1 Democracy and Growth: A Lexical Approach • John Gerring, Boston University • Svend-Erik Skaaning, University of Aarhus • Matthew Maguire, Boston University • Eitan Tzelgov, Gothenburg University ABSTRACT Scholars have long debated whether regime-type has any impact on growth. Empirical work based on established indices of democracy suggests that the relationship is weak or nonexistent. In this paper, we present a theory for why regime-type might matter for growth based on the differential incentives facing leaders in democratic and non-democratic settings. To test this theory, we develop a new index of electoral democracy based on a “lexical” approach to scaling, in which levels of the index represent successive necessary conditions of an ordinal scale. In the empirical sections of the paper we show that this index bears a positive and highly robust relationship to growth, reconcile this result with previous findings based on other measures of democracy, and explore evidence of an accountability mechanism.
77
Embed
Democracy and Growth: A Lexical Approach - BU Blogsblogs.bu.edu/jgerring/files/2014/10/BJPS_lexical_complete.pdf · 1 . Democracy and Growth: A Lexical Approach • John Gerring,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Democracy and Growth: A Lexical Approach
• John Gerring, Boston University
• Svend-Erik Skaaning, University of Aarhus
• Matthew Maguire, Boston University
• Eitan Tzelgov, Gothenburg University
ABSTRACT
Scholars have long debated whether regime-type has any impact on growth. Empirical work
based on established indices of democracy suggests that the relationship is weak or
nonexistent. In this paper, we present a theory for why regime-type might matter for growth
based on the differential incentives facing leaders in democratic and non-democratic settings.
To test this theory, we develop a new index of electoral democracy based on a “lexical”
approach to scaling, in which levels of the index represent successive necessary conditions of
an ordinal scale. In the empirical sections of the paper we show that this index bears a
positive and highly robust relationship to growth, reconcile this result with previous findings
based on other measures of democracy, and explore evidence of an accountability
mechanism.
2
For many who study patterns of development around the world it is axiomatic that
“institutions matter.”1 Among institutions, a country’s regime-type is often regarded as
fundamental, especially with respect to securing conditions for long-term growth.2
A variety of reasons may be adduced to support this view. A democratic framework
may assure credible commitment to property rights3; set in place incentives for leaders to
provide growth fostering public goods4; encourage technological change and productivity
growth5; enhance prospects for learning and deliberation, tapping into the wisdom of
crowds6; and encourage norms of trust, equality, fairness, and reciprocity.7
Not everyone agrees with this optimistic perspective. Some argue, with an eye to the
East Asian NICs (newly industrializing countries), that economic growth is more likely to be
achieved through strict authoritarian rule, a necessary condition for instilling discipline in the
labor force, prioritizing long-term savings and investment over current consumption, and
resisting rent-seeking pressures from organized groups.8 Democracy, by contrast, has been
associated with populist policies that favor short-term redistribution over long-term growth9;
policy sclerosis and a clientelist, rent-seeking style of politicking in which side-payments to 1 Acemoglu et al. 2005; North 1990; Rodrik et al. 2004.
2 Acemoglu & Robinson 2012; Barzel 2002; Bueno de Mesquita & Root 2000; Halperin et al. 2004; Knack &
Y = GDP per capita growth. OLS = ordinary least squares analysis. RE = random effects. FE = fixed effects. MI = full dataset imputed with Amelia II algorithm (Honaker et al. 2011). GMM = generalized method of moments (Blundell & Bond 1998). TSLS = two-stage least squares. Standard errors clustered by country except in Model 10. *** p<.01, ** p<.05, * p<.1 (two-tailed test) Model 1 is understood as the benchmark model.
19
than cross-sectional variation. Year fixed-effects provide assurance that results are not driven
by period effects. Information about variables in this table and subsequent tables – including
coding, sources, and descriptive statistics – is contained in Appendix B.
Model 1 in Table 1 includes a global sample of countries observed from 1820 to
2008. It is worth noting that although historical data is routinely employed when testing the
modernization thesis47, it is rarely employed when democracy sits on the right side of a
causal model. To measure growth across two centuries we rely on estimates of per capita
GDP drawn from Angus Maddison.48 The resulting analysis suggests that a one-point
increase in the Lexical index translates into a 0.4% increase in growth, a considerable impact
when judged over the long term. Subsequent models introduce variations in this benchmark
model.
Model 2 focuses on the contemporary era (1960-). Here, we adopt widely used
measures of GDP and GDP growth contained in the World Development Indicators.49
While this analysis suggests a somewhat smaller impact of regime-type on growth it is
important to bear in mind that the variation contained within the contemporary sample is
truncated, with 43% of the cases classified as L6 (see Table A1). This may account for the
reduced coefficient, relative to the benchmark model (Model 1).
The next series of tests focus on model specification. In Model 3, we add the
following covariates: Urban (percent living in urban areas), Population, and Capability (an
index of state strength that combines iron/steel production, energy use, military
47 E.g., Epstein et al. 2006.
48 Maddison 2010. An updated and slightly expanded version of the original Maddison dataset (Bolt & Van
Zanden 2013) shows similar patterns when the benchmark model is tested.
49 World Bank 2007.
20
expenditures, military personnel, and total and urban population). The inclusion of
Capability is especially important. This factor, which changes over time, may affect the
propensity of a state to develop democratic institutions as well as to achieve strong growth
performance, thus serving as a potential confounder. It might also be viewed as endogenous
to regime-type, in which case it is correctly regarded as a mechanism rather than a potential
confounder. (For this reason, we do not include it in our benchmark model.)
In Model 4, we add (to those in the previous model) a set of non-varying covariates
including European language (percent speaking a European language), English legal origin
(former British colony), Latitude (distance from equator, natural logarithm), Landlock (lack
of access to an ocean), and regional dummies (Africa, Asia, Latin America, Middle East).
Because these covariates are static we employ a random effects estimator. Results from the
specification tests in Models 3 and 4 are consistent, though there is some instability in the
estimated effect of electoral democracy on growth, as one might expect.
In Model 5, we return to the benchmark model, this time imputing a full sample –
i.e., all sovereign states from 1820 to 2004 – using the Amelia II multiple-imputation
algorithm.50 Sample size increases modestly – from just under 12,000 observations (Model 1)
to just over 13,000 observations (Model 5). Results indicate a slight decrease in the
coefficient for the Lexical index relative to the benchmark model, suggesting that there is
little sample bias.
In Model 6, we introduce a lagged dependent variable (along with annual fixed
effects), which may help to block confounders and to solve problems of autocorrelation.51
The length of our panel obviates concerns about bias often generated when a lagged
50 Honaker et al. 2011.
51 Beck, Katz 1995.
21
operator is combined with unit fixed effects. Results for our key variable are virtually
identical to those reported in our benchmark model. Moreover, results for the key variable
are virtually unaffected when country fixed effects are removed and a random-effects
estimator is adopted (not shown).
In Model 7, the unit of analysis shifts from annual to five-year periods. This is
accomplished by calculating the mean value for growth across a moving five-year period and
running a panel analysis with every fifth year. (Since right-side variables are scarcely affected
by this moving-average they remain in their accustomed format.) Results are diminished but
persistent relative to the benchmark model.
An ever-present threat to inference is the possibility of endogeneity between X and
Y. Evidently, if growth causes democratization then any apparent relationship is spurious.
One approach to this problem is to lag right side variables several periods behind the
outcome. In Model 8, the Lexical index is given a five year lag. While the coefficient is
somewhat diminished, the relationship retains significance. Of course, regime-type changes
slowly so that a country’s Lexical score at T is highly correlated with its score at T-5.
Another approach to modeling endogeneity is to exchange X and Y – from one side
of the model to the other – leaving all other elements intact. Here, we find that any possible
impact of growth on democracy disappears after a single lag (one year), suggesting that any
X/Y endogeneity is at best short-lived and cannot serve as a confounder in Model 8. By and
large, the literature on democratization corroborates this interpretation. While a country’s
level of development probably affects its propensity to create and/or sustain a democratic
form of government52, studies suggest that annual growth performance has little impact on
these outcomes (though it may have repercussions for leadership turnover, as shown in 52 Epstein et al. 2006.
22
Table 4). A more plausible conjecture is that growth performance has a positive, proximal
effect on stability for any regime-type, democratic or autocratic.53
Yet another approach to X/Y endogeneity employs instruments for all right- and
left-side variables, drawn from their lagged and differenced values. The GMM estimator
developed by Blundell & Bond arrives at estimates for the Lexical index that are virtually
identical to our benchmark model, as shown in Model 9.54 (Because GMM estimators are
designed for panels that are short and wide, we replicate this approach with the
contemporary sample and specification shown in Model 2. Here, the estimated effect of
Lexical on Growth is even stronger than that reported in Model 9 [available upon request].)
A final approach to model identification enlists an instrumental variable in a two-
stage analysis. Previous studies suggest that a country’s regime-type is strongly influenced by
neighborhood effects.55 A country in a predominantly democratic neighborhood may be
subjected to strong peer pressure to adopt a democratic form of government, while a
country in a predominantly autocratic neighborhood may face considerable adversity if it
wishes to transition to democracy. A simple approach to modeling diffusion regards all
countries within 500 kilometers as neighbors56; their mean level of democracy (measured by
the Lexical index) becomes the basis for a diffusion variable. This variable is lagged one-
period in order to further exogenize the instrument.57
the Lexical index on growth. All specification tests are passed (at conventional statistical
thresholds).
Several features of our data may help explain the stability of results across various
samples, specifications, estimators, and codings. First, our sample is long (T~200 years) and
wide (N~186), offering nearly 12,000 observations. Second, there is considerable year-to-
year variation in the outcome of interest. Indeed, the correlation between growth at T and T-
1 is nearly zero. (It is slightly higher in the nineteenth-century, due to the linear interpolation
of missing data; however, these provide only a small portion – roughly one quarter – of the
total sample.)
Serial correlation in the predictor is somewhat more troublesome, as countries do
not alter regime-types frequently. Even so, among 16,899 country-years, 1,336 regime
changes are registered by the Lexical index, roughly eight percent of the sample. This means
that the average country experiences at least six regime changes during the observed period –
a fair bit of variation. To be sure, the pattern of change trends upward: 880 changes
correspond to an increase in a country’s lexical score while 456 changes correspond to a
decrease, a roughly 2:1 differential. Yet, 456 instances of democratic reversal are still a
considerable number. Note also that since growth does not exhibit a strong secular-historical
pattern we need not fear accidental correlation between left- and right-side variables with
parallel trends. More problematic are year-to-year variations, a feature that annual dummies
should overcome.
In sum, many of the problems traditionally associated with panel data – short panels,
sluggish variables, long-term trends that co-vary75 – are less problematic in this particular
setting. 75 Bertrand et al. 2004.
27
IV. Other Measures of Democracy
We have shown that when features of electoral democracy are arranged in a “lexical” fashion
– as a sequence of necessary conditions within an ordinal scale – a positive relationship to
growth emerges that is robust to a variety of samples, specifications, and estimators (Table
1). This raises a puzzle, for most studies find little or no relationship between regime-type
and growth, as discussed in our preliminary review of the literature.76 In this section, we 76 Somewhat more optimistic views of the democracy/growth relationship can be found in some recent work
(for an extensive review see Knutsen 2012). Studies that operationalize democracy as a historical (stock) variable
often find a positive relationship to growth (Ferree & Singh 2006; Gerring, Bond, Barndt & Moreno 2005;
Persson & Tabellini 2009). Studies sometimes find a conditional relationship between democracy and growth.
For example, Wu (2012) finds an interactive relationship between regime-type and structural factors such as
external threats and natural resource intensity, namely, democracies perform better than autocracies when these
impediments are present but not when they are absent. In this vein, Heo & Tan (2001) conduct a Granger test
of causality, suggesting that in about one-third of the countries in their sample democracy has a causal effect on
economic growth.76 Some studies discern a curvilinear relationship between democracy and growth. For
example, Barro (1997) finds that as one moves from a low to moderate level of democracy growth increases.
But as one moves to a higher level of democracy the relationship becomes negative (see also Barro 1996;
Plumper, Martin 2003).
Several studies that reach positive conclusions about the nexus between regime-type and economic
performance nexus are marred by questionable research designs and/or non-replicable findings. For example,
Benyishay & Betancourt (2010) examine the relationship between one component of the Freedom House
index, “Personal Autonomy and Individual Rights,” and growth, where they find a positive relationship.
However, this particular component – described by the authors as “the extent of personal economic freedoms
such as the choice of ownership form, employment, residence and education, as well as social freedoms such as
choice of marriage partners and family size” (2010: 282) – is peripheral to the usual meaning of democracy.
Moreover, the results are rather shaky given that data for the key independent variable is available only for a
28
compare the Lexical index with extant indices in a systematic fashion in order to shed light
on this discrepancy.
Our review incorporates extant indices of electoral democracy with prominence in
the social science literature and broad country and temporal coverage. Among ordinal
indices, we include the “Polity2” variable from the Polity IV dataset77 and the Political rights
(“PR”) and Civil liberty (“CL”) indices, both produced by Freedom House.78 Among interval
indices, we include the Democracy Index produced by Vanhanen79, the Contestation and
Inclusive indices produced by Michael Coppedge and collaborators80, and the Unified
single year and is tested for a small subsample of sixty countries. Papaioannou and Siourounis (2008) construct
a binary coding of enduring democratization episodes, which they show to be associated with a significant long-
term improvement in growth rates. The problem with this analysis is its construction of the key independent
variable. Countries receive a positive score only if they manage to maintain a democratic regime indefinitely
(until the end of the observed period). This equates democratization with successful democratization, aka
democratic consolidation, making causal inference especially problematic. Rodrik and Wacziarg (2005) find a
positive relationship between democratization episodes and growth in the short-term (the five years following a
transition). While suggestive, this finding is apparently contingent upon the inclusion of several covariates that
may be endogenous to the causal factor of interest. It also begs the question of what the longer-term impact of
regime-transitions might be. Persson and Tabellini (2006) find a positive relationship when growth is regressed
against a binary measure of democracy generated from the Polity2 scale, which is coded 1 when the score is
strictly positive along the -10 to +10 index. Our analyses suggest that these results are not robust (see Table 3).
77 Marshall & Jaggers 2007.
78 Freedom House 2007.
79 Vanhanen 2000.
80 Coppedge et al. 2008.
29
Democracy Scores (“UDS”).81 Among binary indices, we include the Democracy-
Dictatorship (“DD”)82, “BMR”83, “BNR”84, and “PT”85 indices.
Table 2 summarizes key features of these eleven indices, alongside the Lexical index
(for further details see Appendix B). It will be seen that coverage for the Lexical index –
including 17,054 country-year observations from 1800 to 2008 – is greater than or
comparable to all other indices. The final columns of Table 2 display correlations between
the Lexical index and extant indices. Although they co-vary, the correlations are far from
perfect, especially when the highest scoring cases (country-years in which Lexical=6) are
removed from the sample. In this sub-sample, the mean correlation between the Lexical
index and extant continuous indices is 0.52 (Pearson’s r) and the mean correlation with
extant binary indices is 0.35 (Spearman’s r). The lexical approach to electoral democracy is
not just another flavor of vanilla.
The consequences of these varying approaches to measurement are tested in Table 3.
In this set of tests we impose a common sample by focusing on the contemporary era and
imputing a full panel of observations for all sovereign states from 1960 to 2004 using the
Amelia II algorithm.86 Continuous indices are re-scaled from 0-1 (1=most democratic) so
that coefficients can be directly compared. Each regression model follows the specification
of Model 2 in Table 1: growth (from the WDI) is regressed against an index of democracy
81 Pemstein et al. 2010.
82 Cheibub et al. 2010.
83 Boix, Miller & Rosato 2013.
84 Bernhard, Nordstrom & Reenock 2001.
85 Persson & Tabellini 2006.
86 Honaker et al. 2011.
30
along with per capita GDP (from the WDI) and year and country fixed-effects. Results for
the Lexical index are displayed alongside results for extant indices, introduced above.
*=theoretical range. The final columns show the Spearman’s or Pearson’s correlation coefficient
between the Lexical index and other indices. Full samples include all available observations. Restricted samples include country-years in which Lexical<6.
Regression tests in Table 3 confirm that while the Lexical index is correlated with
growth most extant indicators of democracy are not. There is only one exception – Inclusive
(see Model 10). However, this result is probably driven by X/Y endogeneity or by an
unmeasured common-cause confounder. Further tests reveal that when Inclusive is lagged a
31
single period (one year) behind the outcome it loses statistical significance (see Table D4).
Additional robustness tests focused on other indices also reveal inconsistent relationships
between these indices and growth when alterations in sample, specification, or estimator are
R2 (within) (0.062) (0.062) (0.062) (0.063) (0.062) (0.062) (0.063) (0.062) (0.062) (0.064) (0.062) (0.064) Y = GDP per capita growth (WDI). Variables defined in Appendix B. All non-binary measures converted to a 0-1 scale so that coefficients can be directly compared. Units of analysis = country-years. Sample: all sovereign countries from 1960-2004 (N=6,836), missing data imputed with Amelia II (Honaker et al. 2011). All models include year and country fixed-effects. Estimator = ordinary least squares, standard errors clustered by country. *** p<.01, ** p<.05, * p<.1 (two-tailed test)
33
In order to make sense of these divergent findings we must enter into a detailed
discussion of measurement. Here, we shall identify features of the Lexical index that set it
apart from other indices. Note that each contrast has a somewhat different target – Feature 1
sets the Lexical index apart from extant indices A and B but not C, while Feature 2 sets the
Lexical index apart from indices B and C but not A. Considered together, however, these
features may explain the divergent outcomes displayed in Table 3 and in adjunct models
presented in Appendix D.
First, the coding of many democracy indices includes elements that are not strictly
electoral in character such as rule of law, civil liberties, conflict, corruption, civil society,
participation, and constraints on the executive. Although it is not the objective of this paper
to test alternative institutional sources of growth, analyses suggest that indices that stray
from the electoral core are not robustly associated with growth.
Second, the Lexical index recognizes important interdependencies among the
properties of electoral democracy. That is, the possession of one attribute is presumed to
affect the way in which other attributes function, and hence its implications for growth. A
classic instance of this is the interaction of multiparty competition and suffrage. Arguably,
the meaning and import of an institution like universal suffrage depends upon whether
multiparty competition is allowed. By contrast, continuous measures tend to combine
features of a regime that have no functional relationship to one another, creating “mashup”
indicators of development with no underlying theoretical rationale.87
Third, while binary indices recognize the interactive nature of political institutions
they reduce all relevant factors into a single dichotomous coding. This tosses out
87 Ravallion 2011.
34
information that may be invaluable for sorting out the relationship between electoral
democracy and growth. Of particular importance are distinctions at the lower end of the
scale, e.g., between a no-election regime such as Saudi Arabia (L0), a regime with single-party
elections such as North Korea (L1), a multi-party election regime that does not control the
choice of executive such as Jordan (L2), and a multi-party election regime that extends to the
legislature and executive but is not minimally competitive such as Rwanda (L3). These sorts
of distinctions, as well as those at the high end of the scale, are conflated by binary indices.
Although the findings of this paper contrast starkly with most extant studies, the
divergent findings are not surprising once one looks closely at the construction of various
indices. Choices in measurement are highly consequential, as previous studies have
demonstrated.88 With respect to growth, our findings suggest that the diverse institutions
associated with electoral democracy are best operationalized as a series of necessary
conditions, establishing the cumulative format of a lexical scale.
V. Mechanisms: An Initial Test
Our theory (Section I) suggests that electoral accountability is the chief characteristic
distinguishing democratic and autocratic regime-types. Accountability cannot be directly
observed, as discussed. Nonetheless, we ought to be able to observe its traces. Specifically, if
we observe an association between growth performance and subsequent leadership turnover
in democracies this may be interpreted as an indication that leaders are being rewarded
(punished) for good (bad) economic performance. By the same token, if we observe no such
association (or a much weaker association) in autocracies this confirms our conjecture that
88 Casper & Tufis 2003.
35
no such mechanism exists where conditions of multiparty competition are not present. And
if this differential pattern is present, we will have corroborated our assumption that
democratic leaders face greater incentives to pursue growth-augmenting (universalistic) policies
than autocratic leaders.
A connection between macroeconomic performance and the fates of incumbents in
democratic polities has been located by some studies89 but not others.90 However, all extant
studies are limited in their purview to democracies (variously defined) and to the postwar
era. Most focus on a single country or a small sample of OECD countries.
In order to expand this empirical terrain – to cover a longer period of historical time
and a larger sample of countries (including both sides of the political regime spectrum) – we
turn to a measure of leadership turnover provided by the Archigos dataset.91 The outcome of
interest is a binary measure of leadership change in the top executive office (1=turnover,
0=no turnover), measured annually for sovereign countries across the past century and a
half.
This measure of leadership turnover is regressed in a logit model against the growth
rate (calculated, as previously, from Maddison) along with covariates measuring per capita
GDP92, decade dummies (to capture time effects), and country fixed-effects. The model thus
estimates the probability of a leadership turnover in a given year conditional on covariates,
which are measured in the previous year to avoid simultaneity.
To distinguish between polities with elected leaders versus those without we examine
the disaggregated empirical tests of the Lexical index contained in Table C1 and Figure C1
for potential threshold effects. These analyses suggest that if there is a threshold effect this
effect is registered at the point at which executive and the legislature offices are subjected to
multi-party elections (L3). Accordingly, our sub-sample of non-elective regimes contains
those polities with a Lexical score of 0-2, while our sub-sample of elective (multi-party)
regimes is restricted to polities with a Lexical score of 3-6.
As one might expect, leadership changes are less common in the first group than in
the second group. The first sub-sample contains 741 leadership changes across 5,357
observed country-years, nested within 121 unique country-periods. The mean leadership
tenure in the subset without multi-party elections is approximately 10 years. The sub-sample
of elective regimes contains 1,592 leadership changes across 6,450 observed country-years,
nested in 139 unique country-periods, generating leadership spells that average just over 6
years.
Results of these sub-sample analyses are displayed in Table 4. Among countries
without multi-party elections (Model 1) we find that there is no discernible relationship
between growth and leadership turnover. Indeed, the estimated coefficient for growth is
almost precisely 0. By contrast, among polities with multi-party elections (Model 2), we find
a consistent, robust relationship whereby a decrease in growth rates increase the odd of
leadership tenure. The predicted probabilities estimated in Model 2 are plotted in Figure 1,
along with 95% confidence intervals.
37
Table 4: Growth and Leadership Change
1 2
Sample Autocracies Minimal electoral
democracies
Operationalization Lexical=0-2 Lexical=3-6
Growth (T-1) -0.000 -0.015*** [0.003] [0.005] GDPpc (ln) (T-1) 0.173 -0.156 [0.190] [0.137] Country FE X X Decade FE X X Years 1841-2004 1865-2004 Countries (N) 106 136 Obs (N) 4,086 5,633 Log likelihood -1332.908 -2541.446 Prob >chi2 0.000 0.000
Y: probability of leadership change (Goemans et al. 2010). Estimator: logit, maximum likelihood. *** p<.01, ** p<.05, * p<.1 (two-tailed test).
The estimated causal effect in Model 2 is modest, perhaps due to the fact that
economic growth is imprecisely measured across two centuries. Note also that electoral
defeat is a blunt instrument; incumbents may also be sensitive to the impact of growth
performance on their popularity (measured in the contemporary era by public opinion
surveys). In any case, it is noteworthy that the results shown in Model 2 are robust (a) when
the sample is restricted to the postwar era (1950-2004), (b) when different ways of modeling
time are adopted (e.g., year dummies, a linear trend variable, or no time-periods), (c) when
growth is lagged one or two-periods behind the outcome, (d) when growth is measured with
a three-year moving average, (e) when per capita GDP is omitted from the model, and (f)
when fixed- or random-effects estimators are employed.93
93 Since the analyses in Table 4 hinge on a binary distinction between political regimes it is not surprising to
discover that the point at which the cutoff is inscribed affects the results. Specifically, when the cutoff is moved
38
Figure 1:
-50 0 50 100 150 200
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Marginal effect of growth rates on leadership
growth rates
pred
icte
d pr
obab
ility
of t
urno
ver
up (to L4) or down (to L2) the relationships shown in Table 4 are attenuated. Likewise, when regimes are
classified by other binary indices (e.g., BNR, BMR, DD, PT) the relationship between growth and leadership
turnover among democratic polities is still negative but not always statistically significant. This dovetails with
analyses presented in Table 3, where we saw that democracy matters for growth but only when measured in a
lexical fashion.
39
Although we have by no means provided a full exploration of causal mechanisms, we
have corroborated a key expectation: relationships of accountability between leaders and
citizens exist with respect to growth performance in polities with a minimal degree of
political competition but not in polities without such competition. This suggests that sitting
dictators have nothing to fear from poor economic performance, for they are no more or
less likely to lose office when growth is weak. By contrast, poor growth performance lowers
the probability of an incumbent retaining office in a polity with multi-party elections.
More broadly, the evidence gathered in this study suggests that the structure of
incentives facing political leaders depends upon the institutions they are situated within.
Specifically, leaders subjected to multi-party elections should be more strongly motivated to
achieve growth than autocratic leaders. And this, in turn, may help to account for why we
see a difference in growth performance across different political regimes, as measured by the
Lexical index of electoral democracy.
40
VI. References
Acemoglu, Daron, and James A. Robinson. 2012. Why Nations Fail: The Origins of Power,
Prosperity, and Poverty. New York: Crown Publishers.
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2002. Reversal of Fortune:
Geography and Institutions in the Making of the Modern World Income Distribution.
Quarterly Journal of Economics 117(4):1231-1294.
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2005. Institutions as a
Fundamental Cause of Long-run Growth. Pp. 385-472 in Handbook of Economic Growth,
edited by Philippe Algion and Steven N. Durlauf. Amsterdam: North-Holland.
Acemoglu, Daron, Simon Johnson, James A. Robinson, and Pierre Yared. 2008. Income and
Democracy. American Economic Review 98(3):808-842.
Alesina, Alberto, and Roberto Perotti. 1997. The Politics of Growth: A Survey. Pp. 11-49 in
Government and Growth, edited by V. Bergstrom. Oxford: Clarendon.
Alvarez, R. Michael, Delia Bailey, and Jonathan N. Katz. 2011. An Empirical Bayes
Approach to Estimating Ordinal Treatment Effects. Political Analysis 19(1):20-31.
Amsden, Alice H. 1989. Asia’s Next Giant: South Korea and Late Industrialization. Oxford:
Oxford University Press.
Ashworth, Scott. 2012. Electoral Accountability: Recent Theoretical and Empirical Work.
Annual Review of Political Science 15:183–201.
Baier, Scott L., Gerald P. Dwyer, and Robert Tamura. 2006. How Important are Capital and
Total Factor Productivity for Economic Growth? Economic Inquiry 44(1):23-49.
Banks, Arthur S. 1994. Cross-National Time-Series Data Archive. Center for Social Analysis,
State University of New York at Binghamton. Binghamton, New York.
41
Barbieri, Katherine; Omar Keshk. 2012. Correlates of War Project Trade Data Set
Codebook, Version 3.0. Online: http://correlatesofwar.org.
Barro, Robert J. 1996. Democracy and Growth. Journal of Economic Growth 1(March):1-27.
Barro, Robert J. 1997. Determinants of Economic Growth. Cambridge: MIT Press.
Barzel, Yoram. 2002. A Theory of the State: Economic Rights, Legal Rights, and the Scope of the State.
Cambridge: Cambridge University Press.
Baum, Matthew A., and David A. Lake. 2003. The Political Economy of Growth:
Democracy and Human Capital. American Journal of Political Science 47(2):333-347.
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):634-647.
Benyishay, Ariel, and Roger R. Betancourt. 2010. Civil Liberties and Economic
Development. Journal of Institutional Economics 6(3):281-304.
Bernhard, Michael, Timothy Nordstrom, and Christopher Reenock. 2001. Economic
Performance, Institutional Intermediation, and Democratic Survival. Journal of Politics
63:775-803.
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. How Much Should We
Trust Difference-in-Differences Estimates? Quarterly Journal of Economics 119(1):249-275.
Besley, Timothy, and Masayuki Kudamatsu. 2006. Health and Democracy. American Economic
Review 96(2):313-318.
Blundell, Richard, and Stephen Bond. 1998. Initial Conditions and Moment Restrictions in
Dynamic Panel Data Models. Journal of Econometrics 87:115-43.
Boix, Carles, Michael Miller, and Sebastian Rosato. 2013. A Complete Dataset of Political
Regimes, 1800-2007. Comparative Political Studies 46:1523-1554.
Distribution of Countries Across the Lexical Index of Electoral Democracy, 1800-2008
64
Appendix B: Information about Variables
65
Table B1:
Variables, Definitions, Sources
DEMOCRACY INDICES
BMR. To qualify as democratic a country must satisfy the following conditions: “(1) The executive is directly or indirectly elected in popular elections and is responsible either directly to voters or to a legislature, (2) The legislature (or the executive if elected directly) is chosen in free and fair elections, (3) A majority of adult men has the right to vote” (Boix, Miller, Rosato 2013: 1531). Democracy_Boix
BNR. A regime is coded as democratic if it satisfies the polyarchy criterion (Dahl 1971), i.e., regimes “that permit a high level of contestation and include a large part of the adult population” (Bernhard et al. 2001: 783), with some recoding to convert the original “breakdown” variable into a democracy/non-democracy variable (authors). BNR_democracy
CL. Civil liberties, an index measuring the legal and practical protections of human rights (Freedom House 2007), reversed scale. Civil_liberty_FH_reverse
Contestation. An index derived from the first component of a principal components analysis including a large number of democracy indicators (Coppedge et al. 2008). cam_contest
DD. Democracy-dictatorship index. To qualify as democratic a country must satisfy the following conditions: “(1) The chief executive must be chosen by popular election or by a body that was itself popularly elected, (2) The legislature must be popularly elected, (3) There must be more than one party competing in the elections, (4) An alternation in power under electoral rules identical to the ones that brought the incumbent to office must have taken place” (Cheibub et al. 2010: 69). DD
Inclusive. An index derived from the second component of a principal components analysis including a large number of democracy indices (Coppedge et al. 2008). cam_inclusive
Lexical. Lexical index of electoral democracy, as described in text and Appendix A. Lexical
Lexical change. Coded 1 for any year in which a country’s score on the Lexical index changes from the previous year (including the first year of a series), 0 otherwise (authors). Lexical_change
Lexical stock. A country’s cumulative democratic history. Constructed by adding up a country’s score on the Lexical index from 1800 (or independence) to the present year, with a 1% annual depreciation rate (authors). Lexical_stock
Lexical diffusion. The mean level of Lexical for all countries within 500 kilometers of the country being coded (Gleditsch, Ward 2006). Lexical_a
PR. Political Rights, an index measuring the extent of political rights (Freedom House 2007), reversed scale. Pol_Rts_FH_reverse
Polity2. Polity2 index, combining Autocracy and Democracy variables, from the Polity IV dataset (Marshall, Jaggers 2007). Polity2
PT. Coded 1 for all years in which Polity2 > 0, and 0 otherwise (coded by authors but derived from Persson, Tabellini 2006). Polity2_dich_0
UDS. Unified Democracy Score, derived from an IRT model including a large number of democracy indicators (Pemstein et al. 2010). uds_mean
Vanhanen. Democracy index, the product of (1) the vote-share or seat-share of all but the largest party and (2) the share of adult population that voted (Vanhanen 2000). Democracy_Index_Vanhanen
OTHER VARIABLES
Capability. Capability index, combining iron/steel production, energy use, military expenditures, military personnel, and total and urban population (COW). capability_cow
English legal origin. English legal origin (La Porta et al 1999). English_legal_origin
66
European language. Percent speaking a European language (CIA WorldFactbook). European_language
GDP per cap (ln) (contemporary). GDP per capita (World Bank 2007), transformed by natural logarithm. GDPpc_ln_07
GDP per cap (ln) (historical). GDP per capita (Maddison 2010), missing data within a time-series interpolated and missing data for contemporary countries imputed from World Bank (2007), transformed by natural logarithm. GDPpc_imputed_Madd_ln
Growth (contemporary). GDP per capita growth (World Bank 2007). GDPpc_Growth_WDI_07
Growth (historical). GDP per capita growth calculated by authors from GDP per capita (historical), as described above. GDPpc_Madd_imp_Growth
Landlock. 1 if country is landlocked, 0 otherwise (Acemoglu, Johnson, Robinson 2002). Landlock
Latitude (ln). Distance from the equator, transformed by natural logarithm (authors). Latitude_ln
Population (contemporary). Total population (World Bank 2007), transformed by natural logarithm. Pop_Total_WDI_07_imp_ln
Population (historical). Total population (Maddison 2010), missing data within a time-series interpolated, transformed by natural logarithm. Pop_Madd_ipo_ln
Y = GDP per capita growth. Estimator = OLS (ordinary least squares) with standard errors clustered by country or Bayesian hierarchical (Alvarez et al. 2011). All models include year and country fixed effects. Samples include 186-187 countries observed from 1822-2004. *** p<.01, ** p<.05, * p<.1 (two-tailed test)
72
Figure C1:
Model Comparisons ─ Bayesian Shrinkage, Pooled and Unpooled models
The solid line depicts the linear trend from the pooled model (Model 1, Table 1). Estimates from the Bayesian shrinkage model (accompanied by 95% confidence intervals) are drawn relative to this trend. The triangles (jittered to the right for visibility) represent the estimates from the unpooled model (Model 4, Table C1).
73
Appendix D:
Additional Tests of Alternate Indices
74
Table D1:
Tests of the Polity2 Index 1 2 3 4 5 6 7 8 9 Growth Madd WDI Madd Madd Madd Madd Madd Madd Madd Period 1822-2004 1962-2005 1822-2001 1822-2001 1822-2004 1823-2004 1825-1995 1822-2004 1822-2004 Time-periods Annual Annual Annual Annual Annual Annual 5-year Annual Annual MI No No No No Yes No No No No Estimator OLS OLS OLS RE OLS OLS OLS OLS GMM Polity2 0.116*** -0.041 -0.004 0.017 0.100** 0.117*** 0.029 0.124*** [0.042] [0.026] [0.021] [0.021] [0.042] [0.041] [0.021] [0.037] Polity2 0.576 (T-5) [0.481] GDPpc (ln) -6.767*** -3.345*** -2.572*** -0.626** -5.070*** -7.023*** -2.275*** -6.621*** -8.107*** (T-1) [1.410] [0.637 [0.558] [0.267] [1.148] [1.389] [0.394] [1.493] [0.490] Urban -4.696* -2.581 [2.652] [1.726] Population -1.452*** -0.120 (ln) [0.526] [0.126] Capability 10.962*** 3.459** [3.693] [1.649] European 2.719*** language [0.606] English -0.106 legal origin [0.394] Latitude (ln) 0.113 [0.264] Landlock -0.486 [0.466] Yt-1 0.030** -0.007 [0.012] [0.007] Regional FE X Country FE X X X X X X X X Year FE X X X X X X X X X Countries (N) 165 153 152 149 212 165 162 163 165 Obs (N) 11,094 5,302 9,426 9,311 13,125 11,041 2,034 10,539 11,041 R2 (within) 0.131 0.094 0.099 0.090 0.095 0.131 0.198 0.129 Wald Chi2 1713.60 Y = GDP per capita growth. OLS = ordinary least squares analysis. RE = random effects. GMM = generalized method of moments (Blundell & Bond 1998). FE = fixed effects. MI = full dataset imputed with Amelia II (Honaker et al. 2011). Standard errors clustered by country. *** p<.01, ** p<.05, * p<.1 (two-tailed test)
75
Table D2:
Tests of the PT Index 1 2 3 4 5 6 7 8 9 Growth Madd WDI Madd Madd Madd Madd Madd Madd Madd Period 1822-2004 1962-2005 1822-2001 1822-2001 1822-2004 1823-2004 1825-1995 1822-2004 1823-2004 Time-periods Annual Annual Annual Annual Annual Annual 5-year Annual Annual MI No No No No Yes No No No No Estimator OLS OLS OLS RE OLS OLS OLS OLS GMM PT Index 1.293** 0.088 0.269 0.404 1.250*** 1.284** 0.438* 1.495*** [0.519] [0.324] [0.245] [0.252] [0.470] [0.514] [0.257] [0.471] PT Index 0.576 (T-5) [0.481] GDPpc (ln) -6.709*** -3.282*** -2.609*** -0.650** -5.064*** -6.956 -2.277*** -6.621*** -8.039*** (T-1) [1.399] [0.637] [0.555] [0.265] [1.144] [1.379] [0.401] [1.493] [0.489] Urban -4.758* -2.560 [2.654] [1.740] Population -1.439*** -0.121 (ln) [0.524] [0.127] Capability 11.171*** 3.480** [3.803] [1.626] European 2.699*** language [0.595] English -0.143 legal origin [0.391] Latitude (ln) 0.097 [0.265] Landlock -0.477 [0.467] Conflict
Yt-1 0.029** -0.007 [0.013] [0.007] Regional FE X Country FE X X X X X X X X Year FE X X X X X X X X X Countries (N) 165 153 152 149 212 165 162 163 165 Obs (N) 11,094 5,302 9,426 9,311 12,125 11,041 2,034 10,539 11,041 R2 (within) 0.131 0.093 0.099 0.090 0.095 0.131 0.198 0.129 Wald Chi2 1712.38 Y = GDP per capita growth. OLS = ordinary least squares analysis. RE = random effects. GMM = generalized method of moments (Blundell & Bond 1998). FE = fixed effects. MI = full dataset imputed with Amelia II (Honaker et al. 2011). Standard errors clustered by country. *** p<.01, ** p<.05, * p<.1 (two-tailed test)
76
Table D3:
Tests of the DD Index 1 2 3 4 5 6 7 8 9 Growth Madd WDI Madd Madd Madd Madd Madd Madd Madd Period 1947-2004 1962-2005 1947-2001 1947-2001 1947-2004 1947-2004 1955-1995 1952-2004 1947-2004 Time-periods Annual Annual Annual Annual Annual Annual 5-year Annual Annual MI No No No No Yes No No No No Estimator OLS OLS OLS RE OLS OLS OLS OLS GMM DD 0.736 0.097 -0.152 0.216 0.720 0.798 -0.222 0.698 [0.851] [0.330] [0.327] [0.292] [0.857] [0.844] [0.298] [0.777] DD (T-5) -0.561 [0.803] GDPpc (ln) -9.869*** -3.665*** -3.337*** -0.110 -8.897*** -10.273*** -2.946*** -10.506*** -11.937*** (T-1) [1.956] [0.659] [0.802] [0.252] [1.889] [1.937] [0.563] [2.326] [0.699] Urban -6.855* -1.829 [3.468] [1.716] Population -2.311** 0.034 (ln) [0.948] [0.135] Capability 11.843 -1.864 [26.406] [9.012] European 1.907*** language [0.552] English -0.247 legal origin [0.369] Latitude (ln) 0.081 [0.228] Landlock -0.332 [0.420] Yt-1 0.034*** -0.019** [0.011] [0.009] Regional FE X Country FE X X X X X X X X Year FE X X X X X X X X X Countries (N) 187 175 154 151 198 187 184 185 187 Obs (N) 7,772 5,865 6,491 6,392 7,930 7,742 1,289 6,936 7.742 R2 (within) 0.128 0.087 0.083 0.060 0.113 0.129 0.170 0.127 Wald Chi2 1206.81 Y = GDP per capita growth. OLS = ordinary least squares analysis. RE = random effects. GMM = generalized method of moments (Blundell & Bond 1998). FE = fixed effects. MI = full dataset imputed with Amelia II (Honaker et al. 2011). Standard errors clustered by country. *** p<.01, ** p<.05, * p<.1 (two-tailed test)
77
Table D4:
Tests of the Inclusive Index 1 2 3 4 5 6 7 8 9 Growth WDI WDI WDI WDI WDI WDI WDI WDI WDI Period 1962-2000 1962-2000 1962-2000 1962-2005 1962-2000 1965-2000 1962-2001 1962-2005 1963-2000 Time-periods Annual Annual Annual Annual Annual 5-year Annual Annual Annual MI No No No Yes No No No No No Estimator OLS OLS RE OLS OLS OLS OLS OLS GMM Inclusive 0.425** 0.406** 0.276* 0.337** 0.361** 0.099 0.490** [0.178] [0.179] [0.148] [0.147] [0.154] [0.147] [0.191] Inclusive 0.153 (T-1) [0.167] Inclusive -0.013 (T-5) [0.104] GDPpc (ln) -4.430*** -4.795*** 0.176 -2.634*** -4.771*** -4.451*** -3.993*** -3.168*** -7.362*** (T-1) [0.771] [0.903] [0.211] [0.458] [0.721] [0.750] [0.726] [0.686] [0.308] Urban -11.870** -2.071 [4.852] [1.915] Population -4.575*** -0.083 (ln) [1.376] [0.154] Capability 55.428 5.474 [55.411] [10.289] European 1.916*** language [0.625] English -0.333 legal origin [0.421] Latitude (ln) -0.089 [0.220] Landlock -0.181 [0.479] Yt-1 0.272*** 0.170*** [0.043] [0.011] Regional FE X Country FE X X X X X X X X Year FE X X X X X X X X X Countries (N) 171 142 140 198 171 170 171 173 171 Obs (N) 4,873 4,216 4,170 6,921 4,854 1,066 4,970 5,310 4,754 R2 (within) 0.088 0.110 0.069 0.065 0.148 0.211 0.079 0.070 Wald Chi2 1298.99 Y = GDP per capita growth. OLS = ordinary least squares analysis. RE = random effects. GMM = generalized method of moments (Blundell & Bond 1998). FE = fixed effects. MI = full dataset imputed with Amelia II (Honaker et al. 2011). Standard errors clustered by country. *** p<.01, ** p<.05, * p<.1 (two-tailed test)