, I cannot control for a full set of county-year xed eects, since the variation I use to estimate cannot control for a full set of county-year xed eects, since the variation I use to estimate cannot for aterms full set xed eects, since the and variation I regressions use to estimate the key control interaction is of at county-year this level. However, the President House just the key interaction terms is at this level. However, the President and House regressions just the key interaction terms is at this level. However, the President and House regressions just examined can be considered placebo regressions to rule out this interpretation. If important examined can be considered placebo regressions to rule out this interpretation. If important examinedtrends can bewere considered regressions to rule outfor this interpretation. important partisan drivingplacebo the results, observed eects the vote share ofIfthe incumpartisan trends were driving the results, observed eects for the vote share of the incumpartisan trends were driving the results, observed e ects for the vote share of the incumbent should be mirrored in specications with the vote share of fellow party members as a bent should be mirrored in specications with the vote share of fellow party members as a bent should be mirrored in specications the vote of credibility fellow party as of a dependent variable instead. They do not, with and this gives share further to members the results dependent variable instead. They do not, and this gives further credibility to the results of dependent Section 3. variable instead. They do not, and this gives further credibility to the results of Section 3. Section 3. Information spillovers, however, may take several other forms than simply the contagion Information spillovers, however, may take several other forms than simply the contagion Information spillovers, however, may take several forms than simply the counties contagion of representatives from the same party and state. other Specically, out-of-state are of representatives from the same party and state. Specically, out-of-state counties are of representatives same party states and state. Specically, out-of-state counties exposed relatively from more the to neighboring political information. Thus, there may are be exposed relatively more to neighboring states political information. Thus, there may be exposed relatively more to neighboring states political information. Thus, there may be partisan spillovers from neighboring states. Imagine an out-of-state county c of state s sitting partisan spillovers from neighboring states. Imagine an out-of-state county c of state s sitting 0 out-of-state county c of state s sitting partisan spillovers neighboring states. Imagine an . There are at least two questions that in a media market from dominated by neighboring state s in a media market dominated by neighboring state s00 . There are at least two questions that . There are least two questions that in a media market dominated by neighboring state s seem worth asking. First, to examine the hypothesis that voters inat these out-of-state counties seem worth asking. First, to examine the hypothesis that voters in these out-of-state counties seem worth asking. First, to examine hypothesis that in these out-of-state are on average misinformed votersthe who must rely on voters a partisan cue, one could counties ask: do are on average misinformed voters who must rely on a partisan cue, one could ask: do are on average misinformed voters who must rely on a partisan cue, one could ask: do voters in county c punish the candidate of party P in state s when the Herndahl of the voters in county c punish the candidate of party P in state s when the Herndahl of the 0 voters in county c punish the candidate of sparty P Second, in stateasdi when the Herndahl of the rises? erent hypothesis suggests candidate for party P in neighboring state candidate for party P in neighboring state s00 rises? Second, a dierent hypothesis suggests Second, dierent hypothesis candidate for party P inare neighboring state s rises? that out-of-state voters not misinformed, but are bettera informed on average suggests because that out-of-state voters are not misinformed, but are better informed on average because 0 that get out-of-state voters areboth not states, misinformed, are better informed on average because . Hence, they may rely on comparisons, and, they information about s and sbut they get information about both states, s and s00 . Hence, they may rely on comparisons, and, . Hence, they rely onof comparisons, and, they get information about both states, and sin for instance, it is possible to examine if s voters c punish themay candidate party P in state for instance, it is possible to examine if voters in c punish the candidate of party P in state for instance, it is possible to examine if voters in c punish the candidate of party P in state 28 than that in state s0 . s if the Herndahl of party P in state s is larger 28 28table focuses on out-of-state counties only26 . I examine both questions in Table 12 . The In line with previous results, columns 1 and 2 verify that the dependent variable, the incumbent senators vote share, does not react to his Herndahl index or that of his opponent in out-of-state counties. Columns 3 and 4 ask if it responds instead to the concentration indices of the neighboring states candidates. The coecients are again not signicant, with small point estimates. Finally, columns 5 and 6 check whether it could be the dierence between the incumbents Herndahl and the Herndahl of his fellow party member running in a neighboring state (and similarly for the opponent) that aects voting patterns in out-of-state counties. Once again, the point estimates are close to zero and statistically insignicant. Overall, this section suggests that partisan trends are unlikely to drive the main results 30 of the paper. Additionally, information spillovers across members of the same party, whether in the same or neighboring states, are not rst-order.

small point estimates. Finally, columns 5 and 6 check whether it could be the dierence besmall point estimates. Finally, columns 5 and 6 check whether it could be the dierence between the incumbents Herndahl and the Herndahl of his fellow party member running in a tween the incumbents Herndahl and the Herndahl of his fellow party member running in a neighboring state (and similarly for the opponent) that aects voting patterns in out-of-state neighboring state (and similarly for the opponent) that aects voting patterns in out-of-state counties. Once again, the point estimates are close to zero and statistically insignicant. counties. Once again, the point estimates are close to zero and statistically insignicant. Overall, this section suggests that partisan trends are unlikely to drive the main results Overall, this section suggests that partisan trends are unlikely to drive the main results of the paper. Additionally, information spillovers across members of the same party, whether ofif the paper. Additionally, members of the s the Herndahl of partyinformation P in state sspillovers is larger across than that in state s0 . same party, whether in the same or neighboring states, are not rst-order. in the same or neighboring states, are 12 not I examine both questions in Table . rst-order. The table focuses on out-of-state counties only26 . In line with previous results, columns 1 and 2 verify that the dependent variable, the incumbent senators vote share, does not react to his Herndahl index of or that of his opponent 5 Further robustness checks: denition media mar-

in out-of-state counties. Columns 3 and 4 ask if it responds instead to the concentration kets kets indices of the neighboring states candidates. The coecients are again not signicant, with

Further robustness checks: denition of media mar-

small point estimates. Finally, columns 5 and 6 check it could be the di erence beBefore concluding, I demonstrate the robustness of the whether main result to alternate denitions of Before concluding, I demonstrate the robustness of the main result to alternate denitions of tween the incumbents Herndahl and the of his fellow member running in a the media market. Recall that a county is Herndahl in-state if the state the party county is in has more than the media market. Recall that a county is in-state if the state the county is in has more than neighboring state (andof similarly for the opponent) that aects if voting patterns in (100 out-of-state x % of the population that media market, and out-of-state it has less than x)%. x% of the population of that media market, and out-of-state if it has less than (100 x)%. counties. Once again, theare point estimates close zero x. and statistically insignicant. Results reported thus far for x = 66. In are Table 13,to I vary I focus on the most demanding Results reported thus far are for x = 66. In Table 13, I vary x. I focus on the most demanding Overall, this section suggests partisan are unlikely to drive theat main results specication with county xed ethat ects and a fulltrends set of controls. Panel A looks all in-state specication with county xed eects and a full set of controls. Panel A looks at all in-state of the paper. Additionally, information spillovers across members of the same and party, whether and out-of-state counties, whereas Panel B restricts attention to the in-state contiguous and out-of-state counties, whereas Panel B restricts attention to the in-state and contiguous in the same or neighboring states, counties. are not rst-order. counties compared to out-of-state In odd columns, the Herndahl index is at the counties compared to out-of-state counties. In odd columns, the Herndahl index is at the PAC level. In even columns, it is computed at the industry level. Finally, x = 50 in columns PAC level. In even columns, it is computed at the industry level. Finally, x = 50 in columns 1-2, = 60 in columns 3-4, and x = checks: 70 in columns 5-6. The tableof shows that themarresults 5 x robustness denition media 1-2, xFurther = 60 in columns 3-4, and x = 70 in columns 5-6. The table shows that the results do not depend on the exact threshold of population to dene media markets. In all cases, do not depend on the exact threshold of population to dene media markets. In all cases, is negative, typically signicant at more than a 99% condence level, and the point inhIkets inhI is negative, typically signicant at more than a 99% condence level, and the point estimate is very close in all specications to results presented in Table 5. estimate is very close in all specications to results presented in Table 5. Before concluding, I demonstrate the robustness of the main result to alternate denitions of An alternative approach to check the robustness of the results to the media market alternative approach to check the robustness the the results to is the media market the An media market. Recall that a county is in-state if theof state county in has more than denition is to rely directly on the share of in-state media market population as the key, denition to rely directly the market, share of and in-state media market as the x% of the is population of that on media out-of-state if it haspopulation less than (100 xkey, )%. 26 continuous, measure of availability of information in each county. More concretely, for each all regressions in are this for table include a Table full set 13, of county party on and year times party xed Results reported thus far x= 66. In I varytimes x. I focus the most demanding 26 Again, Again, all regressions in this table include a full set of county and times party and yearcontrols times party xed eects, as well as the campaign nance controls, county controls, state Herndahl described county cwell in state scampaign and media market (Area ofset Dominance Inuence) m, dene the ratio e ects, as as the nance controls, county controls, and state Herndahl controls described specication with county xed e ects and a full of controls. Panel A looks at all in-state above. above. in-state as the counties, ratio of the population of restricts media market m residing in state s to the total and out-of-state whereas Panel B attention to the in-state and contiguous population of market m. This variable can then be columns, used, in place of the in state dummy, in 29 odd counties compared to out-of-state counties. In the Herndahl index is at the 29 specications analogous to (1) and (2). The presented Table 14, are with PAC level. In even columns, it is computed atresults, the industry level.in Finally, x= 50 in in line columns the ones presented previously, both for PAC-level concentration indices (Panel orresults at the 1-2, x = 60 in columns 3-4, and x= 70 in columns 5-6. The table shows thatA) the industry-level Regardless we focus on state eects do not depend (Panel on the B). exact thresholdof ofwhether population to dene mediaxed markets. Inregressions all cases, (columns 1 and 2), county xed regressions (columns 3 and 4), or county xed eects re inhI is negative, typically signicant at more than a 99% condence level, and the point gressionsis among the set of specications more comparable contiguous counties as dened by the in-state estimate very close in all to results presented in Table 5. dummy (columns 5 and 6), the message the same. of An increase concentration of An alternative approach to check theisrobustness the resultsin tothe the media market campaign is contributions for the incumbent leads to a media fall in market his votepopulation share in counties denition to rely directly on the share of in-state as the with key, media markets that have a larger proportion of their population in the countys state.
Again, all regressions in this table include a full set of county times party and year times party xed eects, as well as the campaign nance controls, county controls, and state Herndahl controls described above.
26

Final Remarks

29 The evidence presented in this paper supports 31 the idea that mass media, by informing voters, may reduce the inuence of special interest groups. Better access to mass media

industry-level (Panel B). Regardless of whether we focus on state xed eects regressions (columns 1 and 2), county xed regressions (columns 3 and 4), or county xed eects regressions among the set of more comparable contiguous counties as dened by the in-state dummy (columns 5 and 6), the message is the same. An increase in the concentration of campaign contributions for the incumbent leads to a fall in his vote share in counties with media markets that have a larger proportion of their population in the countys state. politicians is inevitably a noisy measure of the degree of capture27 . Second, notice that if concentrating contribution patterns likely hurt politicians, those politicians who are willing to fund their campaigns money from a narrow setidea of interest groups should also possess The evidence presented with in this paper supports the that mass media, by informing other resources to counteract the of negative this entails. Better In thisaccess sense,to the estimated voters, may reduce the inuence specialsignal interest groups. mass media e ects likely the impact on an average politician of the public disclosure of allows votersunderestimate to react to negative information about their candidates, and in particular the his links to inuence of special narrow interests. interests in policy as proxied by their campaign nance proles. For this Also related,others) notice that the reduction in candidate vote mass share media is onlyis one of the channels reason (among the presence of a free, independent a key component whereby mass political media may reduce special interest group inuence in politics. To provide just of democratic institutions. one Hence, example, media disclosure of potentially undue interest group role inuence may, even themass paper contributes to the growing evidence on the crucial of the media in when not aecting behavior,accountability. trigger or facilitate the so, action of otherainstitutions, empowering voters voter and improving In doing it suggests channel thatsuch has as courts, to hold politicians accountable. received relatively little attention in the empirical literature. As noted in the Introduction, Finally, alonga these lines the obtained here are While many there is by now fair amount of results evidence on the impact ofcontext-specic. free media on voter turnout, studies have documented the bias in political in the choices. US, the extent of misinformation voter knowledge, and representative behavior news and policy But special interests are 28 In mass such and bias in many other countries in the The world is a farpresented more serious fundamental players in any democracy. evidence here problem suggests . that contexts, it is notreduce clear that greater access to mass media voters at large to put media may help their inuence. Candidates who will are enable more likely to be connected a check on potentially undue less inuence narrowin interest with interest groups receive votes from on average places groups. where they face greater media exposure. This relation provides a direct eect on reduced interest group inuence. Admittedly, the magnitude of the eects found is relatively small, and typically would not decide the election outcome for the average incumbent. Nonetheless, there are various reasons why the estimated eect may underestimate the impact, for voters, of revealed capture of politicians by special interests. First, the concentration of campaign contributions to politicians is inevitably a noisy measure of the degree of capture27 . Second, notice that if 30 politicians, those politicians who are willing concentrating contribution patterns likely hurt to fund their campaigns with money from a narrow set of interest groups should also possess other resources to counteract the negative signal this entails. In this sense, the estimated eects likely underestimate the impact on an average politician of the public disclosure of his links to special interests. Also related, notice that the reduction in candidate vote share is only one of the channels whereby mass media may reduce special interest group inuence in politics. To provide just one 27example, mass media disclosure of potentially undue interest group inuence may, even For a more recent period, a more direct measure of interest-group connectedness based on searches for news stories interest group inuence could provide a more precise of measure allow a reexamiwhen not aregarding ecting voter behavior, trigger or facilitate the action other and institutions, such nation of the magnitude of the eects. Still, the results presented in this paper are interesting in their own as courts, to hold politicians accountable. right, as they allow us to study the ways in which campaign contribution patterns directly and indirectly transmit information to voters. Finally, along these lines the results obtained here are context-specic. While many 28 In 2009, for instance, the US ranked 20 (of 175) on the Press Freedom Index published yearly by studies have documented the bias in political news in the US, the extent of misinformation Reporters Without Borders. and bias in many other countries in the world is a far more serious problem28 . In such 32 31 contexts, it is not clear that greater access to mass media will enable voters at large to put

Final Remarks

to fund their campaigns with money from a narrow set of interest groups should also possess to fund their campaigns with money from a narrow set of interest groups should also possess other resources to counteract the negative signal this entails. In this sense, the estimated other resources to counteract the negative signal this entails. In this sense, the estimated eects likely underestimate the impact on an average politician of the public disclosure of eects likely underestimate the impact on an average politician of the public disclosure of his links to special interests. his links to special interests. Also related, notice that the reduction in candidate vote share is only one of the channels Also related, notice that the reduction in candidate vote share is only one of the channels whereby mass media may reduce special interest group inuence in politics. To provide just 27politics. To provide just whereby mass media may special interest group inuence in politicians is inevitably a reduce noisy measure of the degree of capture . Second, notice that if one example, mass media disclosure of potentially undue interest group inuence may, even one example, mass media disclosure of potentially undue interest group inuence may, even concentrating contribution patterns likely hurt politicians, those politicians who are willing when not aecting voter behavior, trigger or facilitate the action of other institutions, such when not aecting voterwith behavior, trigger or facilitate the action groups of other institutions, such to fund their campaigns money from a narrow set of interest should also possess as courts, to hold politicians accountable. as courts, to hold accountable. other resources to politicians counteract the negative signal this entails. In 27 this sense, the estimated politicians inevitably noisy measure the degree capture . Second, notice if Finally, is along these a lines the results of obtained hereofare context-specic. While that many Finally, along these lines the results obtained herepolitician are context-specic. While many e ects likely underestimate the impact on an average of the public disclosure of concentrating contribution patterns hurt politicians, those politicians are willing studies have documented the bias inlikely political news in the US, the extent of who misinformation studies have documented the bias in political news in the US, the extent of misinformation his links to special interests. to fund their campaigns money from narrow of interest groups should 28 also possess and bias in many otherwith countries in the aworld is set a far more serious problem 28 . In such . channels In such andAlso biasrelated, in many other countries in the in world is a far more serious problem notice that the reduction candidate vote share is only one of the other resources to clear counteract the negative signal this entails. In thisvoters sense,at the estimated contexts, it is not that greater access to mass media will enable large to put contexts, it is not clear that greater access to mass media will enable voters To at large to just put whereby mass media may reduce special interest group inuence in politics. provide echeck ects likely underestimate the impact from on annarrow average politician of the public disclosure of a on potentially undue inuence interest groups. a check on potentially undue inuence from narrow interest groups. one example, mass media disclosure of potentially undue interest group inuence may, even his links to special interests. when not aecting voter behavior, trigger in orcandidate facilitate vote the action of only other institutions, such Also related, notice that the reduction share is one of the channels as courts, to hold politicians accountable. whereby mass media may reduce special interest group inuence in politics. To provide just along lines the results obtainedundue here are context-specic. While oneFinally, example, massthese media disclosure of potentially interest group inuence may,many even studies have documented the bias in political news in the theof extent misinformation when not aecting voter behavior, trigger or facilitate the US, action other of institutions, such 28 and bias in other countries in the world is a far more serious problem . In such as courts, to many hold politicians accountable. contexts, it is not clear access to mass here media will enable voters at While large to put Finally, along these that linesgreater the results obtained are context-specic. many a check have on potentially undue from narrow interest groups. studies documented the inuence bias in political news in the US, the extent of misinformation and bias in many other countries in the world is a far more serious problem28 . In such contexts, it is not clear that greater access to mass media will enable voters at large to put a check on potentially undue inuence from narrow interest groups.

a more recent period, a more direct measure of interest-group connectedness based on searches for a more recent interest period, a moreinuence direct measure of interest-group connectedness on searches for newsFor stories regarding group could provide a more precise measure based and allow a reexaminews stories regarding interest group inuence could provide a more precise measure and allow a reexamination of the magnitude of the eects. Still, the results presented in this paper are interesting in their own nationas of they the magnitude ofstudy the e ects. Still, results presented in this paper are interesting in their own right, allow us to the ways in the which campaign contribution patterns directly and indirectly right, as they allow us to study the ways in which campaign contribution patterns directly and indirectly transmit information to voters. transmit information to voters. 28 28 In 2009, for instance, the US ranked 20 (of 175) on the Press Freedom Index published yearly by In 2009, for instance, Reporters Without Borders.the US ranked 20 (of 175) on the Press Freedom Index published yearly by Reporters Without Borders.

27 27 For

31 31
For a more recent period, a more direct measure of interest-group connectedness based on searches for news stories regarding interest group inuence could provide a more precise measure and allow a reexamination of the magnitude of the eects. Still, the results presented in this paper are interesting in their own right, as they allow us to study the ways in which campaign contribution patterns directly and indirectly transmit information to voters. 27 28 For a more recent period, a more direct measure of interest-group connectedness based on searches for In 2009, for instance, the US ranked 20 (of 175) on the Press Freedom Index published yearly by news stories regarding interest group inuence could provide a more precise measure and allow a reexamiReporters Without Borders. nation of the magnitude of the eects. Still, the results presented in this paper are interesting in their own right, as they allow us to study the ways in which campaign contribution patterns directly and indirectly 31 transmit information to voters. 28 33 In 2009, for instance, the US ranked 20 (of 175) on the Press Freedom Index published yearly by Reporters Without Borders.
27

References
Acemoglu, D., Robinson, J.A., 2012. Why Nations Fail: The Origins of Power, Prosperity, and Poverty. Crown Publishers (Random House), New York, N.Y. Ansolabehere, S., Behr, R., Iyengar, S., 1993. The media game: American politics in the television age. Macmillan, New York. Ansolabehere, S., De Figueiredo, J.M., Snyder, J.M., 2003. Why is there so little money in US politics? The Journal of Economic Perspectives 17, 105130. Ansolabehere, S., Snowberg, E.C., Snyder, J.M., 2006. Television and the incumbency advantage in U.S. elections. Legislative Studies Quarterly 31, 469490. Ansolabehere, S., Snyder, J.M., 2002. The incumbency advantage in U.S. elections: An analysis of state and federal oces, 1942-2000. Election Law Journal: Rules, Politics, and Policy 1, 315338. Banerjee, A.V., Kumar, S., Pande, R., Su, F., 2010. Do informed voters make better choices? Experimental evidence from urban India. Unpublished manuscript. MIT, Yale and Harvard. Baron, D.P., 1994. Electoral competition with informed and uninformed voters. American Political Science Review 88, 3347. Besley, T., Burgess, R., 2002. The political economy of government responsiveness: Theory and evidence from India. The Quarterly Journal of Economics 117, 14151451. Coate, S., 2004. Pareto-improving campaign nance policy. American Economic Review 94, 628655. Dharmapala, D., Palda, F., 2002. Are campaign contributions a form of speech? Evidence from recent US House elections. Public Choice 112, 81114. Feddersen, T.J., 2004. Rational choice theory and the paradox of not voting. Journal of Economic Perspectives 18, 99112. Feddersen, T.J., Pesendorfer, W., 1996. The swing voters curse. American Economic Review 86, 40824.

32
34

Fedderson, T.J., Pesendorfer, W., 1999. Abstention in elections with asymmetric information and diverse preferences. American Political Science Review 93, 381398. Ferraz, C., Finan, F., 2008. Exposing corrupt politicians: The eects of Brazils publicly released audits on electoral outcomes. Quarterly Journal of Economics 123, 747793. Gentzkow, M., 2006. Television and voter turnout. The Quarterly Journal of Economics 121, 931972. Gentzkow, M.A., Shapiro, J.M., Sinkinson, M., 2009. The eect of newspaper entry and exit on electoral politics. NBER Working Paper Series, Working Paper 15544. Glaeser, E.L., Gentzkow, M.A., Goldin, C., 2004. The rise of the Fourth Estate: How newspapers became informative and why it mattered. NBER Working Paper Series, Working Paper 10791. Grossman, G., Helpman, E., 1996. Electoral competition and special interest politics. The Review of Economic Studies 63, 265286. Grossman, G.M., Helpman, E., 2001. Special Interest Politics. The MIT Press, Cambridge, MA. Houser, D., Stratmann, T., 2008. Selling favors in the lab: experiments on campaign nance reform. Public Choice 136, 215239. Johnson, S., Kwak, J., 2010. 13 bankers: the Wall Street takeover and the next nancial meltdown. Pantheon Books, New York. Kahn, K.F., 1993. Incumbency and the news media in U.S. Senate elections: An experimental investigation. Political Research Quarterly 46, 715740. Levitt, S.D., Snyder, J.M., 1997. The impact of federal spending on House election outcomes. Journal of Political Economy 105, 3053. Morton, R., Cameron, C., 1992. Elections and the theory of campaign contributions: a survey and critical analysis. Economics and Politics 4, 79108. Palda, F., Palda, K., 1998. The impact of campaign expenditures on political competition in the French legislative elections of 1993. Public Choice 94, 157174.

33

35

Prat, A., 2006. Rational voters and political advertising, in: Weingast, B.R., Wittman, D. (Eds.), The Oxford handbook of political economy. Oxford University Press, Oxford, UK, pp. 5063. Prat, A., Puglisi, R., Snyder, J.M., 2006. Is private campaign nance a good thing? Estimates of the potential informational benets. papers.ssrn.com . Prat, A., Strmberg, D., 2010. The political economy of mass media. Unpublished, London School of Economics and Stockholm University. Snyder, J.M., Str omberg, D., 2010. Press coverage and political accountability. Journal of Political Economy 118. Stratmann, T., 2005. Some talk: Money in politics. A (partial) review of the literature. Public Choice 124, 135156. Str omberg, D., 2001. Mass media and public policy. European Economic Review 45, 652 663. Str omberg, D., 2004a. Mass media competition, political competition, and public policy. Review of Economic Studies 71, 265284. Str omberg, D., 2004b. Radios impact on public spending. Quarterly Journal of Economics 119, 189221. Vanberg, C., 2008. One Man, One Dollar? Campaign contribution limits, equal inuence, and political communication. Journal of Public Economics 92, 514531.

34
36

A A A A A

A.1 Variable Sources and Denitions A.1 Variable Sources and Denitions A.1 Variable Sources and Denitions A.1 Variable Sources and See Appendix Table A.1 for the sources and denitions of the variables used in the analysis. A.1 Variable Sources and Denitions Denitions
See See See See

Data Data Data Data Data

Appendix Appendix Appendix Appendix Appendix

A.1 A.1 A.1 A.1 for for for for the the the the

As noted in the main text, to classify the PAC contributions in industry categories, I follow As noted infor theResponsive main text, Politics to classify the PAC contributions industry categories, I follow the Center coding system. This is ain hierarchical system, with ve As noted infor theResponsive main text, Politics to classify the PAC contributions industry categories, I follow the Center coding system. This is ain hierarchical system, with ve As noted in the main text, to classify the PAC contributions in industry categories, I follow the Center for Responsive Politics coding system. This is a hierarchical system, with ve super-categories (Business, Labor, Ideological/Single-Issue, Other and Unknown) divided the Center for Responsive Politics coding system. This is aOther hierarchical system, with ve super-categories (Business, Labor, Ideological/Single-Issue, and Unknown) divided the Center for Responsive Politics coding system. This is aOther hierarchical system, with ve super-categories (Business, Labor, Ideological/Single-Issue, and Unknown) divided into 13 sectors, about 100Labor, industries, and about 400 categories as shown in Appendix super-categories (Business, Ideological/Single-Issue, Other and Unknown) divided into 13 sectors, about 100 industries, and about 400 categories as shown in Appendix super-categories (Business, Ideological/Single-Issue, Other and Unknown) divided into 13 sectors, about 100Labor, industries, and about 400 categories as shown in Appendix Table A.2. into 13 sectors, about 100 industries, and about 400 categories as shown in Appendix Table A.2. into 13 sectors, about 100 industries, and about 400 categories as shown in Appendix Table A.2. The CRP has been classifying PACs since 1990, whereas I study campaign contributions Table A.2. The CRP has been classifying PACs since 1990, whereas I study campaign contributions Table A.2. The CRPresults has been classifying PACs since 1990, whereas I study contributions andThe election since 1980. From 1990 onwards, there are nearlycampaign 50, 000 PAC-candidate CRPresults has been classifying PACs since 1990, whereas I study campaign contributions andThe election since 1980. From 1990 onwards, there are nearly 50 , 000 PAC-candidate CRPresults has been classifying PACs since 1990, whereas I study campaign contributions and election since 1980. From 1990 onwards, there are nearly 50 , 000 PAC-candidate pairs in the sample precisely, there are 49, 739 PAC-candidate from 3, 295 PACs and election results (more since 1980. From 1990 onwards, there are nearlypairs, 50, 000 PAC-candidate pairs in the sample (more precisely, there are 49 , 739 PAC-candidate pairs, from 3, 295 PACs and election results (more since 1980. From 1990 onwards, there are nearlypairs, 50, 000 PAC-candidate pairs in the sample precisely, there are 49 , 739 PAC-candidate from 3, 295 PACs contributing to any of the Senate candidates in the sample after 1990). The classication of pairs in the sample (more precisely, there are 49 ,the 739sample PAC-candidate pairs, from 3, 295 PACs contributing to any of the Senate candidates in after 1990). The classication of pairs in the sample (more precisely, there are 49 , 739 PAC-candidate pairs, from 3 , 295 PACs contributing to any of the Senate in the sample after The classication of the interest group sector for thesecandidates PACs is facilitated by the fact1990). that the Federal Elections contributing to any of the Senate candidates in the sample after 1990). The classication of the interest group sector for thesecandidates PACs is facilitated by the fact1990). that the Federal Elections contributing to any of the Senate in the sample after The classication of the interest group sector for these PACs is facilitated by the fact that the Federal Elections Commission assigns a unique identication number to each committee, and this identication the interest group sector for these PACs is facilitated by the fact that the Federal Elections Commission assigns a unique identication to each committee, identication the interest group sector for these PACs is number facilitated by the fact that and the this Federal Elections Commission assigns unique identication number each committee, this identication number is also part a of the CRP dataset. Hence, Ito can classify every and PACs interest group Commission assigns a unique identication number to each committee, and this identication number is also part of the CRP dataset. Hence, I can classify every PACs interest group Commission assigns a unique identication numberIto each committee, and thisinterest identication number is also part of the CRP dataset. Hence, can classify every PACs group after 1990. number is also part of the CRP dataset. Hence, I can classify every PACs interest group after 1990. number is also part of the CRP dataset. Hence, I can classify every PACs interest group after 1990. To classify PAC contribution money before 1990, I proceed in two steps. First, I check after 1990. To classify PAC contribution money before 1990, I proceed in two steps. First, I check after 1990. To classify PAC contribution before 1990, I proceed two steps. First, I check whether the pre-1990 contributingmoney PAC survives after 1990. If so, in I assign the sector allocated To classify PAC contribution money before 1990, I proceed in two steps. First, I check whether the pre-1990 contributing PAC survives after 1990. If so, I assign the sector allocated To classify PAC contribution money before 1990, I proceed in two steps. First, I check whether the pre-1990 contributing PAC survives after 1990. If so, I assign the sector allocated by CRP to the respective PAC to the earlier period. Prior to 1990, there are nearly 41, 783 whether the pre-1990 contributing PAC survives after 1990. If so, I assign the sector allocated by CRP to the respective PAC to the earlier period. Prior to 1990, there are nearly 41, 783 whether the pre-1990 contributing PAC survives after 1990. If so, I assign the sector allocated by CRP to the respective PAC to the from earlier period. Prior to 1990, there are of nearly 41 , 783 PAC-candidate pairs in the sample, 3 , 232 PACs contributing to any the Senate by CRP to the respective PAC to the from earlier period. Prior to 1990, there are of nearly 41 , 783 PAC-candidate pairs in the sample, 3 , 232 PACs contributing to any the Senate by CRP to the respective PAC to the from earlier Prior to 1990, there are of nearly 41 , 783 PAC-candidate pairs in the sample, 3, period. 232 PACs contributing to any the Senate candidates before 1990. Of these PAC-candidate pairs, 38 , 712 (that is, 93% of the total PAC-candidate pairs in the sample, from 3, 232 PACs contributing to any of of the Senate candidates before 1990. Of these PAC-candidate pairs, 38 , 712 (that is, 93% the total PAC-candidate pairs in the sample, from 3, 232 PACs any of of the Senate candidates before 1990. Of these PAC-candidate pairs, contributing 38, 712PACs) (thatto is, the total PAC-candidate pairs, which correspond to 2, 625 contributing can93% be classied by candidates before 1990. Of these PAC-candidate pairs, 38 , 712 (that is, 93% of the total PAC-candidate pairs, which correspond to 2 , 625 contributing PACs) can be classied by candidates before 1990. Of these PAC-candidate pairs, 38, 712PACs) (that is, 93% of the total PAC-candidate pairs, which correspond to 2 , 625 contributing can be classied by this rst step alone. PAC-candidate pairs, which correspond to 2 , 625 contributing PACs) can be classied by this rst step alone. PAC-candidate pairs, which correspond to 2, 625 contributing PACs) can be classied by this rst step alone. for the set of unmatched PACs, I assigned the sector in the following way. First, thisNext, rst step alone. Next, for the set of unmatched PACs, I assigned the sector in the following way. First, thisNext, rst step alone. for the set of unmatched PACs, I assigned sector in PAC the following way. when possible I use the name of the sponsor of thethe unmatched to search for First, other Next, for the set of unmatched PACs, I assigned the sector in PAC the following way. when possible I use the name of the sponsor of the unmatched to search for First, other Next, for the set of unmatched PACs, I assigned the sector in the following way. First, when possible I use the name of the sponsor of the unmatched PAC to search for other matched committees sponsored by the same organization, and assign the corresponding when possible I use the name of the sponsor of the unmatched PAC to search for other matched committees sponsored by the same organization, and assign the corresponding when possible I use the name of the sponsor of the unmatched PAC to search for other matched committees sponsored by the same organization, and assign the corresponding industry. matched committees sponsored by the same organization, and assign the corresponding industry. matched committees sponsored by the same organization, and assign the corresponding industry. When this procedure does not permit the classication of the PAC, I follow the guidelines industry. When this procedure does not permit the classication of the PAC, I follow the guidelines industry. 29 the classication of the PAC, I follow the guidelines When this procedure does not permit identify the committees interest group. In some for When PAC classication from the CRP 29 to this procedure does not permit the classication of the PAC, I follow the guidelines identify the committees interest group. In some for When PAC classication from the CRP 29 to this procedure does not permit the classication of the PAC, I follow the guidelines for PAC classication from the CRP29 to identify the committees interest group. In some for29PAC classication from the CRP29 to identify the committees interest group. In some http://www.opensecrets.org/action/ftm/index.php for29PAC classication from the CRP to identify the committees interest group. In some
29 http://www.opensecrets.org/action/ftm/index.php 29 http://www.opensecrets.org/action/ftm/index.php 29 http://www.opensecrets.org/action/ftm/index.php

A.2 Campaign nance data industry sector classication A.2 Campaign nance data industry sector classication A.2 Campaign nance data industry sector classication A.2 Campaign nance data industry sector As noted in the main text, to classify the PAC contributions in industry categories, I follow A.2 Campaign nance data industry sector classication classication

Appendix Appendix Appendix Appendix

Table Table Table Table

sources sources sources sources

and and and and

denitions denitions denitions denitions

of of of of

the the the the

variables variables variables variables

used used used used

in in in in

the the the the

analysis. analysis. analysis. analysis.

http://www.opensecrets.org/action/ftm/index.php 35

35 35 35 35

37

cases, the PAC sponsor name does not provide enough clues to classify the interest group sector. Hence, while the classication is very complete, it is below 100%. More specically, of the 3, 071 PAC-candidate pairs that remained to be classied (corresponding to 607 unclassied PACs), 1, 930 PAC-candidate pairs are classied after these steps (corresponding to 372 PACs). Hence, overall I am able to classify 40, 642 PAC-candidate pairs (from 2, 997 PACs) before 1990. This amounts to 97% of the total number of PACs before 1990. Also importantly, the average contribution for unclassied PACs is smaller than the average. Thus, in terms of contribution money, the share of unclassied contributions is even smaller. For instance, in 1980, the year with least success in the classication, 94% of the contributions are classied, yet this represents 96% of the contribution money. By 1988, 98.4% of the contributions are classied, and this amounts to 99.4% of the contribution money.

36

38


Figure 1. The Denver-CO and Lincoln-NE Areas of Dominance Influence

Wyoming Nebraska Colorado

Note: Counties in the Denver-CO Area of Dominance Influence (for 1980) are shown hatched, and those in the Lincoln-NE Area of Dominance Influence (for 1980) are shaded.

39

Table 1. Summary Statistics Mean County-level variables 0.58 0.12 5,042 323,054 597.41 22.69 4.32 1.61 12.46 29.72 10.92 0.28 0 84 0.13 13.19 1.38 32.75 0 0 25.9 0 69,910 9,542,574 16539.57 84.4 34.7 57.44 86.81 100 95.5 1 0.13 0.92 20,597 90,655 146.28 46.72 14.33 50.62 6.63 38.59 69.18 0.84 9532 9532 9532 9532 9532 9532 9532 9532 9532 9532 9532 Standard Deviation Minimum Maximum Observations

Variable

Incumbent share of the two party vote

per capita personal income, 2000 US$ population population density (population per square mile) % under 18 % over 65 % female % black % urban % 12 years of more of school (persons 25+) ratio of in-state population for county's media market

40
5,185,977 3,076,456 1,036,460 312,506 0.25 0.13

Dummy=1 if incumbent is Democratic Herfindahl index PAC contributions -incumbent Herfindahl index PAC contributions -opponent Herfindahl index PAC contributions at industry level -incumbent Herfindahl index PAC contributions at industry level -opponent

Candidate-level variables 0.49 0.02 0.15 0.06 0.23 0.5 0.09 0.24 0.11 0.23 3,550,866 4,133,369 450,742 387,474 0.12 0.11 0.34 0.5

0 0 0 0.03 0.03 0 17,047 0 37 0 0 0 0

1 1 1 1 1 27,301,767 34,851,988 3,063,172 1,690,014 0.64 0.48 1 1

180 180 180 180 180 180 180 180 180 180 180 1759 542

Campaign receipts (2000 US $)-incumbent Campaign receipts (2000 US $)-opponent Receipts from special interest PACs (2000 US $)-incumbent Receipts from special interest PACs (2000 US $)-opponent Share of connected PACs money-incumbent Share of connected PACs money-opponent

County Media Market Classification In-state dummy (1 if in-state) 0.86 In-state and contiguous dummy (1 if in-state and adjacent to out-of-state) 0.56

Summary statistics are for the base sample in the analysis: excludes counties that were not dominated by the same state throughout the period and counties in overwhelmed states (an "overwhelmed" state has less than two-thirds of the population living in in-state dominated media markets). For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. For each county in state s and media market m, the ratio of in-state population refers to the proportion of the total media market population residing in state s. The in-state dummy equals 1 if the state the county is in has more than two-thirds (66%) of the population of that media market, and zero if it has less than one-third (33%). In the in-state and contiguous dummy, in addition to the latter conditions the in-state county must border an out-of-state county in the same state for the dummy to be equal to 1. See the text and Data Appendix for detailed variable definitions and sources.

Table 2. Media markets and electoral response to campaign finance concentration, US Senate 1980-2002 Dependent variable: share of the two-party vote received by the incumbent (1) (2) (3) (4) (5) (6) (7) (8) (9) State fixed effects herfindahl at PAC level herfindahl at interest group (industry) level (10) 0.00782 (0.00365) 0.171 (0.0221) 0.0775 (0.0165) -3.414 (0.503) -0.171 (0.0236) 0.0205 (0.0173) -1.465 (0.470) 9,527 0.490 9,532 0.293 9,532 0.302 -1.277 (0.457) -1.395 (0.448) -1.246 (0.456) 0.0242 (0.0174) 0.0333 (0.0168) 0.0281 (0.0168) 0.0108 (0.0154) 0.0242 (0.0195) -0.168 (0.0242) -0.15 (0.0230) -0.154 (0.0228) -0.129 (0.0208) -0.163 (0.0274) 0.763 (0.521) 0.692 (0.510) -0.169 (0.507) -3.533 (0.499) -0.156 (0.0285) 0.0259 (0.0196) -1.382 (0.471) 0.0728 (0.0167) 0.00175 (0.0162) 0.00745 (0.0162) 0.0267 (0.0148) 0.107 (0.0186) 0.104 (0.0186) 0.158 (0.0228) 0.116 (0.0214) 0.125 (0.0217) 0.128 (0.0207) 0.122 (0.0272) 0.108 (0.0283) 0.0871 (0.0245) 0.0114 (0.0184) 0.553 (0.512) -0.135 (0.0252) 0.0359 (0.0192) -1.15 (0.455) 0.00537 (0.00363) 0.00575 (0.00362) 0.00617 (0.00363) -0.024 (0.0105) 0.00738 (0.00364) 0.00504 (0.00364) 0.00552 (0.00362) 0.00595 (0.00363) 0.0978 (0.0249) 0.0181 (0.0184) 0.480 (0.501) -0.14 (0.0253) 0.0308 (0.0192) -1.271 (0.447)

in state dummy

-0.0254 (0.0106) 0.0982 (0.0239) 0.043 (0.0169) -0.330 (0.502) -0.113 (0.0236) 0.0101 (0.0176) -1.188 (0.456)

herfindahl incumbent

herfindahl opponent

herfindahl state

in state dummy x herfindahl incumbent

in state dummy x herfindahl opponent

41
9,532 0.286 9,532 0.294 9,532 0.363 9,527 0.369

in state dummy x herfindahl state

Campaign finance controls County controls Campaign finance controls x in-state dummy County controls x incumbent party

Observations R-squared

9,532 0.364

9,527 0.370

9,527 0.491

Standard errors, clustered by county, in parenthesis. All regressions include a full set of state and year fixed effects and interactions of the state and year fixed effects with a dummy that equals 1 if the incumbent is a Democrat. Campaign finance controls in columns 3-6 and 8-10 include the total campaign receipts, and total campaign receips from connected PACs, for the incumbent and his opponent (in 2000 US$ thousands of dollars). County controls in columns 4-5 and 9-10 include real per capita income (in 2000 US$ dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, percent urban, percent with a high 12 or more years of education (among people with 25 or more years of age), and state income growth. The in-state dummy equals 1 if the state the county is in has more than twothirds (66%) of the population of that media market, and zero if it has less than one-third (33%). For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. "Herfindahl state" corresponds to a similar index computed from the contribution of each of several economic sectors to the Gross State Product. Demeaned variables are shown in bold. I demean herfindahl variables before interacting them, so that the coefficient on the main effect for the in-state dummy shows the impact measured at the mean values of the herfindahl indices. See the text and Data Appendix for detailed variable definitions and sources.

Table 3. Media markets and electoral response to campaign finance concentration, US Senate 1980-2002 Dependent variable: share of the two-party vote received by the incumbent (1) (2) (3) (4) (5) (6) (7) (8) (9) County fixed effects herfindahl at PAC level herfindahl at interest group (industry) level 0.113 (0.0214) 0.0859 (0.0206) -3.864 (1.011) -0.0995 (0.0222) 0.0116 (0.0212) -0.942 (1.024) 9,527 0.843 9,532 0.755 9,532 0.763 -0.0542 (1.061) -0.0607 (1.027) 0.682 (1.154) 0.0126 (0.0212) 0.033 (0.0186) 0.0302 (0.0188) 0.00772 (0.0185) 0.000863 (0.0241) -0.0939 (0.0235) -0.0721 (0.0185) -0.065 (0.0182) -0.048 (0.0184) -0.0997 (0.0330) -0.378 (1.087) -0.736 (1.056) -1.956 (1.162) -3.652 (1.021) -0.0899 (0.0336) 0.00116 (0.0241) -1.221 (1.033) 0.0835 (0.0206) 0.00115 (0.0182) 0.00324 (0.0185) 0.0281 (0.0175) 0.128 (0.0232) 0.126 (0.0233) 0.0206 (0.0212) -0.399 (1.089) -0.065 (0.0246) 0.0246 (0.0217) -0.138 (1.063) 0.098 (0.0228) 0.0532 (0.0189) 0.0513 (0.0191) 0.0687 (0.0201) 0.0661 (0.0331) 0.0495 (0.0339) 0.0261 (0.0251) 0.0269 (0.0248) 0.0233 (0.0216) -0.765 (1.053) -0.0592 (0.0239) 0.0222 (0.0219) -0.146 (1.026)

(10)

herfindahl incumbent

0.0449 (0.0257) 0.0567 (0.0214) -2.01 (1.148) -0.0456 (0.0243) -0.00792 (0.0227) 0.602 (1.142)

herfindahl opponent

herfindahl state

in state dummy x herfindahl incumbent

in state dummy x herfindahl opponent

42
9,532 0.748 9,532 0.756 9,532 0.824 9,527 0.827

in state dummy x herfindahl state

Campaign finance controls County controls Campaign finance controls x in-state dummy County controls x incumbent party

Observations R-squared

9,532 0.825

9,527 0.828

9,527 0.844

Standard errors, clustered by county, in parenthesis. All regressions include a full set of state and year fixed effects and interactions of the county and year fixed effects with the party of the incumbent. Campaign finance controls in columns 3-6 and 8-10 include the total campaign receipts, and total campaign receips from connected PACs, for the incumbent and his opponent (in 2000 US$ thousands of dollars). County controls in columns 4-5 and 9-10 include real per capita income (in 2000 US$ dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, percent urban, percent with a high 12 or more years of education (among people with 25 or more years of age), and state income growth. The in-state dummy equals 1 if the state the county is in has more than two-thirds (66%) of the population of that media market, and zero if it has less than one-third (33%). For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. "Herfindahl state" corresponds to a similar index computed from the contribution of each of several economic sectors to the Gross State Product. See the text and Data Appendix for detailed variable definitions and sources.

Table 4. In-state versus out of state counties, differences in observable characteristics (1) (2) (3) (4) (5) (6) (7) (8) population 55,533 (11,584) 51,014 (8,626) 9,528 0.095 12.35 (7.880) 49.26 (7.854) 2,623 0.310 2,623 0.984 72.46 (0.266) -0.00876 (0.119) -0.0438 (0.300) 15.61 (0.241) 2,623 0.337 9,528 0.978 9,528 0.263 9,528 0.213 0.118 (0.112) 50.73 (0.0933) 2,623 0.357 76.01 (15.71) 25.65 (0.127) 15.66 (0.223) 50.26 (0.0884) 98.23 (18.56) 0.162 (0.0764) -1.158 (0.228) 0.0421 (0.0886) 1.855 (0.703) 5.201 (0.656) 9,528 0.558 1.401 (0.886) 5.506 (0.606) 2,623 0.482 population density % under 18 % over 65 % female % black

Real per Dependent variable: capita income Panel A: in-state versus out of state 1,704 (212.8) 22,877 (219.0)

% urban 10.6 (1.810) 31.69 (1.713) 9,527 0.151 3.002 (2.214) 27.89 (1.717) 2,623 0.154

(9) % with 12+ years of school (people 25+) 2.873 (0.423) 69.2 (0.387) 9,527 0.588 1.233 (0.535) 69.47 (0.371) 2,623 0.672

in state dummy

Constant

43
782.8 (236.7) 19,686 (212.0) 2,623 0.551 2,623 0.108 23,169 (9,231) 12,529 (4,773)

Observations 9,528 9,528 R-squared 0.390 0.106 Panel B: in-state and contiguous versus out of state

in state and contiguous dummy

Constant

Standard errors, clustered by county, in parenthesis. All regressions include a full set of state and year fixed effects. The in-state dummy equals 1 if the state the county is in has more than two-thirds (66%) of the population of that media market, and zero if it has less than one-third (33%). In the in state and contiguous dummy, in addition to the latter conditions the in-state county must border an out-of-state county in the same state for the dummy to be equal to 1.

Observations R-squared

Table 5. Media markets and electoral response to campaign finance concentration, US Senate 1980-2002 In-state contiguous counties vs out-of-state counties only Dependent variable: share of the two-party vote received by the incumbent (1) (2) (3) (4) (5) (6) (7) (8) (9) herfindahl at PAC level herfindahl at interest group (industry) level 0.0645 (0.0254) 0.0955 (0.0216) -5.511 (1.067) -0.0769 (0.0309) 0.0231 (0.0285) -1.885 (1.196) 2,623 0.844 2,623 0.771 2,623 0.782 -0.812 (1.084) -0.838 (1.052) -0.229 (1.067) 0.0251 (0.0273) 0.0344 (0.0270) 0.0323 (0.0265) 0.0286 (0.0260) 0.0125 (0.0324) -0.0625 (0.0310) -0.0657 (0.0241) -0.0573 (0.0233) -0.0414 (0.0246) -0.0752 (0.0371) -0.0624 (0.0379) 0.0111 (0.0311) -1.867 (1.199) -2.32 (1.031) -2.606 (1.002) -3.151 (1.042) -5.176 (1.085) 0.086 (0.0212) 0.0282 (0.0205) 0.0307 (0.0206) 0.0356 (0.0183) 0.13 (0.0243) 0.122 (0.0239) 0.0484 (0.0235) -2.266 (1.042) -0.0582 (0.0284) 0.0240 (0.0307) -0.788 (1.079) 0.0702 (0.0267) 0.0319 (0.0254) 0.0264 (0.0270) 0.0404 (0.0292) 0.0153 (0.0346) 0.0232 (0.0356) -0.00126 (0.0300) -0.00436 (0.0305) 0.0521 (0.0237) -2.587 (1.006) -0.0499 (0.0272) 0.0228 (0.0301) -0.812 (1.047)

(10) 0.0128 (0.0327) 0.0621 (0.0217) -3.216 (1.037) -0.0400 (0.0287) 0.0155 (0.0310) -0.159 (1.053)

herfindahl incumbent

herfindahl opponent

herfindahl state

in state dummy x herfindahl incumbent

in state dummy x herfindahl opponent

44
2,623 0.764 2,623 0.777 2,623 0.825 2,623 0.831

in state dummy x herfindahl state

Campaign finance controls County controls Campaign finance controls x in-state dummy County controls x incumbent party

Observations R-squared

2,623 0.827

2,623 0.833

2,623 0.845

Standard errors, clustered by county, in parenthesis. All regressions include a full set of county and year fixed effects and interactions of the county and year fixed effects with the party of the incumbent. Campaign finance controls in columns 3-4 and 7-8 include the total campaign receipts, and total campaign receips from connected PACs, for the incumbent and his opponent (in 2000 US$ thousands of dollars). County controls in columns 4 and 8 include real per capita income (in 2000 US$ dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, percent urban, percent with a high 12 or more years of education (among people with 25 or more years of age), and state income growth. Like in Table 2, the in-state dummy equals 1 if the state the county is in has more than two-thirds (66%) of the population of that media market, and zero if it has less than one-third (33%). However, in addition to the latter conditions the in-state county must border an out-of-state county in the same state for the dummy to be equal to 1. For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. "Herfindahl state" corresponds to a similar index computed from the contribution of each of several economic sectors to the Gross State Product. See the text and Data Appendix for detailed data definitions and sources.

Table 6. Media markets and electoral response to campaign finance concentration, US Senate 1980-2002 Placebo out of state vs in-state contiguous counties only Dependent variable: share of the two-party vote received by the incumbent (1) (2) (3) (4) (5) (6) (7) (8) herfindahl at PAC level herfindahl at interest group (industry) level -0.0107 (0.0337) 0.103 (0.0160) -5.913 (0.828) 0.00834 (0.0387) 0.0153 (0.0238) 0.541 (1.126) 2,505 0.734 2,505 0.744 2,505 0.807 2,505 0.811 2,505 0.741 2,505 0.750 -0.191 (0.994) -0.553 (1.011) 0.0133 (0.0224) 0.00524 (0.0232) 0.00524 (0.0231) 0.00847 (0.0266) 0.00376 (0.0391) 0.0147 (0.0281) 0.0135 (0.0268) 0.0150 (0.0405) -1.437 (0.720) -1.84 (0.728) -5.667 (0.830) 0.00866 (0.0417) 0.00726 (0.0249) 0.642 (1.123) 0.0999 (0.0152) 0.0415 (0.0171) 0.0423 (0.0176) 0.127 (0.0184) 0.124 (0.0176) 0.00474 (0.0344) -0.0344 (0.0297) -0.0369 (0.0305) -0.0671 (0.0363) -0.0494 (0.0372) -0.0673 (0.0292) 0.048 (0.0196) -1.375 (0.739) 0.0204 (0.0281) 0.00514 (0.0258) -0.155 (0.992) 2,505 0.808 -0.0667 (0.0288) 0.0499 (0.0201) -1.811 (0.742) 0.0185 (0.0262) 0.00472 (0.0258) -0.527 (1.009) 2,505 0.812

herfindahl incumbent

herfindahl opponent

herfindahl state

in state dummy x herfindahl incumbent

in state dummy x herfindahl opponent

45

in state dummy x herfindahl state

Campaign finance controls County controls

Observations R-squared

Standard errors, clustered by county, in parenthesis. All regressions include a full set of state and year fixed effects, as well as interactions of the state fixed effects with the party of the incumbent. Campaign finance controls in columns 3-4 and 7-8 include the total campaign receipts, and total campaign receips from connected PACs, for the incumbent and his opponent (in 2000 US$ thousands of dollars). County controls in columns 4 and 8 include real per capita income (in 2000 US$ dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, percent urban, and percent with a high 12 or more years of education (among people with 25 or more years of age). The in-state dummy equals 0 for in-state and contiguos counties as defined in Table 4, and 1 for their in-state neighbors. For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a herfindahl of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such herfindahls. "Herfindahl state" corresponds to a similar index computed from the contribution of each of several economic sectors to the Gross State Product. See the text and Data Appendix for detailed variable definitions and sources.

Table 7. Media markets and electoral response to share of connected-PACs money, US Senate 1980-2002 Dependent variable: share of the two-party vote received by the incumbent (1) (2) (3) (4) (5) (6) County fixed effects, instate and contiguous only

State fixed effects in state dummy share incumbent share opponent in state dummy x share incumbent in state dummy x share opponent State herfindahl Campaign finance controls County controls Observations R-squared 9,532 0.561 0.00433 (0.00422) -0.00219 (0.0186) -0.0253 (0.00846) -0.00929 (0.0203) 0.02 (0.00928) 0.0121 (0.00307) -0.00674 (0.0158) -0.00699 (0.00552) 0.0101 (0.0165) 0.0172 (0.00580) 9,527 0.749

County fixed effects

-0.0162 (0.0149) -0.0112 (0.00634) 0.00803 (0.0158) 0.00389 (0.00656)

-0.0116 (0.0153) 0.00173 (0.00609) 0.00550 (0.0159) 0.00247 (0.00631)

-0.0200 (0.0156) -0.00958 (0.00659) -0.00869 (0.0185) -0.0104 (0.00819)

-0.0169 (0.0165) 0.00386 (0.00677) -0.00574 (0.0182) -0.0143 (0.00811)

9,532 0.923

9,527 0.931

2,623 0.919

2,623 0.926

Standard errors, clustered by county, in parenthesis. In columns 1-2, regressions include a full set of state x party and year x party fixed effects. In columns 3-6, regressions include a full set of county x party and year x party fixed effects. In columns 5 and 6 only contiguous counties are included. In even columns, campaign finance controls include total campaign receipts, and total campaign receipts from connected PACs, for the incumbent and his opponent (in 2000 US$ thousands of dollars); county controls include real per capita income (in 2000 US$ dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, percent urban, percent with a high 12 or more years of education (among people with 25 or more years of age), and state income growth; state herfindahl corresponds to a herfindahl index computed from the contribution of each of several economic sectors to the Gross State Product and the interaction of this measure with the in-state dummy. The in-state dummy equals 1 if the state the county is in has more than two-thirds (66%) of the population of that media market, and zero if it has less than one-third (33%). Share incumbent and share opponent refer to the total share of connected-PACs money in campaign receipts. In columns 1 and 2, these shares are demeaned before interacting them with the in-state dummy. See the text and Data Appendix for detailed data definitions and sources.

46

Table 8. Concentration indices and news about campaign contributions Dependent Variable: herfindahl (PAC level) herfindahl (industry level) (1) (2) (3) (4) (5) (6) 0.993 (0.161) -0.000154 (0.000138) 175 0.192 175 0.324 175 0.404 175 0.309 -0.000253 (0.000127) 0.887 (0.155)

total PAC contributions (7) (8) (9) -205,697 (144,480)

PAC stories/Total stories

Total number of candidate stories 175 0.322

0.984 (0.159) -9.23e-05 (0.000119)

0.869 (0.156) -0.000198 (0.000112) 175 0.411

1,245 (419.2) 175 0.521 175 0.561

-91,968 (158,422) 1,239 (423.0) 175 0.561

47

Observations R-squared

Robust standard errors in parenthesis. All specifications include a full set of year fixed effects, a dummy variable that equals 1 if the candidate is Democratic and a dummy variable that equals one if the canidate is the incumbent. Total number of stories refer to the number of stories in which keywords identifying the candidate name and the word "senate" or "senator" appears in a story. Total number of PAC stories are those in which, in addition, the story contains the keywords "PAC" and "money". The search window is two months before each election. Appendix Table 3 shows the sample of states included each year (and the number of newspapers on which the automated search was done), as well as basic summary statistics for total stories, PAC stories, and the ratio.

(1) state and year 0.00473 (0.00415) 0.0168 (0.00534) 0.017 (0.00537) 0.0205 (0.00413) -0.0123 (0.00503) 0.00461 (0.00417) 0.0155 (0.00361) state x party year x party state x party year x party state and year state x party year x party state x party year x party state and year

Table 9. Media markets and turnout, US Senate 1980-2002 Dependent variable: ratio of total votes in senate race to population (2) (3) (4) (5) (6) (7) (8)

(9) state x party year x party state x party year x party

Fixed effects:

in state dummy

in state and contiguous dummy

48

placebo in state and contiguous dummy

-0.0123 (0.00506)

-0.00207 (0.00389)

Population control

Observations 10,331 10,331 10,331 2,782 2,782 2,782 2,675 2,675 2,675 R-squared 0.546 0.564 0.650 0.520 0.534 0.689 0.554 0.570 0.707 Standard errors, clustered by county, in parenthesis. In columns 1-3, the in-state dummy equals 1 if the state the county is in has more than two-thirds (66%) of the population of that media market, and zero if it has less than one-third (33%). In columns 4-6, to the latter conditions the in-state county must border an out-ofstate county in the same state for the dummy to be equal to 1. In columns 7-9 the placebo in-state dummy equals 0 for in-state and contiguos counties as defined in columns 4-6, and 1 for their in-state neighbors. Columns 3, 6 and 9 include total population in the right hand side. The fixed effects included in each regression are noted in the title of each column.

Table 10. Media markets and turnout response to campaign finance concentration, US Senate 1980-2002 Dependent variable: ratio of total votes in senate race to population (1) (2) (3) (4) (5)

(6)

State fixed effects in state dummy herfindahl incumbent herfindahl opponent in state dummy x herfindahl incumbent in state dummy x herfindahl opponent State herfindahl Campaign finance controls County controls Observations R-squared 9,532 0.561 0.00433 (0.00422) -0.00219 (0.0186) -0.0253 (0.00846) -0.00929 (0.0203) 0.02 (0.00928) 0.0121 (0.00307) -0.00674 (0.0158) -0.00699 (0.00552) 0.0101 (0.0165) 0.0172 (0.00580) 9,527 0.749

County fixed effects

County fixed effects, instate and contiguous only

-0.0162 (0.0149) -0.0112 (0.00634) 0.00803 (0.0158) 0.00389 (0.00656)

-0.0116 (0.0153) 0.00173 (0.00609) 0.00550 (0.0159) 0.00247 (0.00631)

-0.0200 (0.0156) -0.00958 (0.00659) -0.00869 (0.0185) -0.0104 (0.00819)

-0.0169 (0.0165) 0.00386 (0.00677) -0.00574 (0.0182) -0.0143 (0.00811)

9,532 0.923

9,527 0.931

2,623 0.919

2,623 0.926

Standard errors, clustered by county, in parenthesis. In columns 1-2, regressions include a full set of state x party and year x party fixed effects. In columns 3-6, regressions include a full set of county x party and year x party fixed effects. In columns 5 and 6 only contiguous counties are included. In even columns, campaign finance controls include total campaign receipts, and total campaign receipts from connected PACs, for the incumbent and his opponent (in 2000 US$ thousands of dollars); county controls include real per capita income (in 2000 US$ dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, percent urban, percent with a high 12 or more years of education (among people with 25 or more years of age), and state income growth; state herfindahl corresponds to a herfindahl index computed from the contribution of each of several economic sectors to the Gross State Product and the interaction of this measure with the in-state dummy. The in-state dummy equals 1 if the state the county is in has more than two-thirds (66%) of the population of that media market, and zero if it has less than one-third (33%). For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. In columns 1 and 2, these indices are demeaned before interacting them with the in-state dummy. See the text and Data Appendix for detailed data definitions and sources.

49

Table 11. Spillovers: Media markets and response of House and Presidential vote to Senate race campaign finance concentration, 1980-2002 (1) (2) (3) (4) (5) (6) Dependent variable: share of two-party vote to presidential candidate of incumbent's party Herfindahl at level: PAC Industry herfindahl incumbent herfidnahl opponent in state dummy x herfindahl incumbent in state dummy x herfindahl opponent Control for house candidate herfindahl Observations R-squared 1,324 0.964 1,324 0.964 1,321 0.849 1,321 0.841 -0.240 (0.195) -0.00422 (0.0246) 0.190 (0.149) 0.0102 (0.0239) -0.457 (0.249) -0.0151 (0.0266) 0.234 (0.169) 0.00513 (0.0267)

share of two-party vote to house candidate of incumbent's party PAC Industry PAC Industry -3.466 (0.774) -0.179 (0.0640) -0.594 (0.316) 0.128 (0.0801) -1.845 (0.995) -0.141 (0.0683) -0.595 (0.445) 0.0992 (0.0787) -2.538 (0.738) -0.0621 (0.0483) -0.210 (0.198) 0.0622 (0.0527) 1,169 0.899 -0.0227 (1.394) -0.0289 (0.0550) -0.110 (0.316) 0.0349 (0.0559) 1,169 0.902

Standard errors, clustered by county, in parenthesis. All regressions include a full set of county x party and year x party fixed effects. Campaign finance controls, county controls, and state herfindahl controls as described in Table 2 are included in all regressions. The in-state dummy is limited to contiguous counties and equals 1 if the state the county is in has more than two-thirds (66%) of the population of that media market, and zero if it has less than one-third (33%). For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. Note that these measures as well as the campaign finance controls in the right hand side of the regression refer to the senate candidates, whereas the dependent variable is for house candidates and presidential candidates as described in the column titles. See the text and Data Appendix for detailed data definitions and sources.

50

Table 12. Spillovers: Out-of-state counties response to neighboring state campaign finance profiles in Senate races, 1980-2002 (1) (2) (3) (4) (5) (6) Dependent variable is incumbent share of the two-party vote Industry Industry Industry level level level PAC level PAC level PAC level -0.0224 (0.130) 0.0393 (0.0516) -0.00475 (0.0575) -0.0125 (0.0395) -0.0182 (0.0386) -0.00282 (0.0503) 0.0433 (0.0315) 551 0.901 551 0.902 574 0.902 574 0.902 551 0.902 0.00276 (0.0555) 0.0403 (0.0316) 551 0.902 0.0240 (0.0644) 0.0559 (0.0591) -0.0687 (0.0982)

herfindahl incumbent

herfindahl opponent

herfindahl for neighboring state, incumbent's party

herfindahl for neighboring state, opponent's party

51

Difference in herfindahls (incumbentincumbent's party in neighboring state)

Difference in herfindahls (opponentopponent's party in neighboring state)

Observations R-squared

Standard errors, clustered by county, in parenthesis. All regressions include a full set of county x party and year x partyfixed effects. All regressions include campaign finance controls, county controls, and the state herfindahl as defined in Table 2. Only out-of-state counties are included, that is if the state the county is in has less than one-third (33%) of the population of the media market. For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. See the text and Data Appendix for detailed data definitions and sources.

Table 13. Media markets and electoral response to campaign finance concentration, US Senate 1980-2002 Robustness to media market definition Dependent variable: share of the two-party vote received by the incumbent

herfindahl at level

In-state= if the state the county is in has more than x% of the population of that media market; out-of-state: if it has less than (100-x)%. x=50 x=60 x=70 (1) (2) (3) (4) (5) (6) PAC industry PAC industry PAC industry

PANEL A: in state versus out of state -0.0653 (0.0172) -0.00204 (0.0156) 12,922 0.836 12,922 0.837 11,946 0.833 11,946 0.835 -0.00891 (0.0172) 0.0148 (0.0168) 0.0108 (0.0181) -0.0619 (0.0227) -0.0688 (0.0175) -0.0652 (0.0237) -0.0827 (0.0190) 0.0184 (0.0220) 8,379 0.835 -0.0786 (0.0262) 0.00127 (0.0243) 8,379 0.836

in state dummy x herfindahl incumbent

in state dummy x herfindahl opponent

State Herfindahl, Campaign finance, and county controls

Observations R-squared

52
-0.064 (0.0241) 0.0165 (0.0194) 3,999 0.841 3,999 0.843 0.0116 (0.0210) -0.0628 (0.0287) -0.0627 (0.0238) 0.0356 (0.0226) 3,580 0.840

PANEL B: in state and contiguous versus out of state -0.0615 (0.0285) 0.0309 (0.0235) 3,580 0.843 -0.0697 (0.0256) 0.0160 (0.0313) 2,084 0.840 -0.0644 (0.0302) -0.000127 (0.0335) 2,084 0.842

in state dummy x herfindahl incumbent

in state dummy x herfindahl opponent

State Herfindahl, Campaign finance, and county controls

Observations R-squared

Standard errors, clustered by county, in parenthesis. All regressions include a full set of county x party and year x partyfixed effects . Campaign finance controls include the total campaign receipts, and total campaign receips from connected PACs, for the incumbent and his opponent (in 2000 US$ thousands of dollars). County controls include real per capita income (in 2000 US$ dollars), population, population density, percent of the population under 18, percent of the population over 65, percent female, percent black, percent urban, percent with a high 12 or more years of education (among people with 25 or more years of age), and state income growth. The in-state dummy is defined as described in each column title. However, in Panel B in addition to the stated conditions the in-state county must border an out-of-state county in the same state for the dummy to be equal to 1. For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. State herfindahl corresponds to a similar index computed from the contribution of each of several economic sectors to the Gross State Product. The direct effects of the herfindahl of the incumbent and the opponent, as well as the direct effect of the state herfindahl and its interaction with the in state dummy, are always included. See the text and Data Appendix for detailed data definitions and sources.

Table 14. Media markets and electoral response to campaign finance concentration, US Senate 1980-2002 Robustness to Media Market Definition Dependent variable: share of the two-party vote received by the incumbent (1) (2) (3) (4) (5)

(6)

Panel A: herfindahl at PAC level ratio in state herfindahl incumbent herfindahl opponent ratio in state x herfindahl incumbent ratio in state x herfindahl opponent Campaign finance, county, and state herfindahl controls Observations R-squared Panel B: herfindahl at interest group (industry) level ratio in state herfindahl incumbent herfindahl opponent ratio in state x herfindahl incumbent ratio in state x herfindahl opponent Campaign finance, county, and state herfindahl controls Observations R-squared

State fixed effects 0.00625 (0.00416) 0.0938 (0.0136) 0.0952 (0.00880) -0.202 (0.0269) -0.00278 (0.0193) 0.00365 (0.00411) 0.0535 (0.0132) 0.0282 (0.00875) -0.166 (0.0258) 0.00986 (0.0185) 10,305 0.279 0.00580 (0.00415) 0.0498 (0.0158) 0.129 (0.00992) -0.18 (0.0284) -0.00234 (0.0219) 10,300 0.363 0.00339 (0.00411) 0.0328 (0.0144) 0.0402 (0.0100) -0.141 (0.0267) 0.0126 (0.0213) 10,305 0.287 10,300 0.363

County fixed effects 0.183 (0.0457) 0.0688 (0.0139) 0.0985 (0.0104) -0.134 (0.0245) -0.0117 (0.0209) 0.0882 (0.0339) 0.0232 (0.0129) 0.0275 (0.00995) -0.0816 (0.0207) 0.00684 (0.0185) 10,305 0.748 0.193 (0.0456) 0.0195 (0.0199) 0.138 (0.0120) -0.113 (0.0352) -0.0286 (0.0247) 10,300 0.829 0.0952 (0.0345) -0.00129 (0.0157) 0.0435 (0.0117) -0.0619 (0.0268) -0.00312 (0.0222) 10,305 0.755 10,300 0.830

County fixed effects, instate and contiguous only -0.0364 (0.118) 0.0314 (0.0225) 0.106 (0.0160) -0.103 (0.0365) 0.0220 (0.0340) 0.0318 (0.0901) 0.00198 (0.0250) 0.0455 (0.0164) -0.0798 (0.0277) 0.0347 (0.0319) 2,623 0.765 -0.0224 (0.122) -0.0147 (0.0262) 0.136 (0.0179) -0.0986 (0.0424) 0.00846 (0.0385) 2,623 0.831 0.0393 (0.0929) -0.0243 (0.0256) 0.0626 (0.0187) -0.067 (0.0326) 0.0237 (0.0362) 2,623 0.771 2,623 0.832

Standard errors, clustered by county, in parenthesis. Regressions in columns 1-2 include a full set of state x party and year x party fixed effects, and in columns 3-6 a full set of county x party and year x party fixed effects. Campaign finance, county controls and state herfindahl controls are as defined in Table 2. In columns 5 and 6, only the set of contiguous in-state and out-of-state counties as defined in Table 3 are included in the regression. For each county in state s and media market m, the variable ratio in state refers to the share of the media market population that is in state s. For the herfindahl at the PAC (industry) level, the contribution from each PAC (industry) is expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. The ratio in state and herfindahl indices are demeaned before interacting. State herfindahl corresponds to a similar index computed from the contribution of each of several economic sectors to the Gross State Product. See the text and Data Appendix for detailed data definitions and sources.

53

Variable

Appendix Table A.1: Main variables sources and definitions Description Electoral results

Source

Incumbent share of the two-party vote

Total votes obtained by the incumbent senator divided by total votes for the Republican and Democratic candidate, 1980-2002, calculated at the county level. Open seats and unchallenged races are dropped from the analysis. Total votes obtained by the presidential candidate of the same party as the incumbent senator divided by total votes for the Republican and Democratic presidential candidates, 1980-2002, calculated at the county level. For the sample of senate races included in the analysis and for the years in which there is a presidential election.

Share of two-party vote to presidential candidate of incumbent's party

Share of two-party vote to house candidate of incumbent's party

Total votes obtained by the Congress (House) candidate of the same party as the incumbent senator divided by total votes for the Republican and Democratic House candidates, 1980-2002, calculated at the county level. For counties with multiple congressional districts the votes for all the Democratic and Republican candidates running for Congress in a county are added to find the Democratic and Republican vote share per county.

Election data and incumbency status is constructed from a variety of sources by Ansolabehere and Snyder (2002). These data run through 2000 and were updated from similar sources and kindly shared by the authors. The main original sources are ICPSR study number 13 (General Election Data for the United States, 1950-1990), and America Votes (1992, 1994, 1996, 1998, and 2000). The House election variables are available from Snyder and Strmberg's (2010) replication dataset.

Campaign contributions Total campaign receipts (2000 US $) Total campaign contributions from any source to the given candidate during the two-year electoral cycle preceeding the election, deflated by the consumer price index to express in constant 2000 US dollars. Sources of funds include connected PACs, ideological/single-issue PACs, individuals, party committess and other candidates. Total contributions from connected PACs to the candidate during the two-year electoral cycle preceeding the election, deflated by the consumer price index to express in constant 2000 US dollars. Connected PACs include all non-party committees classified by the FEC as sponsored by a corporation, labor organization, membership organization, trade association, cooperative, or corporation without capital stock. They exclude ideological and single-issue PACs, as well as party committes and contributions from other candidates. Federal Elections Commission. Committee master files, candidate master files, and itemized contributions files. The itemized committee contributions file containing each PAC contribution is matched to the other files using the ID number of the contributing committee and the ID number of the recipient (see http://www.fec.gov/finance/disclosur e/ftpdet.shtml). Total campaign receipts and total from connected PACs are available from the candidate's master files, whereas for the herfindahl indices the itemized contribution files are necessary. For industry classification, the source is Center for Responsive Politics and author's adjustments, as explained in Appendix section 2.

Receipts from connected (special interest) PACs (2000 US $)

Herfindahl index of PAC contributions (PAC level)

Financial contributions (24K) and in-kind contributions (24Z) from each PAC are expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares. Contributions are for the two-year electoral cycle preceeding the election. Financial contributions (24K) and in-kind contributions (24Z) from each PAC are classified into a single industry as explained in the main text. Industry contributions are expressed as a share of the total PAC contributions to the candidate, and the index corresponds to the sum of the squares of such shares and are for the two-year electoral cycle preceeding the election.

Herfindahl index of PAC contributions (industry level)

54

Variable

Appendix Table A.1 (continued): Variable Sources and Definitions Description Source Media Markets

In-state dummy

Dummy variable equal to 1 if the county is "in-state" and zero if it is out-of-state. Each county in the sample is assigned to one TV market, or Area of Dominance Influence. A county is "in-state" if the state the county is in has more than x% of the population of the media market, and "out-of-state" if the state the county is in has less than (100-x)% of such population, for x greater than or equal to 50. Unless otherwise stated, the analysis in the paper uses x=66. Only counties that were dominated by the same state are considered in the baseline analysis in the paper. Identical to the in-state dummy except that, in addition to the latter conditions, the in-state county must border an out-of-state county in the same state for the variable to be equal to 1. That is, the instate and contiguous dummy is like the in-state dummy but excludes from the analysis all in-state counties that are not contiguous to other out-of-state counties in the same state.

Data and media market definitions from Ansolabehere, Snowberg and Snyder (2006). Areas of Dominance Influence, constructed by Arbitron, are originally from Broadcast and Cable in 1980, 1990, and 2000.

In-state and contiguous dummy

Ansolabehere, Snowberg and Snyder (2006); Broadcast and Cable in 1980, 1990, and 2000.

Placebo in-state and contiguous dummy Overwhelmed states

Dummy variable equal to zero for the in-state and contiguous counties as defined by the in-state and continuous dummy, and one for neighboring in-state counties in the same state. States-years where less than two-thirds of the population lived in in-state dominated media markets, as defined above. Counties in these states are excluded from the analysis. For each county c in state s and media market (Area of Dominance Influence) m, ratio of the population of media market m residing in state s to the total population of market m. Other variables and controls

Ansolabehere, Snowberg and Snyder (2006); Broadcast and Cable in 1980, 1990, and 2000. Ansolabehere, Snowberg and Snyder (2006); Broadcast and Cable in 1980, 1990, and 2000. Ansolabehere, Snowberg and Snyder (2006); Broadcast and Cable 1980, 1990, and 2000.

Ratio in state

Real per capita personal income

County-level per capita personal income in constant prices (2,000 US$). Nominal figures from Bureau of Economic Analysis deflated with the Consumer Price Index. County population, and county population per square mile. Percent of the population in each category at the county level. Except in the case of percent of population with 12 or more years of school which is computed with respect to total county population of 25 or more years of age, percents are with respect to total population. Yearly percentage change in real state per capita personal income. Real state per capita personal income is the nominal value reported by the BEA deflated by Consumer Price Index. Sum of the squares of the contribution shares of each economic sector to the total Gross State Product.

Regional Economic Information System, Bureau of Economic Analysis, U.S. Department of Commerce US Census Bureau US Census Bureau

Population and population density % under 18, % over 65, % female, % black, % 12 years or more of school (persons 25+), % urban

State-level growth

Regional Economic Information System, Bureau of Economic Analysis, U.S. Department of Commerce Regional Economic Information System, Bureau of Economic Analysis, U.S. Department of Commerce Own calculation based on automated news search on NewsLibrary.com. Search limited for newspapers of the state where the candidate runs for office.

State herfindahl

Total Stories and PAC stories

Total stories=number of news stories in the two months leading to an election appearing containing the candidate's name and the words "senate" or "senator." PAC stories=of the total stories for a candidate in a given election, number which include, in addition, the keywords "PAC" and "money."

55

Sector

Agribusiness

Communications/Electronics

Construction

Defense

Energy & Natural Resources

Appendix Table A.2: Industry categories for special-interest PACs Examples of categories Farm bureaus, Florists & Nursery Services, Grain traders & terminals, Veterinarians, Agricultural Services/Products Animal feed & health products, Agricultural services & related industries, Farm organizations & cooperatives, Agricultural chemicals (fertilizers & pesticides) Cotton, Sugar cane & sugar beets, Other commodities (incl rice, peanuts, honey), Crop Production & Basic Processing Vegetables, fruits and tree nut, Crop production & basic processing, Wheat, corn, soybeans and cash grain Dairy Milk & dairy producers Food and kindred products manufacturing, Food wholesalers, Food stores, Food & Food Processing & Sales Beverage Products and Services, Meat processing & products Forestry & Forest Products Paper & pulp mills and paper manufacturing, Forestry & Forest Products Horse breeders, Livestock, Feedlots & related livestock services, Sheep and Wool Livestock Producers Misc Agriculture Agriculture Poultry & Eggs Poultry & eggs Tobacco Tobacco & Tobacco products Computer manufacture & services, Computer software, Computers, components & Computers/Internet accessories, Online computer services, Data processing & computer services Electronics Mfg & Services Electronics manufacturing & services Misc Communications/Electronics Communications & Electronics Greeting card publishing, Book, newspaper & periodical publishing, Printing and Printing & Publishing publishing (printed & online), Commercial printing & typesetting Live theater & other entertainment productions, Recorded Music & music production, TV production & distribution, Cable & satellite TV production & distribution, Movie TV/Movies/Music Theaters, Entertainment Industry/Broadcast & Motion Pictures, Motion Picture Cellular systems and equipment, Other Communications Services, Telecom Services & Equipment Telecommunications, Telephone & communications equipment, Satellite communications Telephone Utilities Telephone utilities, Long-distance telephone & telegraph service Electrical Supply, Stone, clay, glass & concrete products, Construction equipment, Building Materials & Equipment Plumbing & pipe products, Building materials, Lumber and wood products, Other construction-related products Architectural services, Surveying, Engineers - type unknown, Engineering, architecture Construction Services & construction mgmt svcs Dredging Contractors, Construction, unclassified, Construction & Public Works, Public General Contractors works, industrial & commercial construction, Builders associations Home Builders Residential construction, Mobile home construction Landscaping & Excavation Svcs, Plumbing, heating & air conditioning, Special trade Special Trade Contractors contractors, Electrical contractors Defense Aerospace Defense aerospace contractors Defense Electronics Defense electronic contractors Homeland Security contractors, Defense shipbuilders, Defense Research & Misc Defense Development, Ground-based & other weapons systems, Defense-related services Independent power generation & cogeneration, Nuclear energy, Rural electric Electric Utilities cooperatives, Nuclear plant construction, equipment & svcs, Electric Power utilities, Gas & Electric Utilities Environmental Svcs/Equipment Environmental services, equipment & consulting Fisheries & Wildlife Fisheries & wildlife, Fishing, Hunting & wildlife Mining, Mining services & equipment, Coal mining, Metal mining & processing, NonMining metallic mining Industry Misc Energy Oil & Gas Waste Management Accountants Commercial Banks Credit Unions Finance/Credit Companies Insurance Water Utilities, Power plant construction & equipment, Energy production & distribution, Alternate energy production & services, Energy, Natural Resources and Environment Independent oil & gas producers, LPG/Liquid Propane dealers & producers, Natural Gas transmission & distribution, Oilfield service, equipment & exploration, Gasoline service stations, Petroleum refining & marketing Waste management Accountants Commercial banks & bank holding companies, Banks & lending institutions Credit unions Credit agencies & finance companies Property & casualty insurance, Life insurance, Insurance companies, brokers & agents, Insurance, Accident & health insurance Finance, Insurance & Real Estate, Tax return services, Investors, Financial services & consulting, Credit reporting services & collection agencies, Other financial services Mobile home dealers & parks, Mortgage bankers and brokers, Building operators and managers, Other real estate services, Title insurance & title abstract offices, Real estate agents, Real estate, Real Estate developers & subdividers Savings banks & Savings and Loans Payday lenders, Securities, commodities & investment, Commodity brokers/dealers, Security brokers & investment companies, Venture capital, Hedge Funds, Student loan companies, Private Equity & Investment Firms, Stock exchanges

Finance, Insurance & Real Estate

Misc Finance Real Estate Savings & Loans Securities & Investment

Notes: Category definitions from the Center of Responsive Politics. The level of aggregation used to control concentration indsutries in the paper is that of industries. () Not all categories are presented in every industry to save space

56

Sector

Health

Labor

Lawyers & Lobbyists

Misc Business

Appendix Table A.2 (continued): Industry categories for special-interest PACs Examples of categories Industry Psychiatrists & psychologists, Chiropractors, Optometrists & Ophthalmologists, Health Health Professionals professionals, Physicians, Other physician specialists, Dentists, Other non-physician health practitioners, Nurses, Pharmacists Outpatient health services (incl drug & alcohol), Optical services (glasses & contact Health Services/HMOs lenses), Mental Health Services, Health care services, HMOs, Medical laboratories, Home care services, AIDS treatment & testing Hospitals/Nursing Homes Health care institutions, Nursing homes, Hospitals, Drug & alcohol treatment hospitals Misc Health Health, Education & Human Resources Medical supplies manufacturing & sales, Pharmaceutical wholesale, Pharmaceutical Pharmaceuticals/Health Products manufacturing, Biotech products & research, Personal health care products, Nutritional & dietary supplements, Health care products Building Trade Unions Building trades unions Energy-related unions (non-mining), Communications & hi-tech unions, Mining unions, Industrial Unions Manufacturing unions, IBEW (Intl Brotherhood of Electrical Workers), Automotive unions Other unions, Agricultural labor unions, Defense-related unions, Entertainment unions, Misc Unions General commercial unions, Food service & related unions, Retail trade unions, Health worker unions, Commercial service unions, Labor Unions Police & firefighters unions & associations, Federal employees unions, Civil service & Public Sector Unions government unions, State & local govt employee unions, Teachers unions, US Postal Service unions & associations Railroad unions, Other transportation unions, Merchant marine & longshoremen Transportation Unions unions, Transportation unions, Teamsters union, Air transport unions Corporate Lawyers & law firms, Attorneys & law firms, Legal Services, Trial Lawyers & Lawyers/Law Firms law firms Lobbyists Lobbyists & Public Relations, Registered Foreign Agents Beer, Wine & Liquor Liquor stores, Wine & distilled spirits manufacturing, Liquor wholesalers, Alcohol, Beer General business associations, Chambers of commerce, Small business associations, Business Associations International trade associations, Pro-business associations, Business tax coalitions Commercial photography, art & graphic design, Management consultants & services, Business Services Political consultants/advisers, Outdoor advertising services, Advertising & public relations services, Security services, Employment agencies, Direct mail advertis Casinos/Gambling Indian Gaming, Casinos, racetracks & gambling Plastics & Rubber processing & products, Adhesives & Sealants, Paints, Solvents & Chemical & Related Manufacturing Coatings, Explosives, Household cleansers & chemicals, Chemicals Beverages (non-alcoholic), Confectionery processors & manufacturers, Artificial Food & Beverage sweeteners and food additives, Fish Processing, Food catering & food services, Restaurants & drinking establishments, Beverage bottling & distribution Lodging/Tourism Lodging & tourism, Travel agents, Hotels & motels, Resorts Warehousing, Import/Export services, General commerce, Correctional facilities constr Misc Business & mgmt/for-profit, Wholesale trade Industrial/commercial equipment & materials, Fabricated metal products, Precision Misc Manufacturing & Distributing instruments, Clothing & accessories, Electroplating, polishing & related services, Household & office products, Photographic equipment & supplies, Toys, Aluminum Miscellaneous repair services, Physical fitness centers, Video tape rental, Beauty & Misc Services barber shops, Equipment rental & leasing, Services, Funeral services, Laundries & dry cleaners, Pest control Amusement parks, Professional sports, arenas & related equip & svcs, Recreation/Live Entertainment Recreation/Entertainment, Amusement/recreation centers Retail trade, Hardware & building materials stores, Department, variety & convenience Retail Sales stores, Catalog & mail order houses, Miscellaneous retail stores, Vending Machine Sales & Services, Direct sales, Consumer electronics & computer stores Steel Production Steel Textiles Textiles & fabrics Air freight, Air transport, Aviation services & airports, Aircraft parts & equipment, Space Air Transport vehicles & components, Airlines, Aircraft manufacturers, General aviation (private pilots), Express delivery services Automotive Car rental agencies, Auto manufacturers, Auto repair, Truck/Automotive parts & accessories, Automotive, Misc, Auto dealers, new & used, Auto dealers, foreign imports Buses & Taxis, Motor homes & camper trailers, Pleasure boats, Recreational transport, Freight & delivery services, Transportation, Bus services, Bicycles & other nonmotorized recreational transp, Motorcycles, snowmobiles & other motorized vehicles

Transportation

Misc Transport Railroads

Manufacturers of railroad equipment, Railroads, Railroad services, Railroad transportation Ship building & repair, Sea freight & passenger services, Sea transport, Cruise ships & Sea Transport lines Trucking Trucking companies & services, Trucking, Truck & trailer manufacturers Notes: Category definitions from the Center of Responsive Politics. The level of aggregation used to control concentration indsutries in the paper is that of industries. () Not all categories are presented in every industry to save space

57

Appendix,Table,A.3:,Automated,News,Searches Candidate,news,stories, about,candidates,(total, stories) Number,of, newspapers 1 1 18 18 24 36 40 31 95 74 84.6 98.4 105.3 104.9 104.2 88.6 78.0 131.5 87.5 142.7 2.92 1.94 2.70 1.75 1.43 106.9 104.8 1.60 2.11 4.92 2.80 3.06 2.74 3.64 mean 44.5 20.0 109.9 77.3 std.,dev. 34.6 28.3 88.1 90.6 mean 2.50 0.50 3.25 0.93 std.,dev. 0.71 0.71 2.98 1.38 mean 0.07 0.03 0.04 0.01 0.02 0.04 0.01 0.08 0.02 0.01 Candidate,news,stories, discussing,PAC,money, (PAC,stories)

Ratio,of,PAC, stories/total,stories std.,dev. 0.04 . 0.02 0.01 0.02 0.04 0.01 0.23 0.03 0.02

year 1982 1984 1986 1988 1990

1992

58

1994 1996

1998

2000

2002 44 139

States MA OK CA,2FL,2GA,2IL,2IN,2OH,2OK,2WA CA,2IN,2MA,2MN,2OH,2TX,2UT IL,2IN,2LA,2MA,2MN,2NC,2OK,2OR,2TN,2 TX AL,2AZ,2CT,2FL,2GA,2IA,2IL,2IN,2NC,2OH,2 OK,2OR,2WI CA,2CT,2FL,2IN,2MA,2MS,2UT,2WA,2WI IA,2MA,2MI,2MN,2MS,2NC,2NM,2OK,2TN,2 TX AL,2AZ,2CA,2CO,2CT,2FL,2GA,2IA,2IL,2NC,2 NV,2OK,2OR,2UT,2WA,2WI CA,2CT,2IN,2MA,2ME,2MI,2MN,2MS,2OH,2 TN,2TX,2UT,2WA,2WI AL,2CO,2GA,2IA,2ID,2IL,2ME,2MI,2MN,2NE,2 NM,2OK,2OR 59.7 91.5 59.7 97.1 0.69 1.87 1.12 3.07

Total

0.01 0.03

0.02 0.08