Sei sulla pagina 1di 22

American Economic Association

A Time-Series Analysis of Crime, Deterrence, and Drug Abuse in New York City
Author(s): Hope Corman and H. Naci Mocan
Source: The American Economic Review, Vol. 90, No. 3 (Jun., 2000), pp. 584-604
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/117344 .
Accessed: 28/06/2014 18:17

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.

http://www.jstor.org

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
A Time-Series Analysis of Crime, Deterrence, and Drug Abuse
in New York City

By HOPE CORMANAND H. NACI MOCAN*

Since Gary S. Becker's (1968) groundbreak- Contradictoryresults can be explained, at least


ing work on the economics of crime, econo- in part, by the empirical problems inherent in
mists have expanded upon both the theory and crime research, the most significant being the
the empiricalanalysis of crime (e.g., Isaac Ehr- simultaneitybetween crime and criminal-justice
lich, 1973; M. K. Block and J. M. Heineke, sanctions.' Thus, after 30 years of empirical
1975; Ehrlich, 1975; Ann Dryden Witte, 1980). researchthere is no consensus on the impact of
According to the standard theoretical frame- police and arrestson criminal activity.
work, optimizingindividualsengage in criminal The purpose of this study is to provide new,
activities dependingupon the expected payoffs and potentially more refined, evidence on the
of the criminalactivity, the returnto legal labor- crime-deterrencerelationship using a unique
marketactivity, tastes, and the costs of criminal dataset, which consists of monthlyobservations
activity, such as those associated with appre- in New York City for nearly 30 years. This is
hension, conviction, and punishment.Excellent the only data set of its kind, based on high-
reviews of the literatureappearin Daniel Nagin frequencyobservationsof five differentcrimes,
(1978), SharonLong and Witte (1981), Richard the correspondingarrests,the size of the police
Freeman (1983), and Theodore G. Chiricos force, and a povertyindicator,spanningdecades
(1987). While some studies reportedevidence of experience in one city.2 Consequently,this is
that increases in criminal-justicesanctions re- the firstpaperthat employs high-frequencytime
duce criminal activity (Ehrlich, 1975; Witte, series of individualcrime categories to circum-
1980; StephenK. Layson, 1985;JeffreyGrogger, vent many problems found in studies that em-
1991;StevenD. Levitt, 1997), othersfoundeither ploy cross-sectional or low-frequency (e.g.,
a weak relationship,or none at all between the annual)time-seriesdata sets. We also use recent
two (Samuel L. Myers, Jr., 1983; James Peery advancesin time-serieseconometricsto test and
Cover and Paul D. Thistle, 1988; Christopher correct for problems that may have contami-
Cornwell and William N. Trumbull, 1994). nated the results of previous time-series analy-
ses of crime. We find robust evidence for the
deterrenteffects of arrests and police on most
* Corman:Departmentof Economics, Rider University categories of serious felony offenses.
2083 Lawrenceville Road, Lawrenceville, NJ 08648, and Another unique feature of the study is the
National Bureau of Economic Research; Mocan: Depart- additionof drug-useproxies. In the 1980's and
ment of Economics, University of Colorado-Denver,Cam-
pus Box 181, P.O. Box 173364, Denver, CO 80217, and
into 1990 the media focused much attentionon
National Bureau of Economic Research. This research is drug abuse, crime control, and the criminal-
supportedby a grant from the National Institute of Drug justice system. It had been claimed that in-
Abuse to the NationalBureauof Economic Research(Grant
No. 1-R03-DA06764). An earlierversion of this paper was
presented at the 1996 American Economic Association
Meetings in San Francisco, CA. Ofira Schwartz, Keith ' FranklinFisher and Daniel Nagin's (1978) article de-
Amadio, Joseph Bucs, and Ronald Teodoro helped in data scribing the problem is a classic in the field. Recent litera-
collection. Timothy Potter, Jennifer Giellis, Erdal Tekin, ture suggests several new approaches to the simultaneity
Melissa Anderson, Paul Niemann, and Danny Rees pro- problem: using careful empirical analyses of individual
vided assistance in data analysis. John Lott, Jody Overland, ratherthan aggregate data (Grogger, 1991; Helen Tauchen
and especially Michael Grossman provided very valuable et al., 1994), and finding better exogenous instrumentsfor
suggestions. We thanktwo anonymousreferees for helpful identification(Levitt, 1996, 1997).
comments. Any opinions expressed here are those of the 2 Among the advantagesof using just one city is the fact
authors, and should not be assumed to be those of the that there is one unit defining and collecting crime and
grantingagency, Rider University, University of Colorado- deterrencedata, which prevents inconsistencies across ob-
Denver, or NBER. servations.
584

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN:CRIME,DETERRENCE,AND DRUG ABUJSE 585

creases in violence and other crimes were due lationshipbetweendruguse and crimewrittenby
solely to the "drug epidemic." Many policy criminologists.This literaturedoes not, however,
makers believed that the "crime problem"was provideconclusive evidence on the interrelation-
strictly due to the "drugproblem."In response ship betweendruguse and crime.In theirreview
to the crack epidemic and soaring crime rates, of recentstudieson the relationshipbetweendrug
there was a large increase in resources devoted abuse and predatorycrime, Jan M. Chaikenand
to drug control.3 The inclusion of drug-use Marcia R. Chaiken (1990) present surprisingly
proxies allows us to compare the relative mag- guardedconclusionsstatingthat "- thereappears
nitude of the effects of local law-enforcement to be no simple general relation between high
activities on crime with the magnitudeof vari- ratesof druguse andhigh ratesof crime."LanaD.
ations in drug usage on crime. Our results indi- Harison (1992), in another literaturereview,
cate that drug usage has only a small effect on comes to a similarconclusion:thatthe causallink
some property crimes, and that local law- between nondrugcrime and drug usage has not
enforcement effects on crime are stronger and yet been established,althoughthe two appearto
more significant. be correlated.
In this paperwe add drug use to the analysis,
I. The Model a variablenot usually included in economics of
crime studies. Drug use, according to Paul
The crime-supplyequation,can be writtenas: Goldstein (1985), can affect criminal activity
through three channels. The first is the "phar-
(1) CR = f(POL, ARR, POV, Q), macological" effect: drug use may increase ag-
gression and therefore violent crime. The
where CR stands for criminal activity, POL second is the "economic" effect: some users
representsthe size of the police force, ARR is turnto crime to finance expenditureson drugs.5
crime arrests,POV representslegal-marketop- The third is the "systemic"effect: violence oc-
portunities,proxied by poverty, and Q is drug curs in the drug marketbecause the participants
use. Excluding the drug-usevariable,this equa- cannot rely on contracts and courts to resolve
tion is one variant of the model used in the disputes.
economics of crime literature.4It has been so The net effect of these factors on crime de-
thoroughlydiscussed in the articlescited above, pends on their relative magnitudes and on the
that we present only a brief description. The price elasticity of demandfor illegal drugs.6For
model includes police as a determinantof crime example, if the demand for drugs is price in-
because police officers may have an additional elastic,7an increasein drugconsumption,due to
generaldeterrenteffect in additionto arrestsfor
the specific crimes. POL and ARR are endog-
enous variables, where the size of the police 5 According to Bruce D. Johnson et al. (1985), among
force depends on criminal activity, the fiscal heroin users in New York City during the early 1980's,
condition of the city, the extent of drug use, as approximately 33 percent of total income (cash plus in-
well as other exogenous variables,such as elec- kind) was derived from nondrug criminal activity and ap-
toral cycles (e.g., Levitt [1997]). Crime arrests proximately 36 percent of total income was derived from
drug sales. Otherincome was derived from earnings,public
are a functionof the criminalactivity and police support, and support from family members and friends.
force. The economic model of crime predicts Although some users may substitutedrug crimes for non-
that an increase in deterrencevariables (police drugcrimes when the drugmarketexpands,a studyby Peter
and arrests)should reduce criminalactivity, and Reuter et al. (1990) found evidence of complementarity
between drugincome and income from both nondrugcrimes
an increase in poverty should increase it. and legal earningsfor a cross section of individualsarrested
Thereexists a voluminousliteratureon the re- for drug sales in Washington,DC in the mid-1980's.
6 To the extent that the "systemic" effect captures the
violence stemming from the sellers who fight over the
3 See David W. Rasmussenand Bruce L. Benson (1994) producer surplus, the proxies of drug consumption do not
for a discussion of these trends. accuratelycapturethe "systemic"effect.
4 Theoreticaljustificationof the inclusion of drug use in ' Recent studies by Becker et al. (1991) on cigarettes,
the crime equation can be found, among others, in Ehrlich Henry Saffer and FrankChaloupka(1995) on cocaine and
(1973). heroin, Jan C. van Ours (1995) on opium, and Michael

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
586 THEAMERICANECONOMICREVIEW JUNE 2000

a rightwardshift of the supply curve of drugs, imum damagedue to inflation,influencesthe re-


would be associated with a large decrease in portingratesand the categorizationof this crime.
price, and a decrease in drug spending.8If the We include two criminal-justice sanction
economic effect dominates,then this will result variables:arrestsfor the specific crime, and the
in a decrease in crime. The reverse would be numberof police officers. The monthly number
true if the demand for drugs is price elastic.9 of arrestsby crime category was obtainedfrom
the Crime Analysis Unit of the NYPD. The
II. Data numberof police officers was obtainedfrom the
Office of Management Analysis and Planning
This studyutilizesa uniquedataset, which was of the NYPD. The data obtained from the
constructedusing recordsof the Crime Analysis NYPD span the period January 1970-Decem-
Unit of the New York City Police Department ber 1996. Data restrictionspreclude the use of
(NYPD),the Office of ManagementAnalysisand additionalcriminal-justicevariableswhich have
Planningof the NYPD, the New York City De- traditionally appeared in crime-supply func-
partmentof Health,and the New York City De- tions: the probabilityof conviction and the av-
partment of Human Resources. The Crime erage length of sentence.'2
Analysis Unit of the NYPD has collectedconsis- Our measure of drug use is the number of
tent monthlydata on crime commission and ar- deaths in New York City which are due to drug
rests since 1970. These dataform the core of our poisoning. Data on the amount of drugs con-
data set. Almost all of the researchon criminal sumed are not available. Jeffrey A. Miron and
activityhas focused on the seven "index"crimes: Jeffrey Zwiebel (1991, 1995) and Miron (1998)
murder,feloniousassault,rape,robbery,burglary, used the death rate from the cirrhosis of the
grandlarceny,andmotor-vehicletheft.We exam- liver, the deathrate from alcoholism, the drunk-
ine all but rape and grand larceny.We exclude enness arrest rate, and the number of first ad-
rape because reportingfrequenciesvary signifi- mittances to mental hospitals for alcoholic
cantlyover time.10We exclude grandlarcenybe- psychosis as proxies for alcohol consumption
causethe definitionof this crimehas changedover during prohibition for similar reasons. The
time. In 1986, New York increasedthe value of number of drug deaths was obtained from the
the damagefrom$100 to $250 to be includedinto New York City Health Department,and covers
the categoryof grandlarceny." This change, as the period 1970-1996. Although the codes al-
well as the decreasein the real value of the min- low the coronerto specify the type of drug, the
vast majorityof cases were coded as drug-type
unknown. Therefore, we cannot disaggregate
drugdeathsby type of drug.These datahave the
Grossman and Frank Chaloupka (1998) on cocaine find advantageof not requiringhonest self-reporting,
price-elasticityestimates ranging from inelastic to elastic. and of being closely tied to heavy use.
8 Rasmussen and Benson (1994) discuss changes in the

cocaine marketbetween 1979 and 1990. They conclude that


duringthis periodincreasesin cocainesupplyhave been stron-
12
ger and more significantthan increases in cocaine demand. It is importantto ascertainchanges in the severity of
This is because quantityincreases have been coupled with punishmentbecause of the possibility of indirect effects of
price reductions.Among otherreasons,they cite the specific drug usage on crime. That is, if higher drug usage causes
exampleof theintroductionof crackcocaineas a technological greaternumbersof drug criminals to be sent to prison, and
advancementwhichallowedfor the largeincreasein supply.In if this results in prison crowding, then the effect of drug
our case, even thoughwe do not have a clean and continuous usage on nondrug crime may be due to reduced sanctions
price variable,the price measurewe have is negativelycorre- for nondrugcrimes. There is no evidence for reductionsin
lated with the measureof the druguse. imprisonmentsfor nondrug felons. According to the New
9 In addition,drug usage may have an indirecteffect on York State Department of Corrections, during both the
crime. If increased drug usage causes increased levels of 1970's and the 1980's, prison commitmentsrose faster than
criminal-justice resources to be devoted to drug crimes arrestsfor nondrugcrimes in New York. Thus, when drug
(such as more police officers being devoted to drug busts), usage was rising, the probability of imprisonment given
this may reduce the probability of arrest for those who arrestfor nondrugfelonies was also rising.
commit nondrugcrimes. 13 At the beginning of the crack era, before many med-
10 See Jim Galvin (1985). ical doctorswere awareof the drug, some of the deaths may
" Source: New York City Police Department Crime have been diagnosed as nondrugrelated. As the awareness
Analysis Unit personal communication. increased, the probability of missing the drug usage as a

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN: CRIME,DETERRENCE,AND DRUG ABUSE 587

Because of the difficulty and importanceof TABLE 1-MEANS OF THEVARIABLES


measuringdrug usage, we examined two alter-
native measures of drug usage: the number of Monthly means, NYC
January1970-
releases from all hospitals where the primary VARIABLE December 1996
reason of admission was drug dependence and
DRUG USAGE
drugpoisoning,14 and felony drugarrests.15 The Drug deaths 60
results are discussed in the paper, but because POLICE
they were similar to those obtained with drug Number of officers 27,208
deaths, they are not reportedin detail. ARRESTS
It has been claimed that drug deaths mlaybe Murderarrest 99
Assault arrest 1,421
inversely related to drug usage. When drug Robbery arrest 1,809
prices are high, usage would decrease, but ad- Burglaryarrest 1,226
verse reactions by drug users would increase Motor-vehicle theft arrest 782
due to greater adulteration.To examine this CRIMES
Murder 138
possibility, we utilized data on the prices of Assault 2,657
heroin and cocaine from Drug Enforcement Robbery 6,882
Agency purchases of drugs in New York City Burglary 11,975
for 1977 through 1989. Because the data on Motor-vehicle theft 8,112
price are very noisy and have gaps, they are not POVERTY
AFDC cases 250,106
suitablefor use in the regressions.Nevertheless,
we computeda three-monthmoving average of
prices to reducethe noise and then calculatedits until December 1996. Note that we use actual
zero-ordercorrelationwith our three measures crimes committed ratherthan rates because of
of drug use. The correlations ranged from controversiesover the populationin New York
-0.53 to -0.71.16 That is, for both price mea- City duringthis period. The graphspresentedin
sures and for all three drug-usemeasures,there Figures 1-5 display police, drug-relateddeaths,
was a negative and significant correlationbe- and the five crimes. They include the actual
tween price and our usage proxies. Thus, there values of the variables, represented by the
is no evidence that our usage indicator(s) is jagged line, along with the underlying trend
negatively related to actual usage. component, representedas a smooth line. The
Table 1 presents the mean values for the trendcomponentenables the readerto visualize
variables. All time series relate to New York the long-term swings of the variable, with most
City and cover the period from January 1970 of the noise eliminated. The trend components
were calculated using the Hodrick-Prescottfil-
ter.17 The figuresdemonstratethe importanceof
covering a long time span, and thus observing
cause of death should have decreased.Thus, the drug-death sufficient variationin the variables.
data may underestimate the prevalence of the drug use
around1985, when crackwas first introducedto the market.
14 These data, which are obtainedfrom New York State-
17
Wide Planningand ResearchCooperativeSystem, cover the We used the Hodrick-Prescottfilter (RobertJ. Hodrick
period 1980-1996. and EdwardC. Prescott, 1997) to obtainthe slowly evolving
15 These data cover the entire 27-year period. This series trendcomponent.In this procedure,the trendcomponentin
does not have the disadvantageof the other two, but has the variableunderinvestigation,F is obtainedby solving the
another drawback.Drug arrests is a potential policy vari- following convex minimizationproblem:
able, where police decide on the level of drug arrests,
depending on the relative severity of not only the drug T T

problem but also the crime problem. In addition, one may min (X - r)2 +A (X - 2rt, +rt2)2,
expect that increased drug arrestscause a decrease in non- t=1 t =3
drug arrests,holding police constant.Despite these reserva-
tions, the three measures of drug usage appear to move where X is the variable of interest, and A is the weight on
together, as reportedin the data section. squared second difference of growth component, which
16 The correlationswere negative, but higher in absolute penalizes accelerationin the trend. Following previous ex-
value between drug-use measures and prices for five-, sev- amples (e.g., Keith Blackburnand Morten 0. Ravn, 1992;
en-, and nine-monthmoving averages, rangingfrom -0.57 Mocan, 1999), A is set to be 1,600, but the decomposition
to -0.72. was not sensitive to the variationsin the value of A.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
588 THEAMERICANECONOMICREVIEW JUNE 2000

34000 -

32000 -

30000 - _

A)
28000-

26000-

24000

22000-

20000 - r V F
O v- el o) r- 00 C O m t \) r<- 00 (ON cm
r t :) )
- >- >- >- >- > > >
>F . 0o 00 OFF CE0 00NNEw
00 YO
00 RK00C 00 00 C\ O

FIGURE1. POLICEOFFICERS,NEW YORKCITY

Figure 1 presentsthe numberof police officers level. The large variationin the size of the police
in New York City. The variableis "totaluniform force allows us to observe a range that is far
strength,"and consists of the numberof sworn greaterthanthatexperiencedin most cities.
officerson thepayroll.It excludesindividualswho Deaths due to drugs (Figure 2), declined be-
have left the police force but are receivingtermi- tween 1970 and 1978. They rose after 1979,
nal paychecks.Similarly,it does not includecivil- reachinga plateauin the firsthalf of the 1980's.
ians, or officerswho have been hiredbuthave not They began rising again around 1985, reaching
completedtheir six-monthtrainingcourse at the a peak in 1988. In the peak year of 1988 there
Police Academy.18The steepdeclinebetweenthe were an average of 101 drug deaths per month
mid-1970's and early 1980's was an exogenous compared to 72 in the previous peak year of
event due to the fiscal crisis in New York City. 1971-an increase of almost 40 percent. Com-
Due to layoff and attrition,the police force de- paringthe trough,which occuffed in 1978 (with
clined about one-third from a peak of about about 21 deaths per month),to the peak of 1988
32,000 in 1970 to a troughof under22,000 in the provides an increase of almost 400 percent.
early 1980's. The police force had been increased Drug deaths declined between 1988 and 1991,
to about27,500 in 1988, only half the way backto and startedrising again after 1991 and leveling
its former peak. After declining again between off at an average of 92 deaths per month in
1988 and 1991, the police force was increasedto 1993-1994. Another decline began in 1995.
about 31,000 in 1996, almost reachingits 1970 The patternof drugdeathspresentedin Figure
2 is consistentwith other drug-useproxies. The
simple correlationbetween drug deathsand drug
18 The NYPD and the transit and housing police forces arrestsis 0.65 (1970-1996). It is 0.70 between
merged in 1995. We have subtractedthe size of the transit drug deaths and hospital releases (1980-1996),
and housing police from the totals since April of 1995. and 0.85 between drug arrests and hospital

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN:CRIME,DETERRENCE,AND DRUG ABUJSE 589

160 -

140

120

100

PC 80

60

40

fz10 i-7- 7--f I


20

O '- e~ kf 00 ON 0 e - - r - -t- 00 ON 0 -r - --r


(? (F t- t- (
t- ? t~(~ t-(~ 00 00 00 00 00 00 00 00 00 00 O ON ON o ON ON

FIGURE2. DRUG DEATHS,NEW YORKCITY

releases (1980-1996).19 Although these mea- lowest level in 27 years. Assaults (Figure3), ex-
sures are imperfect,they appearto be measuring hibit a seasonal cycle, with increases in the
the same trends. spring and summer months. Assaults began to
Figures 3-5 present the number of com- increase at the beginning of 1982, over three
plaints for five felony crimes: murder,assault, years before the upturnin murdersand druguse,
robbery,burglaryand motor-vehicle theft. Ex- reaching a peak in 1989 and then declining
amining trends in these data allow us to place afterwards.Assaults in 1989 were 67 percent
the "crime epidemic" of the 1980's into per- higher than their previous peak in 1979 and
spective and allows a visual comparison of were 71 percent higher than the most recent
crime trends to trends in the police force and troughin 1983. By 1996, assaults had fallen 25
drug usage. The monthly number of murders percent from the 1989 peak.
(Figure 3) turned up in mid-1985. The total Robberies (Figure 4) exhibit a cyclical pat-
number of murdersin 1990 was 2,262, an in- tern.The latest upswing took place in mid-1987,
crease of 24 percent from the previous peak in and the latest downturnwas in 1991. The num-
1981. Murdersincreasedabout 63 percentfrom ber of robberies in 1990 was about 7 percent
the mid-1980's until 1990. Startingin 1991, the below the previous peak in 1981. Burglaries,
year in which police force began its second also displayed in Figure 4, declined almost 40
expansion, murdersdeclined steadily, until the percentbetween the first and the second half of
average monthly murdersin 1996 was 82, the the 1980's. Burglariesremainedfairly constant
between 1985 and 1989 and then declined. The
1996 level in burglarieswas about29 percentof
19The correlations between seasonally adjusted series the previous peak in 1980. Motor-vehiclethefts
were 0.68, 0.72, and 0.98, respectively. (Figure 5), which exhibit cyclical behavior

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
590 THEAMERICANECONOMICREVIEW JUNE 2000

10000 - - 250

9000 -
Murders
8000-20

7000
-

6000- 5

5000-

4000 -A -0

3000-

2000 -V-
sals5

1000

o4 - I . i 1 00 ON 0I I Cl
.I , o N 0
N N N N N N N N N oN 00 00 00 0 0t 00 00 00 0 CoN C) CC "C

FIGURE 3. MURDERS/FELONY ASSAULTS, NEW YORK CITY

similar to robberies and burglaries, increased III. The SimultaneityProblem


dramaticallyafter 1985. The level in 1990 was
37 percent higher than the previous peak in The main problem with cross-sectional data
1982 and about 85 percent higher than its most is identification. If crime, police, arrests, and
recent trough in 1985. Similar to other crimes, druguse are all determinedsimultaneously,it is
motor-vehicletheft startedto decline after 1990. difficult to find enough exogenous variables to
As Figures 3, 4, and 5 demonstrate,in 1996 be meaningfully excluded from some of the
murders,robberies,burglaries,and motor-vehicle equations to allow identification.Using a time
theft were at their lowest levels since 1970. series of high frequencyallows us to circumvent
A cursory look at the data shows that there most of the simultaneity issues, as well as al-
was a large increase in drug consumptionin the lowing an explorationof some of the dynamics
1980's as proxiedby drugdeaths,and that some of criminalbehavior.
crime categories increased significantly during Considerequation(2), which is the empirical
the same decade, although the timing is not counterpartof equation (1)
perfectly coincidental. In addition, the upsurge
in crime occurredduring a period in which the (2) CRi, = ci + aijCR,_j+ +YikQt-k
police force was growing. Thus, a cursory in-
spection of the graphsmight supportthe notion + 1yimPOLt_m + &ipARRt-p
that the increases in crime were caused by in-
creases in drug consumptionand that local law- + 1iqPOVt-q + ltIinSEASn + Sit,
enforcement efforts were not effective in
combating crime. Any visual relationship be- where CRi stands for ith crime (i = 1: Rob-
tween crime, police, and drug use is only spec- bery, i 2: Burglary,etc.), andj, k, m, p, and
ulative. To isolate the impacts of deterrenceand q are the lag lengths of crime, drug deaths, the
drug use, we present a multivariatestatistical numberof police officers, arrests,and poverty.
analysis. SEAS stands for monthly dummies to control

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN: CRIME,DETERRENCE,AND DRUG ABUSE 591

25000 -

20000 -

rA
~~~~~~~~~~~~~~Burglaries
15000

10000

N N N N N N
O4 N 00OREB00 00 L 00 ES
00 00
OE 00 00
OK 00 00

FIGURE4. ROBBERIES/BURGLARIES,
NEW YORKC1TY

for the impact of seasonality. In equation (2) imum time frame.20Note that neither the six-
j ' 1, k ' 1, m 2 0, p - 1, q 2 0. Put month nor the nine-month lag in testing and
differently, it is postulated that the number of trainingincludes the lag between the increasein
crimes committedin month t depends upon the crime and the policy decisionto increasethe size
past dynamics of the same criminalactivity, the of the force. Thus, the actual lag may be even
past values of drug use, the current and past longer.
values of the numberof police officers, the past We also performedan empirical test to con-
values of arrests,and the currentandpast values firm whether the institutional evidence, cited
of the rate of poverty. Poverty is approximated above, was supported by our data. We ran a
by the numberof cases of Aid to Families with regression of police on its 18 lags, and the
Dependent Children(AFDC). contemporaneousvalues and 18 lags of AFCD
Using monthlydata allows us to eliminatethe recipients,drugdeaths, and total crime (the sum
simultaneityissue betweenpolice and crime.Our of all five crimes).2' We could not reject the
police variable,"totaluniformstrength,"includes hypothesis that 0-6 lags of crime have no in-
only officerswho have completedtheirsix-month fluence on police (X2 statistic was 8.83 with
course at the Police Academy. Thus, it takes a seven degrees of freedom, with a p-value of
minimumof six monthsbetweena policy change 0.26). On the other hand, we strongly rejected
andthe deploymentof officerson the street(Todd the hypothesis that the remaininglags of crime
S. Purdum, 1990). According to the Applicant (7 through18) do not have an impact on current
Processing Departmentof the NYPD, the six-
monthperiodis a minimum,andwill only occurif
20
the Departmentplannedon the new recruits.An Personal communication with Captain William
unplannedpolicy changerequiresthatcertaintests Schmidt of the Applicant Processing Departmenton Sep-
tember 15, 1998.
beyondthe writtenexaminationbe given to appli- 21
Consistentwith the models estimatedin the paper,all
cants before they enter the Academy. In such a variableswere in differences and a co-integrationterm was
case, a nine-monthlag is a more appropriatemin- included.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
592 THEAMERICANECONOMICREVIEW JUNE 2000

14000

12000 I

8000

6000

4000

2000

0 F V 1 ~ TI I, r -I-__-
O C1 eO) dN t 00 CN 0f O L- 00 ON CD - k1 \C
N N N N N N ~~~~~~~~~~t-
00 00 00 00 00 00 00 00 00 00 O7) C01 (01 O1. 01 CO (1

FIGURE5. MOTOR-VEHICLE
THEFTS,NEW YORKCITY

police with a x2 statistic of 29.72 and a p-value throughthe incapacitationeffect. To gauge the
of 0.003. We obtainedthe same resultswhen we magnitude of such an effect, we explored na-
ran the model with 24 lags of all variables. x2 tional-level data. According to the U.S. Depart-
for 0-6 lags of crime was 9.12 with a p-value ment of Justice study (Brian A. Reaves and
of 0.24, and it was 46.99 with a p-value of Jacob Peres, 1994), in the 75 largest counties of
0.0002. Thus, we confirm, empirically,that the the United States, well over half (63 percent)of
police force reacts to crime with a lag of at least those arrested on a state felony charge were
six months. releasedpriorto case disposition in 1992.22 The
Because currentarrestsare likely to be influ- pretrialrelease rate varies by charge. Only 24
enced by currentcriminalactivity, a simultane- percent of those arrestedfor murderswere re-
ity bias is createdif contemporaneousvalues of leased comparedto 68 percentof those arrested
arrestsare included in the crime equation. Ex- for felony assault.For those who were released,
clusion of contemporaneousvalues of arrests 52 percent were released within a day, and 77
helps identify the crime equation and avoid percent were released within a week. In addi-
simultaneity bias. Using monthly observations tion, most felony defendantswaited longer than
provides a rationalefor lagging crime arrestsby a month for their case to go to trial. In 1992,
one month. It is plausible that increased arrests only 14 percent of felony defendantswere ad-
do not immediately affect criminalbehavior. It judicated within one month, the median time to
takes time for criminals and potential criminals adjudicationwas 83 days, and 10 percent of all
to perceive that such a change has occurred.To felony cases had not been adjudicatedwithin
the extent that it takes at least a month for one year. Thus, most of the felony defendants
criminals to process that information and to
change their behavior, crime should depend on
lagged arrests. 22
The rate was 66 percent in 1988, the first year such
Current arrests could affect current crime data were available.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN:CRIME,DETERRENCE,AND DRUG ABUSE 593

were released shortly after arrestand did not go darmonth.23In addition,we apply the detention
to trial within the month. This suggests that the rates for each crime (Reaves and Perez, 1994)
judicial system does not generate immediate as follows: murder, 76 percent; robbery, 50
incapacitationfor most felony offenders. Along percent; assault, 32 percent; burglary, 49 per-
the same lines, Levitt (1998) found that deter- cent; and motor-vehicle theft, 33 percent. Ap-
rence is a more importantfactor than incapaci- plying the Poisson distribution,assuming that
tation to impact criminal activity. release occurs immediately for those not de-
To estimatethe impactof immediateincapac- tained, and allowing for multiple crimes per
itation on crime, we appliedthe following algo- month for each person arrested, the average
rithm. First, we obtainedinformationabout the numberof the same crimes avertedin the same
offense ratesof criminalswho were arrestedand calendarmonth due to the incapacitationeffect
imprisoned.MarkA. Petersonet al. (1980) cal- is 0.008 for murder,0.09 for robbery,0.06 for
culated that criminals commit the following assault, 0.31 for burglary,and 0.07 for motor-
crimes with the following mean annualfrequen- vehicle theft. For each crime category, allowing
cies: murders, 0.27; robberies, 4.61; assaults, arreststo increase 10 percentin a month would
4.47; burglaries, 15.29; and motor-vehicle result in a same-monthreductionin crimes due
thefts, 5.25. To compute the numberof crimes to incapacitationof: 0.08 for murders, 16 for
that would have been committedin the absence robberies,8 for assaults,38 for burglaries,and 6
of incapacitation, an assumption needs to be for motor-vehicle thefts. As a percent of the
made about the time distributionof crimes. If average number of monthly crimes, these con-
criminals space crimes evenly throughoutthe vert to 0.06 percent for murders,0.24 percent
year, then no crimes except burglarywill occur for robberies, 0.32 percent for assaults, 0.31
more than once per calendarmonth. Since bur- percent for burglaries, and 0.07 percent for
glaries occur every 24 days (15.29 per year), motor-vehicle thefts. This translates into the
only burglars who are arrestedin the first six following incapacitation elasticities of crime:
days of the month(20 percentof those arrested), -0.006 for murders, -0.024 for robberies,
would have gone on to commit anotherburglary -0.032 for assaults, -0.031 for burglaries,and
in the same calendar month. According to the -0.007 for motor-vehicle thefts. Thus, even
Bureau of Justice Statistics (Reaves and Perez, using the upper-boundestimates, the immediate
1994), 49 percent of the burglarsare incapaci- incapacitationeffect is small, with same-month
tated postarrest.Thus, for each burglaryarrest, incapacitation elasticities ranging from 0 to
0.098 burglaries(0.2 X 0.49) will be avertedin -0.O3.24
the same calendarmonth due to incapacitation. It is also possible that first-monthdeterrence
The average number of burglary arrests per effects could be large for released arrestees,if
monthis 1,226 in New York City between 1970 they faced harsh penalties for a subsequentar-
and 1996 (see Table 1). This suggests that a rest while awaiting trial. Surprisingly, the of-
10-percent increase in burglary arrests in a fenders who are released prior to trial and who
given month would generate an incapacitation- are rearrestedfor a second felony during the
relateddecreasein burglariesin the same month pretrialperiod face a probabilityof rerelease of
by 12, which is 0.1 percent of the average 61 percent,almost identical to the initial release
numberof burglaries.Thus, under the assump- rate of 63 percent. Thus, marginalsanctions do
tion of even spacing, incapacitation effects not increase immediately following arrest and
would be zero for all crimes otherthanburglary release.
and would be quite small for burglaries. Although immediate incapacitation and de-
To get an upper-boundestimate of the inca- terrenceeffects were estimated to be small, we
pacitationeffect, we use a Poisson distribution, also estimated the crime equations with the in-
and assume that crimes are random and inde- clusion of the contemporaneousvalue of the
pendent events. On average, offenders who are
incarceratedare assumed to be arrestedin the 23 Fifteen days will be the average length of time in jail
middle of the month. We use the same crime for the average person arrestedin the middle of the month.
frequencieswe cite above, and convert these to 24 These elasticities reflect only same-crime effects, and
15-day rates of commission in the same calen- do not include effects due to cross-crime incapacitation.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
594 THEAMERICANECONOMICREVIEW JUNE 2000

arrests to allow for potential instantaneous stochastic trends or whetherbreaks in their un-
effects. Our results were not sensitive to the derlying trends are responsible for the appear-
inclusion of the contemporaneousarrests.This ance of the unit root. Following Eric Zivot and
suggests that neither simultaneitybias (due to Donald W. K. Andrews (1992), who extended
the inclusion of currentarrests)nor model mis- PierrePerron's(1989) test where the breakpoint
specification (due to the exclusion of current is estimated, ratherthan fixed, and Baldev Raj
arrests)is a major factor in the estimation. We and Daniel J. Slottje (1994) and Mocan (1999),
also used two-stage least squares (TSLS) to who applied the test to several inequality mea-
estimate the effect of arrests on crime. Instru- sures, we applied unit-root tests that allow for
mentationand the results from alternativespec- segmented trends with level and slope shifts at
ifications are explained below. Simultaneityin endogenously determined break points. In no
equation(2) could still emerge if the errorswere case could we reject the hypothesis of a unit
serially correlated. This is also tested in the root. This means thatthe properspecificationof
results section. equation (2) should involve regressing the first
A similar issue arises with currentvalues of difference of crime variableson the first differ-
drug use. To eliminate potential simultaneity ence of police, drug use, arrests, and AFDC
between currentdrug use and currentcrime, we cases, and should not include a time trend as a
lagged drug use one period. However, to the regressor.
extent thatdruguse has an immediateimpacton Although the variables are governed by sto-
crime, we have created a misspecification. As chastic trends, if there exists a linear combina-
with arrests, in one specification we estimated tion of them which is stationarywith zero mean,
the crime equation with the inclusion of the then they are co-integrated.In this case, there
contemporaneousvalue of drug use, and ob- exists a common factor in all, and the variables
tained similar results. We also attacked this do not diverge from one anotherin the long run.
problem with a TSLS approach.The results are Co-integrationtests suggested that there is evi-
explained in Section V. dence of co-integrationbetween crime, arrest,
drug use, and the police force in all cases.
Therefore,the estimated regressions include an
IV. Time-SeriesPropertiesof the Variables error-correctionterm.25
JohnH. Cochrane(1991) points out thatthere
It is well known that the usual techniques of exist stationary and unit-root processes for
regressionanalysis can result in highly mislead- which the result of any inference is arbitrarily
ing conclusions when variablescontain stochas- close in finite samples. Similarpoints have been
tic trends (Clive W. J. Granger and Paul raised and the resilience of the unit-root tests
Newbold, 1974; CharlesR. Nelson and Heejoon against trend-stationaryalternatives has been
Kang, 1984; J. H. Stock and Mark W. Watson, questionedby others (e.g., ChristopherA. Sims
1988). In particular,if the dependent variable and HaraldUhlig, 1991; David N. DeJong et al.,
and at least one independent variable contain 1992; Glenn D. Rudebusch, 1993). Nelson and
stochastic trends, and if they are not co- C. J. Murray (1997) state that recent papers
integrated, the regression results are spurious broughtthe literatureto a full circle on the issue
(Grangerand Newbold, 1974; P. C. B. Phillips, of unit root in U.S. real GDP, and they analyze
1986). To identify the correct specification of the robustnessof recent findings with respect to
equation(2), an investigationof the presence of finite sample implications of data-basedmodel
stochastic trends in the variables is needed. specificationsand the effects of test size. Thus,
First, standard augmented Dickey-Fuller tests given the current controversy on the issue of
were applied. They indicated the presence of unit roots, and given that it is not possible to
unit roots in the variables under investigation. determinethe exact structureof the underlying
Given the evidence provided by David F. datageneratingmechanismwith a finite sample,
Hendry and AdrianJ. Neale (1991) that regime the evidence of a unit root, and therefore the
shifts can mimic unit roots in autoregressive
time series, it is important to investigate
25
whether the variables are indeed governed by The results of these tests are available upon request.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN: CRIME,DETERRENCE,AND DRUG ABUJSE 595

need to employ the datain first-differenceform, exception of robberies, drug use also has a
can be consideredan approximationof the exact short-durationimpact on crime.
structure. In Table 2, the first segment for each crime
category presents the estimated coefficients in
V. Results the corresponding equation. Because the ex-
planatory variables generally enter with more
The regressionresults are presentedin Table than one lag, the sum of the estimated coeffi-
2. The optimal lag-length for each variablewas cients is calculated, which representsthe long-
determined by Akaike Information Criterion run impact of the explanatoryvariables on the
(H. Akaike, 1973), and reported next to the crime variable. $a stands for the s-umof the
crime category. The natural logarithms of the lagged crime growthcoefficients; i,`f is the sum
variables were taken before differencing. Esti- of the coefficients of drug-use growth. Eey,$65,
mations were carriedout using a heteroskedas- and 1? representthe sum of the coefficients of
ticity and serial correlation robust covariance the growth in the numberof police officers, the
matrix with serial correlationup to lag twelve. numberof arrests,and the AFDC cases, respec-
Lagrange-multipliertests were used to deter- tively.
mine whether the residuals of the estimated $i5 for murdersis -0.336, indicating that a
models are white noise. The test statistics re- 10-percentincrease in the growthrate of arrests
ported in Table 2, which are distributedas x2 generates a 3.4-percent decrease in the growth
with 12 degrees of freedom, confirmedthe hy- rate of murders. Although the second lag of
pothesis of no serial correlationin errorsup to police growth (y2) is negative and highly sig-
lag 12.26,27In the interestof space, Table 2 does nificant,the sum of the police coefficients is not
not reportthe estimatedcoefficients of the sea- statistically different from zero. Hence, the re-
sonal dummies or the error-correctionterm. sults suggest that arrests constitute deterrence
All five crime categories are influenced by for murders.For assaults,thereis no compelling
the number of police officers with short lags. evidence of a deterrence effect, although the
For example, changes in the contemporaneous contemporaneous value of police is signifi-
value and two past values of the police-force cantly negative. The growthin poverty, approx-
growth (lags = 0-2) influence the currentrate imatedby the rateof growthin the AFDC cases,
of growth of murders;and the growth rate of has a positive and significant impact on both
assaults are affected by the growth rate of the murdersand assaults.
contemporaneousand immediatepast of police. Law enforcement has a significant deterrent
It is interestingto note that arrestshave dif- effect on robberies, burglaries, and motor-
ferent lag structuresbetween violent and non- vehicle theft. For example, for robberies, the
violent crimes. Arrestshave short-livedimpacts sum of the current and lagged arrest growth
for assault and murder:assaults are influenced (`65) is -0.940, and for motor-vehicletheft it is
by assault arrests up to four months ago, and -0.395. An increasein the growthrateof police
murdersare influencedby three lags of murder officers is associated with a reduction in the
arrests. Robberies, burglaries, and motor- growth rate of burglaries and robberies, al-
vehicle thefts, on the other hand, present a though the relationshipis only significantat the
longer dependence on arrests. Robberies and 11-percentlevel in the latter.
motor-vehicle thefts are influenced by arrests Murderand assault growth rates are not re-
that took place up to 12 and 14 months ago, lated to changes in the growth rate of drug use.
respectively. Burglariesexhibit the longest de- This result, which is sornewhat surprising, is
pendence on arrests with 21 lags. With the consistent with the one reportedby Cormanet
al. (1991). Using an interventionanalysis, they
failed to document a structuralupturn in the
26 The critical value at the 10-percentlevel is 18.54. All time-series behavior of murdersin New York
test statistics are below that number. City around 1985. One explanationfor this re-
27 Very similar results are obtained when the data are
aggregated to quarterly,biannual, and annual frequencies,
sult is that the increase in the supply of drugs
although statistical significance is largely diminished in that constituted the crack-cocaine epidemic
biannualand annual data. reduced the producer surplus fought over,

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
596 THEAMERICANECONOMICREVIEW JUNE 2000

RESULTS
TABLE2-REGRESSION

Models with drugdeaths Models with drugdeaths


(January1970-December1996) (January1970-December1996)
Murder (a) Laggedcrime:1-7 Assault (a) Laggedcrime:1-9
(p,) Laggeddruguse: 1-2 (p,) Laggeddruguse: 1-2
(y) Laggedpolice: 0-2 (,y) Laggedpolice: 0-1
(6) Laggedarrest:1-3 (6) Laggedarrest:1-4
(() LaggedAFDC:0-2 (4) LaggedAFDC:0-1

c = 0.1 15*** (3.504) c = --0. 121*** (-3.731)


a1 = -0.659*** (-7.220) a, = -0.405*** (-4.823)
a2 = -0.494*** (-4.601) a2 = 0-280***(-2.688)
a3 = -0.339*** (-3.297) a3 --0.176 (-1.567)
a4 = -0.314*** (-3.527) a4 =-0.208** (-2.061)
a5 = -0.262** (-2.396) a5 -0.184** (-2.118)
a6 = -0.141 (-1.455) a6 = -0.143* (- 1.699)
a7 = -0.078 (-1.086) a7 =0.001 (0.008)
P3I= 0.012 (0.331) a8 = -0.038 (-0.488)
2 = -0.020 (-0.640) a9 = 0.088 (1.373)
= 0.349 (0.650) f1 = -0.012 (-0.925)
=y = -0.587 (-1.107) 12 = -0.013 (-1.189)
72= - 1.148***(-2.996) yo = -0.286* (- 1.681)
,= - -0.127* (-1.779) Y1= -0.002 (-0.013)
62 = -0.131 (- 1.489) 61 = -0.056 (-0.841)
63 = -0.078 (-1.009) 62 - -0.009 (-0.136)
(0 = 1.515(1.008) 63 = -0.052 (-0.736)
(, = 0.869 (0.523) 64 = 0.045 (0.589)
=2= 0.732 (0.583) 0= 0.043 (0.088)
- 1.346*(1.786)
Y,a -2.287*** (-3.876)
Y.Pi= -0.008 (-0.139) ai =- 1.345** (-2.337)
Yi = -1.385 (-1.471) E/3i= -0.025 (-1.234)
E .i = -0-336* (-1.697) Y= -0.288 (-1.297)
Ui= 3.116* (1.913) Y,5= -0.072 (-0.351)
= 1.389* (1.676)
AdjustedR2 = 0.481 AdjustedR2 = 0.685
X2(12)for (Pi = P12 = 0): 7.06 X2(12)for (p1- = P12 - 0): 7.92

Robbery (a) Laggedcrime:1-5 Burglary (a) Laggedcrime:1-9


(1B)Laggeddruguse: 1-13 (13)Laggeddruguse: 1-3
(,y) Laggedpolice: 0-3 (,y) Laggedpolice: 0
(6) Laggedarrest:1-12 (6) Laggedarrest:1-21
(O LaggedAFDC:0-2 (O LaggedAFDC:0-2

c = 0.066*** (4.562) c -0.0 12 (-0.680)


a1 = -0.090 (-1.198) a1 -0.238*** (-3.599)
a2 = -0.099 (-1.341) a2 =-0.106 (-1.558)
a3
= -0.059 (-0.815) a3 = 0.020 (0.359)
a4 = 0.025 (0.415) a4 - -0.119** (-2.038)
a5 = -0.086 (-1.539) a5 = 0.109* (1.869)

P = 0.014 (1.269) a6 0.105 (1.320)


12 = 0.024* (1.823) a7 -0.021 (-0.291)
3 = 0.032***(2.825) a8 0.1 10 (1.410)
4 = 0.018 (1.102) a9 0.142** (2.565)
,BS= -0.012 (-0.747) 1 - 0.013* (1.931)
6 = 0.002 (0.084) 2 = 0.019* (1.812)
7 = 0.035** (2.510) 3 = 0.025*** (2.797)

18 = 0.037** (2.424) o -0.419***


0 (-2.995)
19 = 0.010 (0.924) 61 =-0.094*** (-2.846)
10 = 0.037***(3.397) 82 = -0.100*** (-2.673)
PI = 0.033***(2.966) 63 = -0.095** (-2.381)

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN:CRIME,DETERRENCE,AND DRUG ABUSE 597

TABLE2-CONTINUED

Modelswith drugdeaths Modelswith drugdeaths


(January1970-December1996) (January1970-December1996)
6112 (1.414)
= 0.019 84= --0.047 (-1.102)
,313 0.033*** (4.128)
= 85= -0.124** (-2.575)
,yo= -0.281* (-1.811) 86 = *-0.158*** (--2.904)
,yl = 0.174 (0.693) i7 = -0.066 (-1.536)
Y2 = 0-509*** (-2.771) 88 = -0.120*** (--2.822)
y3= 0.091 (0.510) i9 = 0.010 (0.228)
8, = -0.152*** (-3.469) =
sio 0.022 (0.620)
82 =-0.099 (- 1.572) 81- = 0-078**(2.178)
i3= -0.160*** (-2.985) 812 =0 .089** (2.420)
84= -0.077 (-1.527) 813= 0.135*** (3.325)
85= -0.051 (-1.058) 814 =0.052 (1.348)
86= -0.083* (-1.910) 815 =0.032 (0.854)
87= -0.024 (-0.620) 816 0-062**(1.962)
88 = -0.059* (- 1.709) 817 =0? 1 14*** (3.475)
i9 = -0.046 (-0.974) 818 = 0-064*(1.826)
810= -0.069* (- 1.863) 819= -0-089** (-2.446)
81, = -0.0978** (-2.586) 820 = -0-076** (-2.206)
812 = -0.023 (-0.558) 821 = -0.041 (-1.347)
(o = 1.515***(2.877) (0 = 0.742 (1.427)
(, = -0.365 (-0.716) =]
= -0.729 (-1.401)
=2= -0.579 (-1.231) 62 = 0.262 (0.479)

lai = -0.310* (-1.720) = 0.001 (0.006)


lai
l13i = 0.283** (2.432) f3 = 0.057*** (3.212)
lYi= - 0.526 (- 1.618) lyi = -0.419*** (-2.995)
is = -0.940*** (--3.247) 18i = -0.355 (-0.983)
4 = 0.571 (0.727) 1; = 0.276 (0.400)
AdjustedR2 = 0.663 AdjustedR2 = 0.656
x2(12) for (Pi = - = P12 = 0): 17.88 x2(12) for (Pi =.. = P12 = 0): 18.06

Motor-vehicletheft (a) Laggedcrime:1-2 Motor-vehicletheft concluded


((3) Laggeddruguse: 1-8
(y) Laggedpolice:0-2
(8) Laggedarrest:1-14
(4) LaggedAFDC:0-1

c = -0.040*** (-2.988) 87= --0.067* (- 1.913)


a1 = -0.230*** (-2.749) 88= --0.069** (-2.205)
a2 = 0.095* (1.823) 89 = --0.034 (-1.038)
= 0.007 (0.708) 810= --0.001 (-0.034)
P2 = -0.013 (-1.395) 81, = --0.013 (-0.386)
33 = -0.009 (-0.975) 812 = -0.029 (-0.736)
34 = -0.005 (-0.515) 813 = 0.102***(3.887)
135= -0.024*** (-2.619) 814 = 0.026 (0.928)
16 = -0.032*** (-3.268) 0= 1.646**(2.373)
7 = 0.006 (0.512) = -1-130
. (-1.626)
I8 = 0.011 (1.087)
,yo= -0.222 (-1.270)
,Y1= 0.256 (0.871) Xai --0. 134 (-1.148)
72= -0.486** (-2.162) fi - -0.059 (-1.483)
8, = -0.073** (-2.385) X = - 0.452 (-1.371)
82= -0.128*** (-3.756) =8i --0.395* (- 1.882)
83 = -0.071** (-2.429) 0= .516 (0.674)
84= -0.012 (-0.311)
85 = -0.017 (-0.630) AdjustedR2 = 0.635
86 = -0.010 (-0.344) x2(12) for (p, = = P12 = 0): 10.99

Notes: The valuesin parentheses arethe corresponding t-ratios.


* Statistically significant at the 10-percentlevel or better.
** Statistically significant at the 5-percent level or better.
*** Statistically significant at the 1-percentlevel or better.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
598 THEAMERICANECONOMICREVIEW JUNE 2000

TABLE 3-SENSITIVITY ANALYSIS

Murder Police Arrests AFDC Drug use


Benchmark -1.385 -0.336* 3.116* -0.008
(-1.471) (-1.697) (1.913) (-0.139)
3 lags -1.467 -0.331* 3.252* 0.012
(-1.351) (-1.679) (1.797) (0.211)
4 lags -1.224 -0.420* 3.569* 0.015
(-0.884) (-1.793) (1.927) (0.223)
5 lags -1.124 -0.653** 3.819* 0.007
(-0.843) (-2.249) (1.959) (0.069)
6 lags -1.379 -0.669* 3.696* -0.036
(-0.966) (-1.772) (1.851) (-0.310)

Assault Police Arrests AFDC Drug use


Benchmark -0.288 -0.072 1.389* -0.025
(- 1.297) (-0.351) (1.676) (- 1.234)
3 lags -0.757* -0.130 1.335 0.002
(-1.803) (-0.891) (1.284) (0.062)
4 lags -0.636 -0.058 1.503 -0.024
(- 1.249) (-0.263) (1.395) (-0.652)
5 lags -0.452 -0.327 1.897* -0.041
(-0.726) (-1.235) (1.91) (-0.850)
6 lags -0.431 -0.497* 2.175** -0.060
(-0.668) (-1.709) (2.125) (- 1.199)

Robbery Police Arrests AFDC Drug use


Benchmark -0.526 -0.940*** 0.571 0.283**
(-1.618) (-3.247) (0.727) (2.432)
3 lags -0.437 -0.426*** 1.819** 0.054**
(- 1.260) (-3.629) (2.274) (2.405)
4 lags -0.881*** -0.469*** 1.325 0.078***
(-2.785) (-3.427) (1.446) (2.616)
5 lags -0.855** -0.499*** 1.590 0.021
(-2.238) (-2.881) (1.470) (0.572)
6 lags -0.731* -0.594*** 1.998* -0.020
(-1.664) (-2.935) (1.661) (-0.326)
9 lags -0.928* -0.603*** 2.095 0.038
(-1.703) (-2.922) (1.435) (0.426)
12 lags -1.068* -0.827*** 0.410 0.205**
(- 1.927) (-3.298) (0.356) (2.017)
15 lags -1.569** -0.719** -0.116 0.356***
(-2.173) (-2.540) (-0.088) (3.267)

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN:CRIME,DETERRENCE,AND DRUG ABUSE 599

Table 3-CONTINUED

Burglary Police Arrests AFDC Drug use


Benchmark -0.419*** -0.355 0.276 0.057***
(-2.995) (-0.983) (0.400) (3.212)
3 lags -0.078 -0.199*** 0.253 0.049**
(-0.200) (-2.872) (0.332) (2.553)
4 lags -0.585 -0.197** -0.112 0.050***
(-1.610) (-2.125) (-0.143) (2.959)
5 lags -0.408 -0.256** 0.025 0.002
(-1.083) (-2.431) (0.034) (0.057)
6 lags -0.647 -0.494*** 0.268 -0.051
(- 1.599) (-3.507) (0.368) (-0.847)
9 lags -0.772 -0.698*** 0.297 0.037
(-1.293) (-3.947) (0.352) (0.650)
12 lags -0.987 -0.769*** 0.545 0.169**
(- 1.236) (-3.294) (0.677) (2.171)
15 lags -0.952 -0.630* 0.619 0.331***
(-1.194) (-1.787) (0.710) (3.805)
24 lags -0.472 -0.227 0.264 0.281*
(-0.636) (-0.543) (0.318) (1.854)

Motor-vehicle theft Police Arrests AFDC Drug use


Benchmark -0.452 -0.395* 0.516 --0.059
(-1.371) (-1.882) (0.674) (--1.483)
3 lags -0.570 -0.217*** 0.036 0.015
(- 1.256) (-3.141) (0.036) (0.841)
4 lags -0.645 -0.230*** -0.244 0.020
(- 1.415) (-2.932) (-0.245) (1.031)
5 lags -0.673 -0.268*** -0.239 --0.005
(- 1.278) (-2.914) (-0.237) (--0.198)
6 lags -0.694 -0.262*** 0.063 --0.093**
(-1.183) (-3.329) (0.057) (--2.465)
9 lags -0.680 -0.459*** 0.811 --0.074
(-1.172) (-3.335) (0.661) (--1.378)
12 lags -0.770 -0.559** 0.650 --0.041
(-1.115) (-2.485) (0.527) (--0.555)
15 lags -0.967 -0.247 -0.374 0.101
(- 1.475) (- 1.243) (-0.364) (1.028)

Notes: Lag lengths of explanatoryvariables in benchmarkmodels are identical to those in


Table 2. The entries are the sums of the estimatedcoefficients. The values in parenthesesare
the correspondingt-ratios.
* Statistically significant at the 10-percentlevel or better.
** Statistically significantat the 5-percent level or better.
*** Statistically significant at the 1-percentlevel or better.

offsetting other effects of drugs on violent To investigate the sensitivity of the results
crime. Another explanationis that drug-related to variations in the lag specifications we re-
violence stems mostly from the interactionbe- estimated the models with ad hoc lag lengths.
tween sellers. This may not be strongly related More precisely, we estimated the models for
to our drug-use measure, which is a proxy of murders and assaults where the explanatory
drug consumption. Increases in the growth of variables enter with 3, 4, 5, and 6 lags. Be-
druguse, however, are associatedwith increases cause robberies and motor-vehicle thefts in-
in the growthrate of robberiesand burglaries.28 clude lags up to 13 and 14 in their original
specifications, we estimated them with lag
lengths of 3, 4, 5, 6, 9, 12, and 15. Finally,
28 The results were very similar when hospital releases because arrests enter with 21 lags in the bur-
and drug arrestswere used as proxies for drug use. glary equation, we estimated burglaries with

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
600 THEAMERICANECONOMICREVIEW JUNE 2000

3, 4, 5, 6, 9, 12, 15, and 24 lags. The results with the current value of the growth in drug
are reported in Table 3. arrestsby the NYPD.
In Table 3, for ease of comparison,the first Third, if the contemporaneous feedback be-
row under each crime category reproducesthe tween crime and drug use is stronger than that
lag specification and the results obtained from of crime and arrests, it is meaningful to esti-
the benchmark models, presented in Table mate a crime model where arrests are lagged
2. Below the benchmarkspecificationwe report by one month, and drug use enters with lag
the sums of the estimatedcoefficients and their zero. We estimated these models with TSLS
statisticalsignificance for a variety of lag spec- where contemporaneous drug use is instru-
ifications, described above. mented by drug arrests. Finally, based on
The results are robust. The signs of the Levitt's (1998) discussion of incapacitation
police and arrests are always negative in all and deterrent effects, we used current values
crimes for all lag specifications. Furthermore, of changes in growth rates of murders, as-
arrests are significant in murders, robberies, saults, burglaries, robberies, and burglaries to
burglaries, and motor-vehicle thefts. The sum instrument changes in the growth rates of
of the police coefficients, which was not sig- assault arrests, murder arrests, robbery ar-
nificant in the benchmark model for robber- rests, burglary arrests, and motor-vehicle
ies, turns out to be significant in six out of theft arrests, respectively. Levitt (1998) ar-
seven ad hoc specifications. The poverty in- gues that incapacitation implies that an in-
dicator, AFDC cases, has a positive impact on crease in the arrest rate for one crime will
murders and assaults. The relations between reduce all crimes, and deterrence predicts that
drug use and robberies and burglaries are also an increase in the arrest rate for a particular
consistent with the ones obtained from the crime will generate an increase in other
benchmark model, although the models with crimes as criminals substitute away from the
5, 6, and 9 lags do not yield statistically first crime. This argument suggests that one
significant coefficients. type of crime can be used to instrument a
As discussed in Section III, we estimated different type of arrest. For example, an in-
equation (2) with the inclusion of the contem- crease in robbery arrests would have an im-
poraneous values of arrests and drug use. This pact on burglaries as well, suggesting that the
specification, of course, suffers from the stan- contemporaneous value of burglaries could
dard simultaneity problem, but it provides a act as an instrument for robbery arrests.29In
useful comparison to the results displayed in all these TSLS models the directions of the
Table 2. The results obtained from the inclu- relationships remained the same, although the
sion of current values of arrest and drug use statistical significance of the relationships
were very similar to the ones reported in were reduced considerably. This is not sur-
Table 2. (Theseresults,which arenot reportedin prising given that the variables are employed
the interestof space, are availableupon request.) in first differences, and therefore the instru-
This indicatesthatneithera potentialsimultaneity ments are not highly correlated with the en-
bias due to the inclusionof the contemporaneous dogenous variables.
values,nor a specificationbias due to theirexclu- To put the results into perspective and to
sion, has a significantimpacton the results. clarify the relative importance of deterrence
We also estimatedthe models with TSLS in a and drug use on criminal activity, we used the
number of different ways. First, we instru- sum of the coefficients reported in Table
mented the current value of drug-use growth 2, and calculated the responsiveness of each
with the current value of the growth in drug crime to a 1-percent increase in arrests, po-
arrestsby the NYPD, and the currentvalue of lice, and drug use. The calculated elasticities
criminal arrestgrowth with the sixth or twelfth are reportedin Table 4. We included values of
lag of police growth. Second, we dropped the elasticities only when the baseline equation or
police force from the model, instrumentedthe
currentvalue of criminalarrestgrowth with the
current value of police growth, and instru- 29
As we mentioned earlier, in the same paper, Levitt
mented the current value of drug-use growth finds no significantincapacitationeffect (Levitt, 1998).

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL.90 NO. 3 CORMANAND MOCAN:CRIME,DETERRENCE,AND DRUG ABUSE 601

TABLE4-ARREST, POLICE,AND DRUG-USEELASTICITIES deterrencehypothesis. Murders,robberies,bur-


OF CRIME
glaries, and motor-vehicle thefts decline in re-
Motor-vehicle
sponse to increases in arrests;an increase in the
Murder Robbery Burglary theft size of the police force generates a decrease in
robberiesand burglaries.We find no significant
Arrests -0.34 -0.94 -0.36 -0.40
-0.31 -0.71 -0.39 -0.37
relationships between our drug-use measures
Police -0.53 -0.42 and the violent crimes of felonious assault and
-0.52 -0.41 murder,or between drug use and motor-vehicle
Drug use 0.28 0.06 thefts. On the other hand, we find a positive
0.18 0.04 relationshipbetween druguse and robberiesand
Notes: The first row in each cell reports the elasticity burglaries.We also find that an increase in the
calculated using a zero-growth steady-state scenario for growth rate of poverty, proxied by the number
arrestsand crime. The elasticities reportedin the second row of AFDC cases, generates an increase in the
are calculatedusing the average of the year-to-yeargrowth growth rate of murdersand assaults. These re-
rates for arrests and crimes in the sample as the starting
point.
sults are robust with respect to variations in
model specifications.
The policy implications of our results are
straightforward. To combat serious felony
several of the sensitivity analysis equations crimes, local law-enforcementdecision makers
indicated significance. The first row in each can increase the size of the police force, and
cell reports the elasticity calculated using a they can allocate police in a way thatmaximizes
zero-growth steady-state scenario for arrests deterrenceof serious felonies. It is noteworthy
and crime. The elasticities reported in the that in New York City between 1970 and 1980,
second row are calculated using the average the police force decreased by about one-third,
of the year-to-year growth rates for arrests, but felony arrests increased about 5 percent.
police, drug use, and crimes in the sample as This was accomplishedbecause arrestsfor mis-
the starting point. Table 4 demonstrates, for demeanors, the less serious crimes, decreased
example, that a 10-percent increase in murder almost 40 percent;and arrestsfor violations, the
arrests generates a 3.1- to 3.4- percent reduc- least serious level of offenses, decreased over
tion in murders. Three main conclusions can 80 percent. Thus, police were able to reallocate
be reached from examining this table. First, their increasingly scarce resources to combat
law-enforcement variables consistently deter the most serious crimes.
felonies. Second, law-enforcement elasticities Our results suggest that increased law en-
are smaller than one in absolute value, rang- forcement is a more effective method of crime
ing from -0.3 to -0.9. Third, law-enforce- prevention in comparison to efforts targeted
ment elasticities are always greater than drug- at drug use. This is because, although we have
use elasticities, and drug-use elasticities are statistical evidence that drug usage affects
usually quite small in magnitude. robbery and burglary, the relationship is
weaker in comparison to the one between
VI. Summary and Discussion arrests and crime. For example, a 10-percent
decrease in drug use, proxied by drug deaths,
The purpose of this paper is to provide new generates a 1.8- to 2.8-percent decrease in
evidence on the relationshipamong crime, de- robberies, while a 10-percent increase in rob-
terrence,and druguse. As a potentialsolutionto bery arrests brings about a 7.1- to 9.4-percent
some of the difficult empirical problems en- decrease in robberies.Furthermore,there is no
countered in previous research, we use high- consensus that any criminal-justiceprogram,
frequencytime-series datafrom New York City drug-preventionprogram,or rehabilitationpro-
that span 1970 to 1996. Analyzing data that gramhas resultedin decreasesin druguse of any
cover almost three decades enables us to ob- magnitude.No policy could guaranteea reduction
serve significant variation in five different in drug use by a given magnitude,whereas an
crimes and their determinants. increasein law enforcementis relativelystraight-
The results provide strong support for the forwardto implement.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
602 THE AMERICAN ECONOMIC REVIEW JUNE 2000

REFERENCES Dejong, David N.; Nankervis, John C.; Savin,


Nathan E. and Whiteman, Charles H. "Integra-
Akaike,H. "InformationTheory and the Exten- tion versusTrendStationarityin Time Series."
sion of the Maximum Likelihood Principle," Econometrica, March 1992, 60(2), pp. 423-33.
in Boris N. Petrov and Frigyes Csaki, eds., Ehrlich, Isaac. "Participationin IllegitimateAc-
The Second International Symposium on In- tivities: A Theoreticaland EmpiricalInvesti-
Jbrmation Theory. Budapest: Akademiai gation." Journal of Political Economy, May/
Kiado, 1973, pp. 267-81. June 1973, 81(3), pp. 521-65.
Becker,GaryS. "CrimeandPunishment:An Eco- . "The Deterrent Effect of Capital Pun-
nomic Approach." Journal of Political Econ- ishment: A Question of Life and Death."
omy, March/April1968, 76(2), pp. 169-217. American Economic Review, June 1975,
Becker, Gary S.; Grossman, Michael and Mur- 65(3), pp. 397-417.
phy, Kevin M. "Rational Addiction and the Fisher,Franklinand Nagin,Daniel."Onthe Fea-
Effect of Price on Consumption."American sibility of Identifyingthe CrimeFunctionin a
Economic Review, May 1991 (Papers and Simultaneous Model of Crime Rates and
Proceedings), 81(2), pp. 237-41. Sanction Levels," in Alfred Blumstein, Jac-
Blackburn, Keith and Ravn, Morten 0. "Busi- queline Cohen, and Daniel Nagin, eds., De-
ness Cycles in the United Kingdom: Facts terrence and incapacitation: Estimating the
and Fictions." Economica, November 1992, effects of criminal sanctions on crime rates.
59(236), pp. 383-401. Washington, DC: National Academy of Sci-
Block, M. K. and Heineke, J. M. "A Labor The- ences, 1978, pp. 361-400.
oretic Analysis of the Criminal Choice." Freeman, Richard. "CrimeandUnemployment,"
American Economic Review, June 1975, in James Q. Wilson, ed., Crime and public
65(3), pp. 314-25. policy. San Francisco: ICS Press, 1983, pp.
Chaiken, Jan M. and Chaiken, Marcia R. "Drugs 89-106.
and PredatoryCrime," in Michael H. Tonry Galvin, Jim. "Rape: A Decade of Reform."
and James Q. Wilson, eds., Drugs and crime. Crime and Delinquency, April 1985, 31(2),
Chicago: University of Chicago Press, 1990, pp. 163-68.
pp. 203-39. Goldstein, Paul. "The Drugs/ViolenceNexus: A
Chiricos, Theodore G. "Ratesof Crime and Un- TripartiteConceptualFramework."Journalof
employment:An Analysis of Aggregate Re- Drug Issues, Fall 1985, 15(4), pp. 493-506.
search Evidence." Social Problems, April Granger, Clive W. J. and Newbold, Paul. "Spuri-
1987, 34(2), pp. 187-211. ous Regressionsin Econometrics."Journal of
Cochrane, John H. "A Critiqueof the Applica- Econometrics,July 1974, 2(2), pp. 111-20.
tion of Unit Root Tests." Journal of Eco- Grogger,Jeffrey."Certaintyvs. Severity of Pun-
nomic Dynamics and Control, April 1991, ishment." Economic Inquiry, April 1991,
15(2), pp. 275-84. 29(2), pp. 297-309.
Corman, Hope; Joyce, Theodore and Mocan, H. Grossman, Michael and Chaloupka, Frank. "The
Naci. "Homicides and Crack in New York Demand for Cocaine by Young Adults: A
City," in Melvyn B. Krauss and Edward P. Rational Addiction Approach." Journal of
Lazear, eds., Searching for alternatives: Health Economics, August 1998, 17(4), pp.
Drug-control policy in the United States. 427-74.
Stanford,CA: Hoover InstitutionPress, 1991, Harrison, Lana D. "The Drug-CrimeNexus in
pp. 112-37. the USA." Contemporary Drug Problems,
Cornwell, Christopher and Trumbull, William N. Summer 1992, 19(2), pp. 203-46.
"Estimatingthe Economic Model of Crime Hendry, David F. and Neale, Adrian J. "A Monte
with Panel Data." Review of Economics and CarloStudyof the Effects of StructuralBreaks
Statistics, May 1994, 76(2), pp. 360-66. on Tests for Unit Roots," in Peter Hackl and
Cover, James Peery and Thistle, Paul D. "Time Anders H. Westlund, eds., Economic structural
Series, Homicide, and the DeterrentEffect of change: Analysis and forecasting. Berlin:
Capital Punishment." Southern Economic Springer-Verlag,1991, pp. 95-119.
Journal, January1988, 54(3), pp. 615-22. Hodrick, Robert J. and Prescott, Edward C.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
VOL 90 NO. 3 CORMANAND MOCAN:CRIME,DETERRENCE,AND DRUG ABUSE 603

"PostwarU.S. Business Cycles: An Empiri- eds., Deterrence and incapacitation: Esti-


cal Investigation."Journal of Money, Credit, mating the effects of criminal sanctions on
and Banking,February1997,29(1), pp. 1-16. crime rates. Washington, DC: National
Johnson, Bruce D.; Goldstein, Paul J.; Preble, Academy of Sciences, 1978, pp. 95-1.39.
Edward; Schmeidler, James; Lipton, Douglas; Nelson, Charles R. and Kang, Heejoon. "Pitfalls in
Spunt, Barry and Miller, Thomas. Takingcare the Use of Time as an ExplanatoryVariablein
of business: The economics of crime by her- Regression."Journal of Business and Eco-
oin abusers. Lexington, MA: Lexington nomicStatistics,January1984,2(1), pp. 73-82.
Books, 1985. Nelson, Charles R. and Murray, C. J. "The Un-
Layson, Stephen K. "Homicide and Deterrence: certainTrendin U.S. GDP." Working paper,
A Reexaminationof the United States Time- University of Washington, 1997.
Series Evidence." Southern Economic Jour- Perron,Pierre. "The Great Crash, the Oil Price
nal, July 1985, 52(1), pp. 68-89. Shock, and the Unit Root Hypothesis."Econo-
Levitt,StevenD. "TheEffect of PrisonPopulation metrica,November1989, 57(6), pp. 1361-401.
Size on Crime Rates: Evidence from Prison Peterson, Mark A.; Braiker, Harriet B. and
OvercrowdingLitigation."QuarterlyJournal Polich, Suzanne M. Doing crime: A survey of
of Economics,May 1996, 3(2), pp. 319--51. California prison inmates. Santa Monica,
. "Using Electoral Cycles in Police Hir- CA: Rand Corporation,1980.
ing to Estimate the Effect of Police on Phillips, P. C. B. "Understanding Spurious Re-
Crime." American Economic Review, June gression in Econometrics."Journal of Econo-
1997, 87(3), pp. 270-90. metrics, December 1986, 33(3), pp. 311-40.
. "Why Do Increased Arrest Rates Ap- Purdum,Todd S. "More Police, Eventually:A
pear to Reduce Crime:Deterrence,Incapaci- New York City Plan to Hire Thousands
tation, or Measurement Error?" Economic Doesn't Mean They'clAll SuddenlyAppear."
Inquiry,July 1998, 36(3), pp. 353-72. New YorkTimes,September29, 1990, Sec. I, p.
Long, Sharon and Witte, Ann Dryden. "Current 1:5.
Economic Trends: Implications for Crime Raj, Baldev and Slottje, Daniel J. "The Trend
and CriminalJustice,"in Kevin Wright, ed., Behavior of Alternative Income Inequality
Crime and criminal justice in a declining Measures in the United States from 1947-
society. Cambridge, MA: Gunn and Hain, 1990 and the StructuralBreak." Journal of
1981, pp. 69-143. Business and Economic Statistics, October
Miron, Jeffrey A. "An Economic Analysis of 1994, 12(4), pp. 479-87.
Alcohol Prohibition."Journal of Drug Is- Rasmussen, David W. and Benson, Bruce L. Eco-
sues, Summer 1998, 28(3), pp. 741-62. nomic anatomy of a drug war: Criminaljus-
Miron, Jeffrey A. and Zwiebel, Jeffrey. "Alcohol tice in the Commons.Lanham,MD: Rowman
ConsumptionDuring Prohibition."American & Littlefield, 1994.
Economic Review, May 1991 (Papers and Reaves, Brian A. and Perez, Jacob. "Pretrial Re-
Proceedings), 81(2), pp. 242-47. lease of Felony Defendants, 1992."Bureau of
. "The Economic Case Against Drug Justice Statistics Bulletin, U.S. Department
Prohibition."Journal of Economic Perspec- of Justice,Washington,DC, Novemnber1994.
tives. Fall 1995, 9(4), pp. 175-92. Reuter, Peter; MacCoun, Robert and Murphy,
Mocan,H. Naci. "StructuralUnemployment,Cy- Patrick. Money from crime: A study of the
clical Unemployment, and Income Inequali- economics of drug dealing in Washington,
ty." Review of Economics and Statistics, D.C. Santa Monica, CA: Rand Corporation,
February1999, 81(1), pp. 122-34. June 1990.
Myers, Samuel L., Jr. "Estimating the Economic Rudebusch, Glenn D. "The Uncertain Unit Root
Model of Crime: Employment Versus Pun- in Real GNP." American Economic Review,
ishment Effects." QuarterlyJournal of Eco- March 1993, 83(1), pp. 264-72.
nomics, February1983, 98(1), pp. 157-66. Saffer, Henry and Chaloupka, Frank. "The De-
Nagin, Daniel. "GeneralDeterrence:A Review mand for Illicit Drugs." National Bureau of
of the EmpiricalEvidence," in Alfred Blum- Economic Research(Cambridge,MA) Work-
stein, Jacqueline Cohen, and Daniel Nagin, ing Paper No. 5238, August 1995.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions
604 THEAMERICANECONOMICREVIEW JUNE 2000

Sims, Christopher A. and Uhlig, Harald. "Under- van Ours, Jan C. "The Price Elasticity of Hard
standing Unit Rooters: A Helicopter Tour." Drugs: The Case of Opium in the Dutch
Econometrica, November 1991, 59(6), pp. East Indies, 1923-1938." Journal of Political
1591-99. Economy,April 1995, 103(2), pp. 261-79.
Stock, James H. and Watson, Mark W. "Variable Witte, Ann Dryden. "Estimatingthe Economic
Trendsin Economic Time Series."Journal of Model for Crime with Individual Data."
Economic Perspectives, Summer 1988, 2(3), Quarterly Journal of Economics, February
pp. 147-74. 1980, 94(l), pp. 57-84.
Tauchen, Helen; Witte, Ann Dryden and Griesinger, Zivot,Eric and Andrews,DonaldW. K. "Further
Harriet. "Criminal Deterrence: Revisiting the Evidence on the Great Crash, the Oil-Price
Issue with a BirthCohort."Reviewof Econom- Shock, and the Unit-Root Hypothesis."Jour-
ics and Statistics,August 1994, 76(3), pp. 399- nal of Business and Economic Statistics, July
412. 1992, 10(3), pp. 251-70.

This content downloaded from 185.31.194.107 on Sat, 28 Jun 2014 18:17:05 PM


All use subject to JSTOR Terms and Conditions

Potrebbero piacerti anche