Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
A Historical Perspective
John Gerring
Boston University
Department of Political Science
232 Bay State Road
Boston MA 02215
jgerring@bu.edu
Philip Bond
University of Pennsylvania
Wharton Finance Department
Steinberg Hall-Dietrich Hall
3620 Locust Walk
Philadelphia PA 19104-6367
pbond@wharton.upenn.edu
William T. Barndt
Princeton University
Department of Politics
Corwin Hall
Princeton NJ 08540-1012
wbarndt@princeton.edu
A Historical Perspective
ABSTRACT
Recent studies appear to show that democracy has no robust association with economic growth.
Yet, all such work assumes that the causal effect of democracy can be measured by a country’s
regime status in a particular year (T), which is correlated with its growth performance in a
subsequent period (T+1). We argue that democracy must be understood as a stock, rather than level,
measure. That is, a country’s growth performance is affected by the number of years it has been
democratic, in addition to the degree of democracy experienced during that period (running from
contemporary, variable -- with the assumption that long-run historical patterns may help us to
understand present trends. We speculate that these secular-historical influences operate through
four causal pathways, each of which may be understood as a type of capital: physical capital, human
capital, social capital, and political capital. This argument is tested in a cross-country analysis and is
2
Does regime type affect economic performance?1 The predominant view is that democracy has
either a negative effect on GDP growth or no overall effect. Countries with authoritarian political
systems are thus predicted to grow as rapidly as democracies, perhaps even faster. To be sure,
democracy may have some positive indirect effects – e.g., greater stability or more extensive property
rights. However, the econometric evidence suggests that these positives are balanced by negatives
such that the net effect of democracy on growth performance cross-nationally over the last five
decades is negative or null (e.g., Barro 1996; Feng 1997, 2003; Krieckhaus 2004; Kurzman, Werum,
Burkhart 2002; Przeworski and Limongi 1993, 1997; Przeworski et al. 2000).3 For the most part,
case study approaches to this question confirm the results of crossnational growth empirics (e.g.,
Chan 2002; Rodrik 2003; Woo-Cumings 1999). Thus, although most of the rich countries in the
world are democratic, the direction of causality is unclear. One must keep in mind that many rich
democracy is a luxury to be enjoyed only by countries rich enough to afford it. This, indeed, is a
common argument among authoritarian leaders in the developing world.4 Others might counter
that democracy is worth the price of economic development because civic equality, political
participation, and civil liberties are intrinsically important (Shapiro and Hacker-Cordon 1999).
However, this argument is unlikely to inspire enthusiasm among those currently forced to live on
one dollar a day, roughly one fifth of the world’s population. In summarizing the results of a recent
poll in eighteen Latin American countries the authors note that “the preference of citizens for
democracy is relatively low; many Latin Americans value development above democracy and would
even stop supporting a democratic government if it proved incapable to resolving their economic
problems” (UNDP 2004: 6). Indeed, the demand for “food first” sounds more sensible than the call
3
In this paper, we show that the skeptical view of democracy and development rests on a set
of econometric findings in which democracy is treated as a more or less immediate cause. This
year’s level of democracy is thought to influence growth performance in the following period
(usually a decade or two). We argue that this is an unrealistic assumption. If democracy matters to
growth it is more reasonable to assume that this effect stems from a country’s regime history, not
Democracy is thus best considered as a stock, rather than level, variable. The distant past has
contemporary effects.
We begin by constructing a prima facie case for a historical understanding of democracy and
its relationship to economic growth. Second, we discuss the definition and measurement of the key
concept, democracy. Third, we discuss the method that will be employed to test our proposition,
the time-series cross-section (TSCS) regression. Fourth, we discuss the results of these regression
tests. Fifth, we discuss the “democratic growth effect” – its magnitude, its policy significance, and
directions for future research. We conclude with a discussion of the possible merits of a historical
Work on democracy and growth has usually focused on the possible causal effect of a country’s
contemporary level of democracy on its subsequent growth performance. The subsequent period at
issue might be a single year or some multi-year period (e.g., a decade); in any case, scholars have
conceptualized the problem as the effect of regime type at time T on growth rates at time T plus
some specified period. These studies look forward, but not backward, in time.
4
We propose to consider regimes as historically informed phenomena rather than as
contemporary variables. This means looking backward, as well as forward (via lagged predictors), in
time. In particular, it means measuring a country’s democratic accumulated stock rather than its level
of democracy at a particular point in time. The core insight is that institutional effects unfold over
Nowhere does this seem more likely than when one is considering the causal effects of a
regime type. Democracy and authoritarianism construct deep legacies, extending back several
decades -- perhaps even centuries. Therefore, it is these historical legacies rather than their
contemporary status (as democratic or authoritarian), that ought to be of central concern if we wish
to understand the causal effect of a regime type on a variety of current outcomes – social, cultural,
political, or economic.
measured by GDP per capita growth. Thus, we are impelled to think carefully about the causal
mechanisms that might link a country’s regime history with its current growth performance. Our
Capital, in common usage, implies a fungible resource that accumulates over time (given
favorable circumstances), and promises increased returns in the future. An investment today should
bring a higher yield at some later date. If a democratic regime endures, we argue that it is likely to
foster four types of capital: physical capital, human capital, social capital, and political capital. Extant
studies indicate that all four of these forms of capital have positive impacts on growth performance.
Thus, we anticipate that the longer a country remains democratic, the greater its physical, human,
social, and political capital, and the better its growth performance. These relationships are depicted
in Figure 1. Since political capital is the newest concept in this panoply of capital-isms, our
5
[Figure 1 about here]
The extent to which physical capital is equally distributed throughout a society, is now commonly
regarded as an important component of long-run economic growth (Besley and Burgess 2000;
Sokoloff and Engerman 2000; but see Forbes 2000). Thus, insofar as democracy is successful in
redistributing wealth – either through progressive taxation, social policies, land redistribution, or
simply by opening up markets and institutions in civil society to previously excluded groups (Lenski
1966; Lipset 1959; Meltzer and Richards 1981; Muller 1988) – it seems reasonable to expect that the
longer this regime type is in existence, the greater its aggregate effect will be on the achievement of
social equality, and hence on growth. With the exception of a few dramatic instances of land reform
and income distribution in a society change slowly, if they change at all (Banerjee 1999; de Janvry et
al. 2001). Thus, while democracy per se may have little effect on equality, a prolonged experience
with democratic rule may have important consequences (Gradstein and Milanovic 2000; Tavares and
Wacziarg 2001).
Consider, as well, the role of human capital, which endogenous (“new growth”) theory
identifies as a strong influence on growth performance (Barro 1997; Sala-i-Martin 2002). It seems
capital, as measured by declining fertility rates and improvements in education, public health and life
expectancy. Political elites in a democracy have electoral incentives to avoid famine and improve the
quality of life for the least advantaged, incentives which are not present, or are present to a much
smaller degree, in authoritarian systems. Studies have shown, almost uniformly, that democratic rule
translates into improvements in a society’s human capital resources (Baum and Lake 2003; Dreze
6
and Sen 1989; Feng 2003; McGuire 2005). Because of the (literally) long-lived nature of human
capital, the longer a democracy is in place the more pronounced its impact on the level of human
capital.
Finally, many commentators have argued that social capital also contributes to a society’s
economic well-being. For example, in a recent careful study, Guiso et al. (2004) present evidence
that social capital enhances financial development (which is, in turn, generally regarded as growth-
enhancing). Since democracy has been considered a precondition for the development of social
capital (Putnam 2002; Warren 1999), it stands to reason that a long-term democracy would have
long-term effects on the development of social capital. Again, these observations suggest that
democracy’s effect on growth may become more pronounced over time, as a country’s stock of
POLITICAL CAPITAL
In addition to physical, human, and social capital, we argue that established democracies create a
species of capital that is explicitly political. Just as deferred consumption generates physical capital,
which in turn contributes towards output, so should a country’s political experience today affect
acknowledge the intrinsic ambiguity of the core concept. “Capital” poses severe problems of
conceptualization and measurement by virtue of its theoretical abstraction. It is, after all, a capability
-- a potentiality. As such, it cannot be directly measured. Even so, the ambiguities of capital-based
theories may be redeemed by the theoretical leverage that they offer. In particular, they allow us to
think about the role of social institutions through time. This, we believe, is a significant theoretical
7
Politics, too, may be capital-ized. Recall that physical capital is usually measured in monetary
terms (e.g., accumulated investments), human capital by the spread of education and other job-
related skills, and social capital by patterns of social interaction (although there are a variety of other
possible measures [Knack and Keefer 1997; Woolcock 1998]). We propose that political capital may
be operationalized by various measures of the relative health of a polity – e.g., trust in political
institutions, low levels of corruption, bureaucratic capacity, political consensus, political stability, the
wisdom and far-sightedness of political leaders, and so forth (Englehart and Gerring 2005). These
resources, like other sorts of capital, accumulate through time and may be drawn upon for a range of
purposes.
In some respects, our use of the term political capital is an extension of the colloquial usage,
where it is usually understood as a resource attached to a particular figure -- someone who is said to
have accumulated (or lost) political capital by some action or set of actions. Granted, the parallel is
not exact. An individual’s political capital may be nothing more than a quid pro quo, a favor done
for a friend that is “banked” for some period of time. However, an individual’s political capital may
also refer to something more complex: a pattern of behavior that establishes a reputation for fidelity
and competence which translates, in turn, into a relationship of extended trust. (It is at this point
that a representative becomes a trustee rather than simply a delegate and the transactional costs
usual to politics are lowered -- since explicit quid pro quo agreements between actors are no longer
required.) The reputational quality of political capital plausibly operates for institutions as well as for
persons. But this is not a key point of the argument, for there are obvious differences between
How, then, might democratic political arrangements foster political capital in a country and,
ultimately, enhanced economic growth? To begin with, we surmise that democracy sets in place a
learning process in which policies and institutions are refined. Politicians learn how to govern,
8
electors learn how to vote, and politicians learn how to respond. Second, a sustained experience of
democracy enhances stability and consensus. Both of these processes, which we discuss in much
greater detail below, should have a strong influence on economic performance over the long haul.
To be sure, our notion of political capital overlaps somewhat with those of human and social
capital. Learning about policymaking is in a sense simply a special form of human capital
appreciation; while the emergence of consensus is reminiscent of some conceptions of social capital.
What we have in mind, however, is knowledge and consensus that can only be acquired through
political exposure. It is the nature of the “investment” required to generate political capital that
BETTER POLICIES AND INSTITUTIONS. Consider, first, the realm of economic policy and good
governance more generally. There is no question that public policies have a strong impact on the
growth performance of nation-states, although there remain serious disagreements over which
policies, precisely, are most growth-enhancing. For purposes of discussion we shall assume that
large budget deficits, high inflation, high tariff barriers, and heavy regulatory burdens usually have
negative consequences over the long-term. This does not mean that the state should withdraw from
the marketplace; it means, rather, that economic policies should be “market-augmenting” (Azfar and
Cadwell 2002). The security of property rights depends upon the intelligent engagement of the state
(North 1990, 1994). Similarly, good institutions – as measured by low corruption, high bureaucratic
capacity, and other indicators – should foster stronger growth performance, all other things being
equal (Ades and Di Tella 1997; Keefer and Knack 1997; Mauro 1995). With these assumptions
granted, we may now proceed to a discussion of the role of regime type in economic policymaking.
There is good reason to suppose that the length of a country’s experience with democratic
rule is relevant for the intricacies of economic policymaking. First, the advent of democracy is
9
often associated with the installation of a new set of political elites, perhaps even a wholesale
replacement of politicians and bureaucrats that previously staffed the central government (Gifford
and Louis 1982; Jens-Hesse 1993). Let us take as given that political experience matters for effective
policymaking and that political experience is accumulated over time in a stable polity. (We imagine
that this is particularly the case with respect to economic policy, which tends to be more complex
and less amenable to intuition than other areas of public policy.) Politicians and bureaucrats in an
established democracy are likely to have served a long period of apprenticeship, either in an
established party or a government agency; they will have moved up within some institution’s ranks.
This does not guarantee that merit will govern appointment and promotion, but it does guarantee
that whoever occupies the commanding heights of government will be reasonably seasoned. It
follows that the longer a democratic polity is in place, the more experience its elites will have.
By itself, learning by policy-making elites suggests only that long-lived regimes will tend to
produce better policies than short-lived ones -- regardless of whether the regime in question is
democratic or autocratic. However, the very existence of an electorate may dramatically increase the
importance of learning in democratic regimes. That is, not only must governing politicians learn
what constitutes good policy. Voters must also learn to recognize good policy. There may even be
a third stage of learning in which politicians learn that voters have learned to distinguish good
The most important lesson to be learned by democratically elected elites is that growth
performance matters for their political future and their historical legacy, as well as for the future of
their party or chosen successors. In new democracies, politicians frequently adopt short-term
policies intended to pay off political supporters and stimulate the economy during the electoral
season (Dornbusch and Edwards 1991). However, in established democracies, if elites and voters
have experienced a series of electoral and economic cycles and if political power is channeled into
10
well-established organizations (parties, bureaucracies, and so forth), longer time-horizons may
prevail. Voters may have directly experienced the effects of populist economic policies; if so, they
are likely to be skeptical of claims that soaking the rich, inflating the economy, abrogating debt
agreements, or massive expropriation of property will enhance their livelihoods (Remmer 1991;
Weyland 2002). This certainly seemed the case during the disastrous presidency of Fernando Collor
de Mello in Brazil (Rosenn & Downes 1999). Political elites, by the same token, will have had
Economic growth, so long as it is not monopolized by a few, is a salient political issue in most
national campaigns (Lewis-Beck and Stegmaier 2000). Thus, leaders may be inclined to impose
sacrifices over the short-term to facilitate stronger growth performance over the course of their
administration (Stokes 2001, 2002) and in the long run. Political leaders in a highly institutionalized
polity must be concerned with their party’s economic record in subsequent elections. The more
institutionalized the polity, under conditions of democratic accountability, the longer the time-
More generally, democracies institute a learning process among elites and masses in which
economic performance and electoral contests offer periodic corrections. Of course, this learning
process is driven not only by actual rewards and punishments meted out by the electorate. It is also
driven by anticipated rewards and punishments. In Sartori’s (1987: 152) well-chosen words: “Elected
officials seeking reelection (in a competitive setting) are conditioned, in their deciding, by the
anticipation (expectation) of how electorates will react to what they decide. The rule of anticipated
reactions thus provides the linkage between input and output, between the procedure (as stated by
Schumpeter) and its consequences.” Sartori (Ibid.) refers to this as a “feedback theory of
democracy.”
11
The longer a democracy has been in place, the further this learning curve will extend, all
other things being equal. We expect a slow transition away from a populist style of politics and
democratic histories should institute better (i.e., market-augmenting) policies than transitional
democracies or authoritarian regimes. These policies, in turn, should have strong effects on
aggregate growth performance. Empirical grist for this theoretical argument can be found in a series
of studies indicating that a country’s democratic history is a strong predictor of low corruption,
bureaucratic quality, property rights, and other good-governance outcomes (Gerring and Thacker
POLITICAL CONSENSUS AND STABILITY. Thus far, we have outlined how a long-lived democratic
regime might enhance growth through its effect on capital allocation and formation, and through
society’s learning about what constitutes good policies. However, a large literature on democratic
overload posits that democracy engenders costly and destabilizing power struggles among various
subgroups (e.g., Crozier et al. 1975). As we argue below, this cost of democracy is one that may well
newly democratizing countries – particularly when those countries are poor, ethnically divided, or
where the question of nationality is open to question (Chua 2003; Fein 1995; Mousseau 2001; Snyder
2000). Such countries are burdened with a surfeit of expectations, accumulated over many years.
Citizens have been told to expect great achievements from self-government, and they generally
expect these goods to materialize in a hurry. It is the fashion of political leaders during the long and
dangerous struggle for democracy to over-promise. Thus, when the transition finally occurs it is
greeted with enormous hopes and extravagant expectations. Transitions offer little preparation for
12
the hum-drum nature of everyday politics. Democracy experienced is never quite the same as
democracy envisioned (see, e.g., O’Donnell and Schmitter [1986] on desencanto). The process of give-
and-take among competing priorities may seem to barter away what had initially been gained, a
corruption of the pure democratic ideal into brokerage politics. Needless to say, such
disillusionment does not augur well for political stability and consensus.
previously quiescent, and perhaps actively repressed, groups. These might be lower classes, excluded
ethnic or racial groups, or some other category of out-group (Eckstein 1989; Escobar and Alvarez
1992; Stepan 1989; Tarrow 1998). While beneficial in the long-run, the short-run effects of such
mobilizations from below may be destabilizing and may have adverse effects on the investment
climate, as in Bolivia during the early 1980s (Haggard & Kaufman 1995: 184-86).
If democracy survives its tumultuous youth, however, we anticipate that the extreme nature
of political conflict will moderate over time. A democratic political system is, by definition, open to
the inclusion of all sizeable social groups and interests. Once granted a taste of political power,
elites at the head of radical social movements may find it in their interest to join an existing party or
coalition. Moreover, the relatively open nature of deliberation in an established democracy may
diminish the appeal of conspiracy theories, which tend to flourish in the fog of authoritarian rule.
Generally speaking, social and political consensus should become more likely as democracies
become more open to the participation of citizens and as citizens accumulate experience under
democratic rule. Whatever centripetal tendencies are inherent in democracy are more likely to be in
evidence when those democratic arrangements have been in operation for some time. A regime is
unlikely to elicit much legitimacy among out-groups until they have been included in an elected
government. This requires at least one transfer of power, and probably awaits the outcome of
several electoral contests (Przeworki 1991). For these reasons, the thesis of democratic overload is
13
much more compelling when applied to new democracies than when applied to old. New
to be more restrained. In particular, the norm of incremental change is more likely to be accepted.
Although the greatest threats to political stability presumably arise within segmented
societies where distrust among various social groups is high it is not necessary to presuppose a
1999). Somehow, people must agree upon, or at least agree to respect, the imposition of society-
wide policy solutions that impose uneven costs and benefits. In an authoritarian setting, these
coordination problems are solved by the sovereign, Hobbes’ Leviathan. Rule by coercion, insofar as
it is successful, can be imposed without loss of time and without negotiation. The threat of force is
democratic polity must institutionalize procedures for negotiation (formal or informal) among rival
Thus, the problem of overload, insofar as it exists, probably does not arise from institutional
sclerosis (Olson 1982), but rather from insufficient institutionalization (Huntington 1968). When
political institutions are new and few, one rightly anticipates political conflict. Over time, however,
(1970) words, “habituation” – occurs through multiple arenas including political parties, legislatures,
committees, and electoral contests. It is an iterated process, and can only be conceptualized through
time. Institutionalization, therefore, should coincide with the achievement of greater political
stability, consensus, and social trust. Each of these factors presumably has positive implications for
14
EMPIRICS
Our theoretical expectation is that the length of time a country is democratic – and, secondarily, the
degree of democracy found in that country – is likely to influence its economic performance. In order
to transform this expectation into a testable hypothesis we must face the difficult question of how to
conceptualize and measure the key theoretical concept, democracy. Evidently, democracy is a broad
concept, encompassing a wealth of possible attributes (Sartori 1987; Collier and Levitsky 1997). Any
plausible definition must not violate patterns of ordinary usage and must therefore integrate the core
meaning of democracy -- rule by the people. At the same time, in order to be useful a definition
must strive to capture the essentials of whatever theoretical argument the concept is intended to test.
Concepts are active tools, in addition to serving as (neutral) containers of language. Thus, our
definition, while resonating with ordinary usage, is not intended to capture all nuances of
Since our theoretical question is whether democracy has a positive effect on economic
performance, our definition of this difficult concept must attend to those aspects of democracy that
seem most likely to affect economic performance. So viewed, democracy is perhaps best
understood, following Joseph Schumpeter (1942/1950: 269), as “that institutional arrangement for
arriving at political decisions in which individuals acquire the power to decide by means of a
competitive struggle for the people’s vote.” This feature, rather than the closeness of the vote
(Vanhanen 1990) or actual turnover among ruling elites (Przeworski et al. 2000), seems mostly likely
to affect economic performance. The key question is whether the current in-group stands a realistic
chance of becoming an out-group in the near future. Competition may remain a potentiality; it need
not be actualized over the short-term. Indeed, members of a polity may freely choose to re-appoint
15
a ruling group over many years, and the differentials between winning and losing parties may be
quite large. Botswana, where the Botswana Democratic Party (BDP) has ruled since independence,
exhibits both of these features but is still, according to our reckoning, a country in which the ruling
elites face potential rejection at each election. Thus, we shy away from simple, mechanistic
indicators of democracy.
matters that are sometimes difficult to measure – they are, indeed, analogous to the equally fuzzy
notion of an “open” market. (How does one know for sure that a market is unconstrained?) With
respect to political markets, the following issues seem critical, but are by no means comprehensive.
Elected officials should be sovereign; non-elective bodies such as a military junta, monarchy, or
caste must not exercise real power behind the scenes. There should be regular elections, and these
elections must be open to all citizens (both as candidates and electors), subject to minor restrictions
having to do with age or residency. Suffrage should be broad, though not necessarily universal.
Political liberties should be extensive, particularly as pertains to the task of political organization and
campaigning. Election resources, including access to money, the media, and voters, should be fairly
allocated. Each of these is perhaps best understood as a matter of degrees (can one envision a
perfectly fair allocation of election resources?). This inclines us towards a continuous, rather than
Our central hypothesis concerns the temporally-dependent role of democracy, its long-term
potential to foster economic growth. Thus, we wish to capture not only differences in degree of
propose to measure democracy as a stock, rather than level, variable; that is, to measure the
accumulation of democratic experience, rather than a snapshot view of democracy.6 Since the
16
returns to democracy are presumably not linear – at some point the value of an additional year of
democracy should level off – we transform the democracy (stock) variable by its natural logarithm.
We employ the Polity IV “Polity2” variable as our principal measure of democracy (Marshall
and Jaggers 2000). The variable is constructed by measuring the extent to which democratic or
authoritarian “authority patterns” are institutionalized in a given country. It takes into account how
the executive is selected, the degree of checks on executive power, and the form of political
competition. This indicator is highly sensitive (it employs a 21-point scale), and offers extensive
country coverage (all sovereign polities excepting micro-states) and good historical coverage.
Moreover, it allows us to consider both the degree and duration of democracy in any given country-
year. The Polity dataset, however, imposes two serious costs. First, the rules used to create the key
variable, Polity2, are dizzyingly complex. The Polity User’s Manual makes a valiant effort to
explicate coding procedures, but the methods remain rather difficult to unpack. Second, there are
serious questions regarding measurement error in the index (Bollen and Paxton 2000; Bowman et al.
2004; Munck and Verkuilen 2002; Treier and Jackman 2003). Granted, questions might be raised
with respect to all extant, and all conceivable, democracy indices (see previous citations). Polity2 is
no worse than the rest, and probably better than the average. It is, indeed, the industry standard,
owing largely to the strengths noted above. Reassuringly, the Polity2 variable correlates highly with
other existing measures of democracy.7 There is no reason to suspect systematic errors in this index
To create a stock measurement of democracy from this variable we simply add up each
country’s Polity2 score from 1900 to the present year. The year 1900 is chosen as a threshold year
longer restricted to the US and a few European states), b) in which it is not unreasonable to assume
17
a causal relationship between democracy and growth, and c) in which the data exists to test such a
relationship.
Because the historical component of this index weighs heavily on our understanding of the
concept and because the Polity dataset ignores non-sovereign states in its coding procedures, we
supplement the Polity2 coding with our own coding of several nation-states that were previously
part of contiguous empires. The procedure is as follows. For each year that a nation-state belonged
to an imperial power it receives the same Polity2 score as its imperial ruler; e.g., Estonia receives the
same score as the Soviet Union from 1941 through 1990. We use this procedure only for nation-
states that were contiguous with the empire to which they belonged. We assume that contiguous
colonies are likely to be governed in the same manner as the imperial power itself, a dynamic less
To correct for Polity2’s exclusion of micro-states, an exclusion that might also bias our
sample, we impute democracy scores for these excluded cases using other democracy indices that are
conceptually and empirically close to the Polity2 measure: a) the Freedom House Political Rights
indicator,9 b) Ken Bollen’s Liberal Democracy variable (Bollen 1993), c) Tatu Vanhanen’s
Competition variable (Vanhanen 1990), d) Arthur Banks’s Legislative Effectiveness variables (I and
II), and e) Banks’s Party Legitimacy variable (Banks 1994). These various measures of democracy
take into account the degree to which citizens can participate freely in the political process, the
competitiveness, and the degree to which the legislature effects public policy.
These various additions to the core Polity2 variable increase the number of potential cases in
our analysis only slightly – by about 500 cases (compare models 1 and 2 in Table 2).
ANALYSIS
18
The relationship between democracy and growth may be tested in many ways. We take as our point
of departure a time-series cross-sectional format, with countries as our units of analysis. Even so,
this scarcely limits the methodological field since any of the multitudinous approaches employed in
current growth regressions might also be applied to this particular question (see Temple 1999 for a
recent review). The researcher faces choices about what time-intervals to consider, how to correct
for serial and spatial autocorrelation, and how to resolve issues of specification, simultaneity, and
Fortunately, there is general agreement about how to measure the dependent variable,
economic growth, which is usually understood as the percentage change in GDP per capita. We
employ the World Development Indicators (World Bank 2003) growth variable, measured in
constant dollars. Data for the 1950s is imputed using the Penn World Tables (PWT [Summers and
Heston 1991) 6.1 dataset (Chain index, constant dollars). Our choice of the WDI dataset as the
primary data source for indicators of country growth is motivated by two concerns. First, WDI
country coverage is considerably larger than offered by the PWT dataset. Second, for various
reasons explored by Nuxoll (1994) and Temple (1999: 118-19), the WDI indicator is probably the
Since data comprising the dependent and independent variables of interest are compiled at
annual intervals, and since significant changes in both right- and left-hand variables occur from year
to year, it makes sense to employ annual data in this analysis. Not to do so constitutes a waste of
information and is prone to the aggregation problem (how long should intervals of aggregated data
be?). So long as patterns of temporal autocorrelation and period-specific effects can be corrected,
we see no justification for aggregating data over five- or ten-year periods (or longer), as is sometimes
done. The unit of analysis in all the following data tables is therefore the country-year. (In the
19
appendix we offer a series of additional tests that employ data aggregated over 5- and 10-year
periods.)
assumes a one-period (AR1) temporal disturbance whose value is calculated on the basis of the
Durbin-Watson statistic. (Other methods of calculating rho do not have an appreciable effect on the
results of the regressions.) An alternative approach, with much to recommend it in certain situations,
employs a lagged dependent variable in order to de-trend the data (Beck 2001; Beck and Katz 1995).
We find little difference in substantive results when this alternative method is employed (see
Appendix). Since the interpretation of results is considerably less intuitive when a lagged dependent
variable is introduced on the right side (coefficients represent the change in trend of the dependent
The most serious challenge to cross-country analyses of distal causal relationships is the
problem of simultaneity. Suppose, for instance, that a growth regression identifies Factor X as
positively contributing to growth. Almost invariably the skeptic can plausibly argue that this finding
is a consequence of the existence of some unmeasured Factor Z that impacts both growth rates and
Factor X. In short, other (unmeasured) factors that are correlated with democracy may account for
superior growth performance, rendering our results spurious. Indeed, it is plausible to suppose that
countries that are able to maintain a high level of democracy over a long period of time are also
blessed with other advantages – good institutions, good resources, and so forth. Needless to say, it
is difficult, perhaps even impossible, to measure all such country-specific (unit) effects. (This
supposition is borne out by various tests of spatial autocorrelation comparing the results of fixed
and random effects regressions -- e.g., Breusch and Pagan Lagrangian multiplier tests and Hausman
tests.)
20
To control for possibly severe unit effects (spatial autocorrelation) we employ a fixed-effect
format in all analyses. Although this precludes examining variation across countries, a loss of
explanatory leverage, it effectively removes many of the specification problems that plague cross-
country studies. Note that the fixed-effect format imposes a unique intercept for each country.
Factors that are for the most part constant across the time-period of interest (1950-2000), such as
geography, culture, and ethnicity, are eliminated by this research design. Thus, our results will suffer
from omitted variable bias only if the change in growth rate and the change in democracy stock are
both driven by some other (unmeasured) factor. We attempt to control for such factors in Table 1
(see below).
considered as a level (rather than stock) variable. Democratic/authoritarian changes from year to
year may be contingent upon contemporaneous growth performance, creating circularity in the
argument. However, endogeneity is considerably less problematic when the measure of democracy
is itself longitudinal, extending 50-100 years back in time. As a further precaution, we lag democracy
(as well as all control variables) one year. Longer lags -- of 10 and 20 years respectively – are
introduced for our democracy variable in later tests (see Table 2).
Specification problems pervade all cross-country growth regressions (Levine and Renelt
1992; Sala-i-Martin 1997; Temple 1999). While the fixed-effect format handles the problem of
invariant controls (e.g., dummy variables registering a country’s location, its culture, its degree of
ethnic heterogeneity, its colonial history, and so forth), it does nothing to control for factors that
might vary through time. Our approach is to include in our benchmark model any factor that is
statistically significant in most robustness tests (conducted in other studies) and agreed-upon by
most writers on the subject. As it happens, this reduces the field to exactly one variable: GDP per
capita (natural logarithm), intended to capture the convergence hypothesis -- richer countries grow
21
slower, ceteris paribus.10 Recall that we are interested in the effect of democracy on a country’s
growth rate given its current level of economic wealth; so it makes sense to include this important
control, assuming that there are, in fact, convergence effects. This variable is drawn from the WDI
dataset (World Bank 2003), with missing cases from the 1950s imputed from the PWT 6.1 dataset
Other controls are less obvious by virtue of their possibly endogenous relationship to
democracy, their lack of robustness, or their theoretical status. At the same time, it is vital that we
test as comprehensive a set of alternate controls as possible. These controls must encompass not
only those identified by the prodigious literature on economic growth but also those factors that
might affect the simultaneity problem discussed above. These control variables are introduced
seriatim and then in a series of “full” regressions, so as to test their individual and collective effects
where gi,t is the growth rate in country i in year t, αi is a country-specific intercept term,
POLITYSTOCK i,t is the sum of Polity2 scores since 1900 (as described above) for country i as of
year t, Zi,t is a vector of control variables for country i in year t, and εi,t is the error term. As
discussed, we use an AR1 specification, so that εi,t = ρεi,t-1+υ i,t for some constant ρ, with υ i,t
independently and identically distributed over both countries i and dates t. The main object of
interest is the parameter β, which measures the relationship between the stock of democracy and the
growth rate.
It may be helpful to observe that equation (1) can be equivalently expressed in first-
difference form as
POLITYLEVELi ,t −1
g i ,t − g i ,t −1 = β ln(1 + ) + γ ( Z i ,t −1 − Z i ,t − 2 ) + (ε i ,t − ε i ,t −1 ) (2)
POLITYSTOCK i ,t − 2
22
That is, the change in a country’s growth rate is related to the level of democracy in the proceeding
year. In other words, democracy may only have a detectable effect on growth rates if it persists for a
RESULTS
Table 1 presents a series of specification tests probing the performance of the main independent
variable of interest, democracy. We begin with a level measurement of democracy drawn from the
Polity IV dataset, as discussed above. We include only one all-purpose control in this reduced-form
equation, GDP per capita. The results confirm the standard finding: democracy has no statistically
significant effect on economic growth. This non-relationship is robust across a wide range of
democracy indicators (not shown). It matters not how one measures the level of democracy in a
given year; it still has no effect on subsequent economic performance. There is some variation in
effect depending upon the time-period of the analysis (annual, 5-year, or greater). However, none of
We proceed in the remaining models (and in the remaining tables), to investigate democracy
as a stock variable, as described above. Model 2 presents what we regard as our benchmark model,
with only one control variable (replicating the specification of model 1). Model 3 drops that control
variable, demonstrating that Polity stock enhances growth performance even when convergence
effects are ignored. Model 4 adds a variable measuring trade-weighted growth per capita (each
country is assigned the mean value of the growth rate of all other countries in the world in that year,
weighted by their bilateral trade with the country in question). Model 5 adds dummy variables for
each year in the dataset (1950-2000), minus one. Because of the loss of degrees of freedom that this
23
imposes, and because it has only minor effects on the Polity stock coefficient, these controls are not
Model 6 includes a whole series of time-varying controls that are a) common in the growth
literature, b) available for a large number of country-years, and c) might help to control for the
simultaneity problem that we have discussed. These include: Inflation, understood as annual
percent change in consumer prices (natural logarithm; World Bank 2003); Investment, understood as
the share of real GDP comprised by investment (PWT 6.1 [Summers and Heston 1991]); Foreign
direct investment, understood as net inflows as a share of GDP (World Bank 2003); Openness,
understood as imports and exports as a share of GDP (PWT 6.1 [Summers and Heston 1991]);
Government consumption, understood as the government share of real GDP per capita (PWT 6.1
[Summers and Heston 1991]); Population growth (World Bank 2003); Years independent (based on
data compiled by Robert Jackman [personal communication]; see also Jackman 1993); Regime
durability, understood as the number of years since the last three-point change in the composite
Polity2 score (Polity IV); Instability, including assassinations, general strikes, guerilla warfare,
government crises, purges, riots, revolutions, and anti-government demonstrations (these variables,
drawn from the Banks [1994] dataset, are added together to form a composite index [construction of
index by the authors]); Social conflict, including civil violence, civil war, ethnic violence and ethnic
war (these variables, drawn from Marshall et al. [2003], are added together to form a composite
index); Life expectancy (World Bank 2003), Illiteracy, as a logged function (World Bank 2003); a
trend variable (1950=1, 1960=2, . . .); an oil shock dummy (1950-1973=0, 1974-2000=1); and
decade dummies (not shown).11 In additional analyses (not shown), we introduced these control
variables seriatim into the benchmark equation so as to make sure that their individual effects did
24
Models 7-9 add a series of variables that are highly correlated with one another and reduce
the sample size considerably. For this reason, these controls are tested seriatim, rather than all
together. Expropriation measures the risk that the government will seize private property (Political
Risk Services 2004). Repudiation measures the risk that the government will disclaim its contracts
with the private sector (Political Risk Services 2004). Corruption measures the overall spread of
corruption in government (Political Risk Services 2004). All these variables are constructed so that
high scores indicate better performance (i.e., lower levels of expropriation, repudiation, or
corruption).
Although various specification tests affect the coefficient and standard error of Polity stock,
as one might expect, the variable retains statistical significance at the .01 level. When measured as a
stock variable, democracy appears to have a strong positive relationship to growth performance
regardless of the specification of the growth equation. While it is tempting to speculate on the
causal stories contained in various specifications of this growth regression, such speculation is not
immediately germane to our argument. Nor are these stories immediately apparent. It is by no
means clear, for example, which of the control variables should be regarded as exogenous, and
which endogenous, relative to democracy. Thus, we feel it is safer, all things considered, to conduct
further tests with a reduced-form model including only GDP per capita, as in model 2. This is the
only control which can claim some degree of theoretical consensus, is empirically robust, and – with
Having tested a series of control variables we turn now to alternate conceptualizations of our
key variable, Polity stock, and alternative samples. Table 2 begins with the benchmark equation
(model 1), which replicates model 2 in Table 1. This will provide the basis for easy comparisons
across alternative measures and samples. Model 2 excludes all cases for which we added or imputed
additional data for the Polity stock variable, and thus represents the stock variable drawn directly
25
from the Polity IV dataset (Polity2). Not surprisingly – since we have lost only 507 observations –
It is still possible, however, that Polity stock is measuring something other than democracy
per se. One possibility is that this variable is acting as a proxy for the length of time a country has
been autonomous, i.e., the duration of a nation-state. It could be, in other words, that a high Polity
stock score is indicative of a long historical experience of sovereignty, rather than (or in addition to)
a long experience with democratic elections. We offered an initial test of this hypothesis in Table 1,
where we entered a control variable measuring the duration of sovereignty (the number of years a
country has been independent). Even so, our Polity stock variable is suspect since countries enter
the Polity IV dataset only when they become sovereign nations. Except in the case of contiguous
colonies (discussed above), we have no obvious way to treat non-nations in periods prior to
nationhood. This pertains especially to Africa, where most nations came into existence only in the
postwar period. In model 3, therefore, we re-calculate the benchmark regression including only
those countries for which complete data is available for the entire twentieth century (1900-), without any
imputations or additions. Although the number of countries in the dataset as well as the
corresponding number of observations drops to less than one third, the coefficient and standard
errors are fairly stable, suggesting that our results are not the product of peculiarities introduced by
Models 4 and 5 introduce longer lags than are employed in other regressions. Recall that all
regressions reported in Table 1 systematically lag independent variables by one time-period, i.e., one
year. Model 4 shows a ten-year lag on Polity stock and model 5 a twenty-year lag. Remarkably, the
value of Polity stock’s coefficients diminish only slightly, while the standard errors decrease,
26
indicating that the relationship between democracy (stock) and growth is not an instance of reverse
causality.
indicators, Democracy and Autocracy, that comprise the Polity2 index. Both are defined on a scale
respectively. Recall that in order to allow for the possibility of diminishing returns in the growth
benefits accruing from a country’s stock of democracy we use the logarithm of the sum of Polity
stock scores (from 1900 to the present) in all regressions. This specification is consistent with the
increasing costs of increases in the Autocracy stock. To maintain consistency with our standard
regression, in the disaggregated regression reported in model 6 we include the log of the sum of
Democracy scores as well as the log of one thousand minus the sum of Autocracy scores. (One
thousand is the maximum attainable Autocracy score over the course of the twentieth century.)
From model 6 it is apparent that Polity stock’s effect on growth is not simply an artifact of non-
autocracy, as suggested by some writers (Barro 1996, 1997). Rather, both the absence of autocracy
Models 7-11 test the effects of democracy stock on growth when the concept of democracy
concept we code a country-year democratic if it falls above 4 on the Polity2 scale (from -10 to +10).
The first variable, Continuous stock, is created by adding up all years of continuous democracy
experienced by a country in a given year. In 2000, Chile had experienced only 12 years of
continuous democratic rule, and thus receives a score of 12. The US, by contrast, had experienced
100 years of democracy since 1900, and therefore receives a score of 100. These sums are then
logged, as with our other stock measures, in conformance with the expectation of decreasing returns
27
Democracy might also be considered as a cumulative phenomenon such that a country’s
economic performance depends upon the total number of democratic years since 1900 –
disregarding any authoritarian breaks. In this rendition, Chile receives credit for its previous
democratic history, prior to the Pinochet coup. Again, the summed score is logged. Results for this
cumulative stock – together. Models 10 and 11 reprise models 7 and 8 with the addition of our
preferred measure of democracy, Polity stock, where democracy is understood as a scalar concept
Two important conclusions can be drawn from these five regressions (7-11). First, the
continuous, concept. Prior democratic experience matters, regardless of whether or not such
growth is much stronger when democracy is measured as a scalar, rather than dichotomous, concept.
Polity stock swamps the effects of both dichotomous measures. Apparently, both the degree and the
duration of democratic experience matter when one considers the effect of democracy on growth.
Finally, in Table 3, we offer a series of split-sample tests of our major hypothesis. Evidently,
the results shown here are unlikely to be affected by the results of any single country. With a sample
of 178 countries across the world – virtually the entire universe of independent nation-states -- it is
virtually inconceivable that an individual country would constitute an influential case. However, it is
possible that particular regions of the world might affect the results reported here. In this series of
regressions, we exclude regional or socioeconomic blocs that might serve as an influential set of
cases in the benchmark equation: the Middle East (model 1), sub-Saharan Africa (model 2), Asia
28
(model 3), Latin America and the Caribbean (model 4), and the OECD (model 5). None of these
split-sample tests threaten the statistical significance of the Polity stock variable.
Note that the final model restricts the sample to the developing world. This offers a test of
the possibility that the effects measured in our aggregated results might be driven by peculiarities of
the – mostly Anglo-European – early democratizers. Model 5 indicates that the connection between
democracy and growth captured in the Polity stock variable is as strong among late-democratizers as
it is among early democratizers. Thus, it seems safe to conclude that our results are not the product
of idiosyncratic regional effects. This is not to say that the relationship between democracy and
growth is identical in all regions; most assuredly, it is not. It is merely to point out that the aggregate
We have shown thus far that the relationship between democratic stock and growth is robust in a
variety of plausible specifications and operationalizations. (Further tests are conducted in the
appendix.) We turn now to the question of its practical significance. Is the democratic growth effect
Let us consider the results of model 6 in Table 1, in which we control for a range of other
possible causal factors. We regard this model as offering a conservative estimate of causal effects
since many of the variables introduced as controls in this model may be endogenous to democracy,
Recall that in any given year a country’s Polity score can range from -10 to 10. As such, a
country lying in the middle of the distribution -- neither decisively democratic nor autocratic – will
have a Polity stock approximately equal to zero. If such a country subsequently experiences a full
29
decade of high-quality democracy, its Polity stock increases to approximately 100 points. To
estimate the predicted effect of this change on growth, note that because of the possibility of
negative values Polity stock enters the regression as ln(1000 + Polity stock). So from model 6 in
Table 1, the predicted growth impact of a decade of high quality democracy is approximately
The effect for a country that is exiting from a long period of autocracy is greater, translating into an
Given the well-known cumulative effects of small increases in the growth rate, these changes
are significant. For instance, an increase in the annual growth rate from 2% to 2.77% reduces the
time needed to achieve a doubling of incomes from 35 to 25.4 years; an increase to 3.54% further
mention two. First, our econometric specification implies that the benefits of a year of democratic
experience last forever. This is unlikely to be the case. One would expect “investments” in
democratic stock to depreciate just as is the case for investments in physical (or for that matter,
human) capital. The Czech Republic’s current growth rate is surely more affected by its democratic
experience in 1998 than 1938. We have not attempted to model this, rather different, variety of
time-dependence. Another important issue concerns the pathways by which democratic stock might
influence a country’s current growth rate. In our theoretical discussion we suggested several
mechanisms by which such a causal relationship might be realized. However, we did not attempt to
assign a causal weight to any of these causal pathways, a task we must also defer to future work.
INSTITUTIONS IN TIME
30
This paper has demonstrated that the effect of regime type on growth is mediated by a country’s
secular-historical experience of democracy and authoritarianism. This claim stands in sharp contrast
to the conventional wisdom that there is at best no detectable relationship between growth and
democracy or, at worst, a negative relationship. Our somewhat surprising findings stem from a
conceptual insight that democracy/growth effects materialize from a country’s democratic stock
rather than its contemporaneous regime type. Thus, while a country’s level of democracy in a single
year has no measurable impact on its growth rate, its democratic experience over the course of
twentieth century is positively associated with growth. Long-term democracy leads to stronger
economic performance.
This finding, while important in its own right, may also prompt us to re-conceptualize the
causal role of other institutional variables common in the social sciences. Let us begin by
contrasting institutional explanations with other sorts of causal frameworks. When a causal
explanation is rooted in opinions, interests, or coercion (which may be considered a special variety
of interest-based argument), the relationship between a cause and its putative effect is generally
immediate and monotonic. A person’s interests (or opinions or coercive incentives) today should be
acted upon today and will remain the same tomorrow, unless some other factor changes the nature
of that interest (or value). Thus, temporal delays and non-monotonic causal relationships are
possible only if one moves beyond the realm of interests, opinions, and coercion tout court. For
example, if a person does not fully realize his or her true interests then some delay between cause
and effect might be anticipated. Similarly, if the nature of that interest changes over time, we can
imagine non-monotonic results. But this moves us beyond the realm of a purely interest-based
explanation.
With institutional explanations, however, the relationship between cause and effect is often
quite different. We might not expect a rule or norm to have strong immediate effects on the
31
patterns of behavior and expectations it is thought to generate. Instead, these causal relationships
may be conceptualized over a period of time, perhaps quite a long period of time. Moreover, these
slowly developing causal relationships may change over this period in non-monotonic ways. Thus,
we commonly speak of threshold effects, increasing returns (Pierson 2000), reactive sequences
(Mahoney 2000), and a wide variety of non-monotonic relationships. Arguably, the most important
and far-reaching institutional causes have the fewest immediate effects and the greatest non-
linearities. Constitutional frameworks, electoral laws, and property laws are all examples of
reckoned in decades, centuries, or millennia and where non-linear causal relationships are more or
less assumed (Mahoney and Rueschemeyer 2003; Pierson 2004; Steinmo et al. 1992; Waldner 1999).
However, the long-distance, non-linear approach to time is only beginning to take hold among
scholars who work with large-N datasets. Here, explanatory factors are usually considered as
contemporaneous events, or lagged (but still recent) events. The essentially contemporaneous
treatment of variables is entirely appropriate for proximate variables, whose effects are indeed
immediate. But it may not be an appropriate way to consider political institutions whose effects are
“long-memoried” (Pierson 2004). The causal effects of federalism, an electoral system, or a system
only the current status of these constitutional factors or if one takes no account of their changing
variables into the mix. It is now common to employ variables that measure a country’s colonial
history, climate, and geography (Acemoglu et al. 2001; La Porta et al. 1999). By usual construction,
these are static variables. But it is not self-evident that their effects are static through time. That is,
32
it may be important how long a country was colonized, or how long ago. While this is probably not
with respect to most humanly-created institutions (e.g., colonialism, democracy, electoral systems,
In this paper, we have proposed a particular way of thinking about the causal effect of social
institutions. The key insight is that institutions sometimes have cumulative effects, effects that are
noticeable only if an institution’s history is brought into view. Our analysis of the relationship
between democracy and growth suggests that the significant feature of this relationship is the
accumulated history of the variable in question. In this initial study we have employed the simple
relationship. More nuanced views of the ways in which political experience aggregates over time
clearly exist, and warrant investigation. Nonetheless, we hope our results are sufficient to establish
the viability of a cumulative approach to institutional causes. Since institutions are by definition
enduring, and since the effects of institutions are, almost by definition, constitutional – in the sense
of altering the perceptions and behavior of many actors at a systemic level – a wide array of causal
Whatever the method of operationalization, the key concept of stock, or capital, may be useful
as a general term for this species of causal relationship since it allows one to conceptualize the
cumulative effects of an ongoing causal relationship. It is worth reflecting upon the fact that while
the concepts of investment capital and human capital are well known among economists, and the
concept of social capital increasingly common in all the social sciences, we lack an equivalent
concept pertaining to the political realm. Arguably, a country’s stock of democracy is one way of
33
One concrete implication of this approach concerns the effect of institutional reforms,
reforms that currently occupy the attention of academics and policymakers in the developed and
secular-historical changes in policy outcomes, then it is inappropriate to judge the results of such
reforms on the basis of immediate policy gains. Although this renders the testing of hypotheses
more complicated – since we must await long-run results – it may temper the impatience of those
who jump to premature conclusions about the success or failure of regime change, neoliberal reform,
electoral reform, and other reforms of basic institutions. It is unrealistic to expect that the sweeping
changes brought about by such reforms would show instantaneous results. Indeed, while immediate
effects of institutional change are often negative – since such change introduces uncertainties and
information costs in the short run – positive changes are likely to take longer to materialize, since
they depend upon the establishment of a new equilibrium. The concept of political capital may help
policymakers to conceptualize, and thereby accurately account for, the true causal effects of
institutional reform.
34
REFERENCES
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2001. “Colonial Origins of
(December), 1369-1401.
Ades, Alberto and Rafael Di Tella. 1997. “The New Economics of Corruption: A Survey and Some
New Results.” In Paul Heywood (ed), Political Corruption (Oxford: Basil Blackwell), 80-99.
Azfar, Omar and Charles Cadwell (eds). 2002. Market-Augmenting Government: The Institutional
Banerjee, Abhijit V. 1999. “Land Reforms: Prospects and Strategies.” Conference Paper, Annual
World Bank Conference on Development Economics, Washington D.C.; and MIT Department
Banks, Arthur S. 1994. “Cross-National Time-Series Data Archive.” Center for Social Analysis,
Barro, Robert J. 1996. “Democracy and Growth.” Journal of Economic Growth 1 (March), 1-27.
Baum, Matthew A., and David A. Lake. 2003. “The Political Economy of Growth: Democracy and
Beck, Nathaniel. 2001. “Time-Series Cross-Section Data: What Have We Learned in the Last Few
Beck, Nathaniel and Jonathan Katz. 1995. “What to Do (and Not to Do) with Time-Series-Cross-
Section Data in Comparative Politics.” American Political Science Review 89:3 (March), 634-47.
Besley, Timothy and Robin Burgess. 2000. “Land Reform, Poverty Reduction and Growth:
35
Bhalla, Surjit S. 1997. “Freedom and Economic Growth: A Virtuous Cycle?” In Axel Hadenius
(ed), Democracy’s Victory and Crisis (Cambridge: Cambridge University Press), 195-241.
Bleaney, Michael and Akira Nishiyama. 2002. “Explaining Growth: A Contest between Models.”
Bollen, Kenneth A. 1993. “Liberal Democracy: Validity and Method Factors in Cross-National
Bollen, Kenneth A. and Pamela Paxton. 2000. “Subjective Measures of Liberal Democracy.”
Bowman, Kirk, Fabrice Lehoucq, and James Mahoney. 2004. “Measuring Political Democracy:
Data Adequacy, Measurement Error, and Central America.” Working paper, Department of
Chan, Sylvia. 2002. Liberalism, Democracy and Development. Cambridge: Cambridge University Press.
Chua, Amy. 2003. World on Fire: How Exporting Free Market Democracy Breeds Ethnic Hatred and Global
Collier, David and Robert Adcock. 1999. “Democracy and Dichotomies: A Pragmatic Approach to
Choices About Concepts.” Annual Review of Political Science 2:1 (June), 537-66.
Collier, David and Steven Levitsky. 1997. “Democracy with Adjectives: Conceptual Innovation in
Crozier, Michael J., Samuel P. Huntington, Joji Watanuki. 1975. The Crisis of Democracy. New York:
de Janvry, Alain, Gustavo Gordillo, Jean-Philippe Platteau, and Elisabeth Sadoulet (eds). 2001.
Access to Land, Rural Poverty and Public Action. New York: Oxford University Press.
Dornbusch, Rudiger and Sebastian Edwards (eds). 1991. The Macroeconomics of Populism in Latin
36
Dreze, Jean and Amartya Sen. 1989. Hunger and Public Action. Oxford: Clarendon Press.
Eckstein, Susan (ed). 1989. Power and Popular Protest: Latin American Social Movements. Durham, NC:
Englehart, Neil and John Gerring. 2005. “The State: From Concept to Measurement.”
Forthcoming.
Escobar, Arturo and Sonia Alvarez (eds). 1992. The Making of Social Movements in Latin America.
Fein, Helen. 1995. “More Murder in the Middle: Life-Integrity Violations and Democracy in the
Feng, Yi. 1997. “Democracy, Political Stability and Economic Growth.” British Journal of Political
Feng, Yi. 2003. Democracy, Governance, and Economic Performance: Theory and Evidence. Cambridge: MIT
Press.
Forbes, Kristin. 2000. “A Reassessment of the Relationship between Inequality and Growth.”
Gerring, John and Strom Thacker. 2004. “Political Institutions and Corruption: The Role of
Unitarism and Parliamentarism.” British Journal of Political Science 34:2 (April), 295-330.
Gerring, John and Strom Thacker. 2005. Good Government: A Centripetal Theory of Democratic
Governance. Forthcoming.
Gifford, Prosser and William Roger Louis (eds). 1982. The Transfer of Power in Africa: Decolonization,
Gradstein, Mark and Branko Milanovic. 2000. “Does Liberte=Egalite? A Survey of the Empirical
Links Between Democracy and Inequality with some Evidence on the Transition Economies.”
37
Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2004. “The Role of Social Capital in Financial
Haggard, Stephan and Robert R. Kaufman. 1995. The Political Economy of Democratic Transitions.
Hardin, Russell. 1999. Liberalism, Constitutionalism, and Democracy. Oxford: Oxford University Press.
Huntington, Samuel P. 1968. Political Order in Changing Societies. New Haven: Yale University Press.
Jackman, Robert W. 1993. “Rationality and Political Participation.” American Journal of Political
Jens-Hesse, Joachim (ed.). 1993. Administrative Transformation in Central and Eastern Europe. London:
Blackwell.
Keefer, Philip. 2003. “Democratization and Clientelism: Why are Young Democracies Badly
Governed?” Presented at the annual meeting of the American Political Science Association,
Philadelphia (August).
Keefer, Philip and Stephen Knack. 1997. “Why Don’t Poor Countries Catch Up? A Cross-
Knack, Stephen and Philip Keefer. 1997. “Does Social Capital Have an Economic Payoff?: A Cross-Country
Krieckhaus, Jonathan. 2004. “The Regime Debate Revisited: A Sensitivity Analysis of Democracy’s
Kurzman, Charles K., Regina W. Werum, and Ross E. Burkhart. 2002. “Democracy’s Effect on
La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert W. Vishny. 1999. “The
Quality of Government.” Journal of Economics, Law and Organization 15:1 (March) 222-79.
38
Leblang, David A. 1997. “Political Democracy and Economic Growth: Pooled Cross-Sectional and
Lenski, Gerhard. 1966. Power and Privilege: A Theory of Social Stratification. Chapel Hill: University of
Levine, Ross and David Renelt. 1992. “A Sensitivity Analysis of Cross-Country Growth
Lipset, Seymour Martin. 1959. “Some Social Requisites of Democracy: Economic Development
and Political Legitimacy.” American Political Science Review 53:1 (March), 69-105.
Mahoney, James. 2000. “Path Dependence in Historical Sociology.” History and Theory 29:4 (August)
507-48.
Mahoney, James and Dietrich Rueschemeyer (eds). 2003. Comparative Historical Analysis in the Social
Mainwaring, Scott, Daniel Brinks, and Anibal Perez-Linan. 2001. “Classifying Political Regimes in
Latin America, 1945-1999.” Studies in Comparative and International Development 36:1 (October) 7-65.
(http://www.cidcm.umd.edu/inscr/stfail/).
Marshall, Monty G. and Keith Jaggers. 2000. “Polity IV Dataset Project: Political Regime
Mauro, Paolo. 1995. “Corruption and Growth.” Quarterly Journal of Economics 110:3 (August) 681-
712.
39
McGuire, James W. 2005. Politics, Policy, and Mortality Decline in East Asia and Latin America.
Forthcoming.
Meltzer, A.H. and S.F. Richards. 1981. “A Rational Theory of the Size of Government.” Journal of
Mousseau, Demet Yalcin. 2001. “Democratizing with Ethnic Divisions: A Source of Conflict?”
Munck, Gerardo L. and Jay Verkuilen. 2002. “Conceptualizing and Measuring Democracy:
North, Douglass C. 1990. Institutions, Institutional Change and Economic Performance. Cambridge:
North, Douglass C. 1994. “The Historical Evolution of Polities.” Working Paper, Department of
Nuxoll, Daniel A. 1994. “Differences in Relative Prices and International Differences in Growth
O’Donnell, Guillermo and Philippe Schmitter. 1986. Transitions From Authoritarian Rule: Tentative
Olson, Mancur. 1982. The Rise and Decline of Nations. New Haven: Yale University Press.
Persson, Torsten and Guido Tabellini. 2003. The Economic Effects of Constitutions. Cambridge: MIT
Press.
Pierson, Paul. 2000. “Increasing Returns, Path Dependence, and the Study of Politics.” American
40
Pierson, Paul. 2004. Politics in Time: History, Institutions, and Social Analysis. Princeton: Princeton
University Press.
Political Risk Services. 2004. “International Country Risk Guide” (accessed May 10, 2004).
http://www.prsgroup.com/icrg/icrg.html
Przeworski, Adam. 1991. Democracy and the Market. Cambridge: Cambridge University Press.
Przeworski, Adam and Fernando Limongi. 1993. “Political Regimes and Economic Growth.”
Przeworski, Adam and Fernando Limongi. 1997. “Modernization: Theories and Facts.” World
Przeworski, Adam, Michael Alvarez, Jose Antonio Cheibub and Fernando Limongi. 2000.
Democracy and Development: Political Institutions and Material Well-Being in the World, 1950-1990.
Putnam, Robert D. (ed). 2002. Democracies in Flux. Oxford: Oxford University Press
Remmer, Karren L. 1991. “The Political Impact of Economic Crisis in Latin America in the
Rodrik, Dani (ed). 2003. In Search of Prosperity: Analytic Narratives on Economic Growth. Princeton:
Rosenn, Keith S. and Richard Downes (eds). 1999. Corruption and Political Reform in Brazil: The Impact
Sala-i-Martin, Xavier X. 1997. “I Just Ran Two Million Regressions.” American Economic Review 87:2
(May), 178-83.
41
Sala-i-Martin, Xavier X. 2002. “15 Years of New Growth Economics: What Have We Learnt?”
Sartori, Giovanni. 1987. The Theory of Democracy Revisited. Chatham: Chatham House.
Schumpeter, Joseph A. 1942/1950. Capitalism, Socialism and Democracy. New York: Harper & Bros.
Shapiro, Ian and Casiano Hacker-Cordon (eds). 1999. Democracy’s Value. Cambridge: Cambridge
University Press.
Snyder, Jack L. 2000. From Voting to Violence: Democratization and Nationalist Conflict. New York:
W.W. Norton.
Sokoloff, Kenneth L. and Stanley L. Engerman. 2000. “Institutions, Factor Endowments, and
Paths of Development in the New World.” Journal of Economic Perspectives 14:3 (Summer) 217-32.
Steinmo, Sven, Kathleen Thelen, and Frank Longstreth (eds). 1992. Structuring Politics: Historical
Stepan, Alfred P. 1989. Democratizing Brazil: Problems of Transition and Consolidation. Oxford: Oxford
University Press.
Stokes, Susan Carol. 2001. Markets, Mandates, and Democracy: Neoliberalism by Surprise in Latin America.
Stokes, Susan Carol (ed). 2002. Public Support for Market Reforms in New Democracies. Cambridge:
Summers, Robert and Alan Heston. 1991. “The Penn World Table (Mark 5): An Expanded Set of
Tarrow, Sidney. 1998. Power in Movement: Social Movements and Contentious Politics. Cambridge:
Tavares, Jose and Romain Wacziarg. 2001. “How Democracy Affects Growth.” European Economic
42
Temple, Jonathan. 1999. “The New Growth Evidence.” Journal of Economic Literature (March) 112-
56.
Treier, Shawn and Simon Jackman. 2003. “Democracy as a Latent Variable.” Presented to the
Treisman, Daniel S. 2000. “The Causes of Corruption: A Cross-National Study.” Journal of Public
United Nations Development Programme (UNDP). 2004. Overview: Democracy in Latin America:
Towards a Citizens’ Democracy. New York: United Nations Development Programme. Accessed
Vanhanen, Tatu. 1990. The Process of Democratization: A Comparative Study of 147 States, 1980-88. New
Waldner, David. 1999. State Building and Late Development. Ithaca: Cornell University Press.
Warren, Mark E. 1999. Democracy and Trust. Cambridge: Cambridge University Press.
Weyland, Kurt. 2002. The Politics of Market Reform in Fragile Democracies: Argentina, Brazil, Peru, and
Woo-Cumings, Meredith (ed). 1999. The Developmental State. Ithaca: Cornell University Press.
Woolcock, Michael. 1998. “Social Capital and Economic Development: Toward a Theoretical
Synthesis and Policy Framework.” Theory and Society 27:2 (April), 151-208.
World Bank. 2003. World Development Indicators [data file]. Washington: World Bank.
43
APPENDIX
44
Appendix A:
In the body of the paper we defend a particular method for testing the democracy/growth relationship in a
time-series crossnational format. Our approach relies on annual data, fixed effects, an AR1 correction for
serial autocorrelation, and a century-long measurement of regime history (1900-). In this appendix we show
that our results are robust even when various elements of this methodology are altered.
Table A1 displays the results of simple regressions of growth rates over the 1990s on
democracy, democracy stock, and an assortment of standard control variables. As is clear, even in
this simplest version of the analysis, democracy stock is positively and significantly correlated with
growth rates. As usual, the level of democracy is uncorrelated with growth unless almost all controls
are omitted. Its effect disappears once the democracy stock variable is introduced into the equation
(model 3). Thus, the conclusion that democracy stock significantly impacts growth is not an artifact
of our panel data approach. Of course, the correlations of Table A1 may be a result of the omitted
variable bias --- which is precisely why we concentrate on estimates of fixed-effect regressions in the
main text.
In Table A2, we turn to a more complex set of methodological issues: fixed-effects versus random-
effects, annual versus five-year increments of data, the possible influence of OECD cases (tested this time in
possible peculiarities of the Polity stock variable, and a wide variety of static (time-invariant) control variables.
For each model with annual data we have followed our usual approach of measuring all independent
variables in the year prior to the dependent variable. For each model with five-year increments we have
maintained the same approach, this time measuring the independent variables in the first year of the period
under study. The dependent variable in this case is a five-year average of growth performance during the
subsequent period. So, in both cases the dependent variable is forward-lagged one time-period.
45
The standard method of correcting for autocorrelation is employed where annual data is used (AR1
error correction), but not when 5-year increments are used. (By virtue of the fact that we are dealing with 5-
year increments it should be less of a problem.) No correction for serial autocorrelation is usually necessary
Model 1 is a fixed-effect model with one control (GDPpc) and growth data aggregated across five-
year periods. Model 2 is a random-effects model with annual data and the same all-purpose control. Model 3
is a random-effects model with annual data that includes all large-N controls employed previously (see Table
2). (The inclusion of small-N controls such Capital expenditures, Expropriation, Repudiation, and
Corruption would significantly erode the sample size of the analysis.) Model 4 is a random-effects model
with five-year data increments and the same set of controls. Model 5 is a fixed-effects model with five-year
increments and all relevant (varying) controls. Models 6 and 7 replicate models 3 and 4, this time excluding
OECD cases. Models 8 and 9 test the Beck/Katz approach to TSCS analysis (Beck and Katz 1995), with a
lagged dependent variable included as a way of de-trending the data. Models 10 and 11 test another version
of the political capital thesis. Instead of measuring democracy stock since 1900, we instead employ a moving
sum variable that adds up democracy scores (from the same PolityIV dataset) over 20- and 50-year intervals.
In each of these various tests we find that the Polity stock variable retains statistical significance.
46
Table A1:
Cross-sections, 1990s
Model: 1 2 3 4 5 6
Polity level (ln) .561** .055 .311 -.024
(.260) (.337) (.382) (.397)
Polity stock (ln) 2.463*** 2.430*** 2.009** 2.023**
(.641) (.739) (.892) (.940)
GDPpc (ln) .062 -.099 -.108 -.467 -.915** -.915**
(.143) (.176) (.162) (.392) (.353) (.353)
Investment .056 .069* .069*
(.041) (.038) (.038)
Government .034 .028 .028
consumption (.043) (.042) (.042)
Openness .007* .007* .007*
(.004) (.004) (.004)
Population growth -17.527 -29.835 -29.845
(32.572) (29.255) (29.358)
Life expectancy .074 .100* .101
(.063) (.059) (.063)
Constant -.725 -14.887*** -14.714*** -2.208 -13.403** -13.499**
(1.298) (3.949) (4.325) (2.532) (5.889) (6.247)
Observations 167 167 167 126 126 126
Countries 167 167 167 126 126 126
R-squared .02 .11 .11 .14 .18 .18
Prob>F .0804 .0003 .0006 .0010 .0000 .0001
Ordinary least squares analysis. Units of analysis: country-year. Period of analysis: 1990-2000. Dependent variable: average
growth rate, 1990-2000. Independent variables all measured in 1990. Robust standard errors in parentheses. *** p<.01 ** p<.05
*p<.10 (two-tailed tests) Variables and procedures defined in the text.
47
Table A2:
Alternative Estimators and Models
Model: 1 2 3 4 5 6 7 8 9 10 11
Estimator: fe re re re fe re re fe fe fe fe
Periods of analysis: 5 years Annual Annual 5 years 5 years Annual 5 years Annual Annual Annual Annual
Non-
Sample: All All All All All Non-OECD All All All All
OECD
Polity stock (1900-) (ln) 5.584*** 3.339*** 2.281*** 2.034*** 10.192*** 2.316*** 2.286** 3.946*** 7.679***
(1.171) (0.476) (0.625) (0.686) (1.966) (0.834) (0.913) (0.774) (1.877)
Polity stock (20-year) 0.008**
(0.003)
Polity stock (50-year) 0.441***
(0.140)
GDPpc (ln) -4.319*** -0.910*** -1.078*** -1.211*** -5.483*** -1.113*** -1.200*** -2.766*** -6.152*** -4.267*** -5.723***
(0.326) (0.132) (0.264) (0.285) (0.668) (0.308) (0.327) (0.210) (0.622) (0.399) (0.540)
Growthpc (lagged dep var) 0.259*** 0.150***
(0.012) (0.020)
Inflation (ln) -0.131 0.064 -0.125 -0.077 0.096 -0.259***
(0.095) (0.100) (0.110) (0.110) (0.115) (0.095)
Investment -0.061*** -0.049** -0.051* -0.058** -0.037 -0.090***
(0.023) (0.022) (0.027) (0.027) (0.026) (0.026)
Foreign investment 0.166*** 0.027 0.093* 0.181*** 0.024 0.180***
(0.054) (0.051) (0.053) (0.062) (0.059) (0.052)
Openness 0.013*** 0.012** 0.016* 0.010* 0.009 0.040***
(0.004) (0.005) (0.009) (0.005) (0.006) (0.008)
Government consumption -0.029* -0.033** -0.014 -0.018 -0.023 0.002
(0.016) (0.017) (0.022) (0.019) (0.020) (0.021)
Population growth 4.209 -45.746*** -44.257** 5.913 -43.724** 23.920
(13.368) (17.745) (21.849) (15.293) (21.117) (14.647)
Years independent 0.003 0.002 0.131 0.003 0.003 0.094
(0.003) (0.003) (0.238) (0.004) (0.004) (0.282)
Regime durability -0.014* -0.012 -0.011 -0.017 -0.025* 0.003
(0.008) (0.009) (0.014) (0.012) (0.013) (0.013)
Instability -0.076*** -0.003 -0.002 -0.092*** -0.004 -0.095***
(0.025) (0.023) (0.023) (0.031) (0.029) (0.024)
Social conflict -0.233 0.338 0.057 -0.152 0.215 -1.045*
(0.556) (0.552) (0.634) (0.644) (0.641) (0.606)
Life expectancy 0.161*** 0.152*** 0.215*** 0.167*** 0.159*** 0.184***
(0.031) (0.034) (0.059) (0.036) (0.039) (0.050)
Illiteracy (ln) 0.130 0.024 0.637* 0.171 -0.013 0.712**
(0.178) (0.188) (0.342) (0.238) (0.247) (0.334)
Trend 0.032 0.098** -0.028 0.010 0.055 -0.025
(0.043) (0.048) (0.243) (0.053) (0.061) (0.284)
Oil shock -1.178** -0.948* -0.407 -1.068* -0.659 -0.831*
(0.490) (0.509) (0.505) (0.596) (0.620) (0.430)
Latitude (ln) -0.083 -0.117 0.014 -0.205
(0.391) (0.438) (0.496) (0.542)
English legal origin -0.012 -0.007 -0.009 -0.003
(0.008) (0.009) (0.009) (0.010)
Muslim -1.062 -0.755 -0.943 -0.855
(0.783) (0.889) (0.974) (1.085)
Ethnic fractionalize (Alesina) 3.273*** 3.493*** 3.259*** 3.354***
(0.695) (0.784) (0.926) (1.019)
East Asia 0.849 0.840 0.876 0.700
(0.812) (0.922) (0.930) (1.031)
Middle East -0.948* -0.929 -0.923 -1.362*
(0.523) (0.586) (0.727) (0.794)
Latin America 0.403* 0.219 0.247 0.042
(0.221) (0.253) (0.254) (0.283)
Decade dummies Yes Yes Yes Yes Yes Yes
Constant -4.342 -14.296*** -14.702*** -15.045*** -52.568*** -15.120** -15.711** -5.015 -25.225 34.507*** 47.665***
(8.309) (3.095) (4.634) (5.039) (19.489) (5.956) (6.392) (5.356) (20.770) (1.882) (2.296)
Observations 1084 6107 2459 492 509 1892 378 5980 2516 4309 2660
Countries 180 181 124 117 119 98 91 180 126 172 95
R-squared 0.18 0.00 0.08 0.27 .34 0.08 0.28 0.10 0.14 0.03 0.04
fe = fixed-effects. re = random effects. Period of analysis: 1990-2000. Dependent variable: average growth rate, 1950-2000. All
predictors lagged one time-period. Standard errors in parentheses. *** p<.01 ** p<.05 *p<.10 (two-tailed tests) Variables and
procedures defined in the text.
48
Descriptive Statistics for Key Variables
49
Bivariate Correlations of Key Variables
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
1. Polity Stock (ln) 1.00
2. GDPpc (ln) .41 1.00
3. Investment .31 .57 1.00
4. Governm. Consumption -.19 -.19 -.24 1.00
5. Openness .08 .29 .22 .20 1.00
6. Population growth -.22 -.49 -.24 .06 -.16 1.00
7. Life expectancy .15 .78 .57 -.12 .28 -.50 1.00
8. Growth pc global mean .04 -.07 .05 -.16 -.13 .04 -.19 1.00
9. Inflation (ln) -.17 -.12 -.11 .15 -.07 .00 -.04 -.13 1.00
10. Foreign direct invest. .06 .10 .18 .14 .38 -.12 .12 .03 -.07 1.00
11. Instability .00 -.03 -.01 -.05 -.24 .03 .00 .02 .12 -.09 1.00
12. Social conflict -.07 -.22 -.11 .07 -.20 .09 -.13 -.09 .14 -.10 .35 1.00
13. Years independent .16 .34 .20 -.14 -.23 -.21 .28 -.06 .05 -.08 .17 .06 1.00
14. Regime durability .35 .56 .34 -.19 .12 -.28 .46 -.07 -.17 .07 -.12 -.09 .27 1.00
15. Illiteracy .24 -.52 -.32 .13 -.26 .59 -.67 .05 -.11 -.12 .00 .13 .01 -.03 1.00
16. Primary education .17 .43 .34 -.06 .19 -.26 .56 -.04 -.04 .18 -.04 -.08 .18 .18 -.37 1.00
17. Corruption .45 .64 .51 -.19 .21 -.45 .53 .00 -.10 .12 -.14 -.27 .29 .54 -.36 .29 1.00
18. Repudiation .39 .69 .60 -.34 .27 -.51 .60 .02 -.27 .20 -.11 -.28 .30 .42 -.42 .39 .64 1.00
19. Expropriation .33 .62 .52 -.32 .28 -.49 .55 .05 -.22 .23 -.10 -.30 .23 .38 -.45 .37 .62 .86 1.00
50
TABLES AND FIGURES
51
Figure 1:
Hypothesized Causal Pathways
Physical, Human,
and Social Capital
Democracy Economic
stock growth
Political Capital:
a) Better Policies &
Institutions
b) Stability, Consensus
52
Table 1:
Specification Tests
Model: 1 2 3 4 5 6 7 8 9
Polity level -.030
(.024)
Polity stock (ln) 6.188*** 4.704*** 4.843*** 6.336*** 7.677*** 9.938*** 10.282*** 11.022***
(1.031) (1.058) (1.005) (1.053) (1.859) (3.717) (3.726) (3.614)
GDPpc (ln) -3.606*** -3.920*** -3.648*** -5.268*** -8.351*** -11.317*** -11.562*** -11.594***
(.302) (.303) (.296) (.393) (0.784) (1.607) (1.624) (1.597)
Growth pc .536***
(trade-weighted) (.049)
Inflation (ln) -0.205* -0.217 -0.231 -0.248*
(0.109) (0.141) (0.141) (0.138)
Investment -0.087*** -0.207*** -0.198*** -0.193***
(0.031) (0.055) (0.054) (0.053)
Foreign investment 0.166*** 0.221** 0.217** 0.209**
(0.058) (0.096) (0.097) (0.095)
Openness 0.049*** 0.086*** 0.088*** 0.089***
(0.009) (0.015) (0.014) (0.014)
Government 0.000 0.016 0.013 0.017
consumption (0.025) (0.037) (0.037) (0.036)
Population growth 27.915* 43.085* 40.206 36.348
(15.207) (25.168) (25.106) (24.724)
Years independent 0.032 0.082 0.110 0.090
(0.117) (0.091) (0.089) (0.084)
Regime durability 0.018 0.015 0.015 0.014
(0.015) (0.023) (0.023) (0.023)
Instability -0.101*** -0.151*** -0.150*** -0.148***
(0.026) (0.039) (0.039) (0.038)
Social conflict -1.966*** -2.923*** -3.061*** -3.347***
(0.714) (1.064) (1.060) (1.056)
Life expectancy 0.193*** 0.060 0.068 0.128
(0.063) (0.138) (0.139) (0.136)
Illiteracy (ln) 0.902** -1.008 -1.609 -2.884
(0.418) (2.334) (2.295) (2.143)
Trend 0.041 0.647 0.668 0.637
(0.123) (0.466) (0.467) (0.461)
Oil shock -1.260**
(0.516)
Expropriation 0.271*
(PRS) (0.163)
Repudiation 0.163
(PRS) (0.167)
Corruption -0.213
(PRS) (0.286)
Annual dummies Yes
Decade dummies Yes Yes Yes Yes
Constant 28.753*** -11.204** -30.353*** -5.249 -4.734 -5.018 9.919 8.722 7.589
(1.538) (4.811) (5.020) (4.858) (4.827) (8.365) (11.852) (11.874) (11.447)
Observations 5926 5926 6082 5720 5926 2390 1010 1010 1037
Countries 180 180 186 175 180 121 91 91 91
Rsq (within) .03 .03 .00 .05 .07 0.10 0.15 0.15 0.14
Prob>F .0000 .0000 .0000 .0000 .0000 0.0000 0.0000 0.0000 0.0000
Fixed effect regressions with AR(1) disturbance. Units of analysis: country-year. Period of analysis: 1950-2000. DV: annual per capita growth
rate. All predictors lagged one year. Standard errors in parentheses. *** p<.01 ** p<.05 *p<.10 (two-tailed tests) Variables and procedures
defined in the text.
53
Table 2:
Polity Stock, Variously Operationalized
Model: 1 2a 3b 4 5 6 7 8 9 10 11
GDPpc (ln) -3.920*** -1.554*** -1.292*** -4.013*** -3.840*** -3.710*** -3.829*** -4.039*** -4.012*** -3.990*** -4.188***
(.303) (.190) (.245) (.343) (.365) (.335) (.318) (.322) (.324) (.317) (.321)
Polity stock (ln) 6.188*** 7.327*** 4.753*** 6.089*** 5.805***
(1.031) (1.030) (1.682) (1.049) (1.049)
Polity stock (ln), 10-year lag 3.639***
(.827)
Polity stock (ln), 20-year lag 3.952***
(.692)
POLITY COMPONENTS:
Democracy stock (ln) 1.173***
(.221)
Autocracy stock (reverse scale, ln) 5.494***
(1.239)
DICHOTOMOUS MEASURES:
Continuous stock (ln, Polity>4) .274* -.160 .102
(.164) (.221) (.165)
Cumulative stock (ln, Polity>4) .734*** .877*** .520**
(.221) (.296) (.222)
Constant -11.204** -37.009*** -20.248*** 7.826** 5.505** -11.910* 30.042*** 30.694*** 30.449*** -10.135** -7.481
(4.811) (4.650) (6.854) (3.699) (2.746) (6.157) (1.574) (1.559) (1.575) (4.950) (4.920)
Observations 5926 5419 6.854 5229 4396 5285 5926 5926 5926 5926 5926
Countries 180 156 53 179 173 156 180 180 180 180 180
Rsq (within) .03 .02 .01 .03 .03 .03 .03 .03 .03 .03 .03
Prob>F .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000 .0000
Fixed effect regressions with AR(1) disturbance. Units of analysis: country-year. Period of analysis: 1950-2000. DV: annual per capita growth rate. All predictors lagged one year.
Standard errors in parentheses. *** p<.01 ** p<.05 *p<.10 (two-tailed tests) Variables and procedures defined in the text. a: Includes no imputed scores for Polity2. b: Includes only
countries for which data for Polity stock is available for the non-imputed core variable (Polity2) over the course of the entire twentieth century (Polity2_complete==1).
54
Table 3:
Split-Sample Tests
Model: 1 2 3 4 5
Latin Am/
Excluding: Middle East Africa Asia OECD
Caribbean
Polity stock (ln) 3.616*** 5.474*** 9.626*** 6.496*** 6.291***
(1.312) (1.042) (1.134) (1.137) (1.298)
GDPpc (ln) -3.639*** -3.290*** -5.831*** -3.681*** -4.598***
(.320) (.316) (.375) (.333) (.407)
Constant 4.051 -9.076* -20.344*** -15.226*** -9.064
(5.876) (4.676) (5.153) (5.260) (6.244)
Observations 5292 4292 5192 4685 4576
Countries 161 131 157 145 147
Rsq (within) .02 .03 .05 .03 .03
Prob>F .0000 .0000 .0000 .0000 .0000
Fixed effect regressions with AR(1) disturbance. Units of analysis: country-year. Period of analysis: 1950-2000. DV: annual per
capita growth rate. All predictors lagged one year. Standard errors in parentheses. *** p<.01 ** p<.05 *p<.10 (two-tailed tests)
55
NOTES
1
This research was funded by a generous grant from the Frederick Pardee Center for the
Study of the Longer-Range Future at Boston University. We are grateful to Carola Moreno for her
but the relation turns negative once a moderate amount of political freedom is attained.” A few
recent studies find a positive overall relationship between democracy and growth (e.g., Bhalla 1997;
the concept of democracy over time (e.g., Weede 1996), and none that stretches back to encompass
the whole twentieth century. Persson and Tabellini (2003) look at the effect of democracy --
conceptualized as a dichotomous concept over the twentieth century -- on various economic policies
7 Correlations between Polity2 and other democracy indices (introduced below in the text)
are as follows: “Political Rights” (Freedom House) = -.85; “Liberal Democracy” (Bollen) =.92;
8 This re-coding affects the following countries: Albania (1900-1912, Ottoman Empire),
Iraq (1900-1917, Ottoman Empire), Israel (1900-1917, Ottoman Empire), Kazakhstan (1900-1990,
www.freedomhouse.org/research/freeworld/2000/methodology.htm
10
A recent reevaluation of cross-country growth empirics concludes that the log of GDP per
capita is the only variable that is robust across all models (Bleaney and Nishiyama 2002: 45).
11
We interpolate missing values for illiteracy, life expectancy, and population growth in
order to maintain a consistent sample (failing to do so would have significantly reduced the sample
12 Consider a country that has experienced fifty years of severe dictatorship. The effect of a
decade of full democracy on the log of Polity stock would be ln(600)-ln(500) ≈ 0.2.
57