WPS6336
Policy Research Working Paper 6336
Political Reforms and Public Policy
Evidence from Agricultural and Food Policies
Alessandro Olper
Jan Fałkowski
Johan Swinnen
The World Bank
Development Economics Vice Presidency
Partnerships, Capacity Building Unit
January 2013
Policy Research Working Paper 6336
Abstract
This paper studies the effect of political regime transitions subsidization, or both. The empirical findings are
on public policy using a new data set on global consistent with the predictions of the median voter
agricultural and food policies over a 50-year period model because political transitions occurred primarily
(including data from 74 developing and developed in countries with a majority of farmers. The results are
countries over the 1955–2005 period). The authors robust to different specifications, estimation approaches,
find evidence that democratization leads to a reduction and variable definitions.
of agricultural taxation, an increase in agricultural
This paper is a product of the Partnerships, Capacity Building Unit, Development Economics Vice Presidency. It is part
of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy
discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org.
The authors may be contacted at alessandro.olper@unimi.it, jfalkowski@wne.uw.edu.pl, and Jo.Swinnen@econ.kuleuven.be.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
Political Reforms and Public Policy:
Evidence from Agricultural and Food Policies
Alessandro Olper, Jan Fałkowski, and Johan Swinnen 1
JEL classification codes: D72, F13, O13, P16, Q18
Keywords: Democratic reforms, Agricultural and food policies, Comparative political
economics
Sector Board: Public Sector Governance (PSM)
1
Alessandro Olper (corresponding author) is an associate professor at the
University of Milano, Italy, and a researcher at LICOS; his email address is
alessandro.olper@unimi.it. Jan Fałkowski is an assistant professor at the University of
Warsaw, Poland, and a researcher at CEAPS; his email address is
jfalkowski@wne.uw.edu.pl. Jo Swinnen is the director of LICOS, a professor at the
University of Leuven, Belgium, and a visiting professor at Stanford University; his email
address is Jo.Swinnen@econ.kuleuven.be. The authors would like to thank the
participants at several conferences and meetings and the editor and three anonymous
referees for the World Bank Economic Review for their useful comments and suggestions.
Alessandro Olper and Jo Swinnen are grateful for financial support from the KU Leuven
Research Council (OT, EF, and Methusalem projects). Jan Fałkowski gratefully
acknowledges financial support from the Polish Ministry of Science and Higher Education
(grant no. NN112118639).
Across the world, agricultural and food policies distort incentives for farmers and
food consumers. Historically, governments in wealthy countries have subsidized farmers,
whereas governments in poor countries have taxed farmers and subsidized food
consumers (Anderson and Hayami 1986; Krueger et al. 1988; Anderson 1995). These
observations have puzzled economists and other social scientists and triggered a series of
studies in the 1980s and 1990s on “the political economy of agricultural policies� (see de
Gorter and Swinnen 2002; Swinnen 2010 for reviews).
A recent global study on policy distortions in agriculture concludes that although
policy distortions remain important, since the 1980s, the antiagricultural policy bias in
developing countries and the proagricultural bias in high income countries have declined
substantially (Anderson 2009). Interestingly, this was also the period during which
important political reforms occurred in many countries. For example, the fall of the Berlin
wall triggered democratic transitions across Eastern Europe in the early 1990s.
Furthermore, several developing and emerging countries have become more democratic in
recent decades. In Eastern Europe, political reforms induced radical economic
liberalizations in the food system. In contrast, in the absence of major political reforms in
East Asia, gradual economic liberalization was introduced, including the reduced taxation
of farmers (Swinnen and Rozelle 2006).
Thus, the question arises whether and to what degree political reforms have
affected agricultural and food policies worldwide. Political economy studies have
demonstrated the importance of a variety of factors, such as economic structural factors
and resource endowments, but the role of political institutions and reforms has received
less attention. A few studies have attempted to analyze this issue, but the evidence on the
impact of political reforms on agricultural and food policies is not clear because of
problems with the data (see section II for references and details).
2
This paper employs a novel data set on agricultural distortions that was recently
developed by the World Bank (see Anderson and Valenzuela 2008). This data set offers
consistent and comparable protection indices for a large number of countries over a 50-
year period. Employing this data set allows us to take advantage of not only cross-country
variation but also within-country variation in the data and thus to overcome the strong
identification assumptions that characterize previous cross-country studies.
Because the relationship between democracy and public policy may conceal
potential feedback effects, we also study the reverse causality problem and exploit the
timing of democratization. To control for potential nonlinearities and to better address
unobserved heterogeneities, we estimate both linear specifications and semiparametric
models. Specifically, we study the effects of democratic reform using the difference-in-
difference (D-in-D) technique combined with propensity score matching methods, as in
Persson and Tabellini (2008).
By studying agricultural and food policies, our analysis contributes to a broad
body of literature on the impact of political reforms on economic policies (see Rodrik and
Wacziarg 2005; Giavazzi and Tabellini 2005; Eichengreen and Leblang 2008).
The remainder of this paper is organized as follows. Section I presents stylized
facts on trends in agricultural and food policies over time and across political regimes.
Section II discusses conceptual issues and summarizes previous findings linking political
reforms to public policies. Section III presents our empirical strategy. Section IV presents
the data and key variables. In section V, the empirical results are presented and discussed,
and in section VI, we test the robustness of our findings. Finally, section VII concludes.
3
<>I. POLICY INDICATORS AND STYLIZED FACTS
We employ two different indicators of agricultural and food policies: the nominal
rate of assistance (NRA) to agriculture and the relative rate of assistance (RRA), both from
the World Bank’s agricultural distortions database (see Anderson and Valenzuela 2008,
for calculation details). This database reports the most consistent and comparable
estimates of agricultural protection across countries and over time. In our econometric
analysis, we use a sample of 74 countries, comprising yearly data from 1955 to 2005 (see
table S.1 in the supplemental appendix, available at http://wber.oxfordjournals.org/). The
average number of observations per country is 35. We work with an unbalanced panel
with more than 2,600 observations.
The NRA measures total transfers to agriculture as a percentage of the undistorted
unit value. The NRA for agriculture is obtained as a weighted average of the NRA at the
product level, using the undistorted value of production as a weight. The NRA is positive
when agriculture is subsidized, negative when it is taxed, and zero when net transfers are
zero. The NRA includes both the assistance provided by all tariff and nontariff trade
measures applied to agricultural products and any domestic price-distorting measures. 1
The price equivalent of any direct intervention regarding inputs or outputs is also
included. 2
To account for the protection of manufacturing sectors, which is an important
source of indirect taxation on agriculture, especially in developing countries, we use both
the NRA and the RRA, which is calculated as the ratio of agricultural NRA to
nonagricultural NRA. 3 The RRA is a useful indicator for international comparisons of anti-
or proagricultural policy regimes. There are fewer observations for the RRA because the
country and time-series coverage is smaller than for the NRA. Specifically, the RRA data
contain five fewer countries (69 instead of 74).
4
Table 1 summarizes the NRA and RRA for democracies and autocracies (see below
for definitions). The table indicates that autocracies are associated with negative levels for
both NRA and RRA, whereas democracies have positive levels. Moreover, the differences
are significant. The average NRA (RRA) is −0.15 (−0.26) for autocracies and +0.45
(+0.31) for democracies, a difference of 0.60 (0.57) or 60 (57) percentage points.
TABLE 1. NRA and RRA over Time and across Political Regimes
Full sample Autocracy Democracy
NRA RRA NRA RRA NRA RRA
1956–1959 0.41 0.18 −0.13 −0.29 0.66 0.41
1960–1964 0.28 0.08 −0.16 −0.30 0.54 0.30
1965–1969 0.27 0.07 −0.13 −0.27 0.51 0.27
1970–1974 0.10 −0.01 −0.24 −0.33 0.46 0.26
1975–1979 0.10 0.02 −0.23 −0.31 0.44 0.31
1980–1984 0.09 0.03 −0.22 −0.29 0.38 0.28
1985–1989 0.29 0.20 −0.06 −0.22 0.59 0.47
1990–1994 0.23 0.18 −0.14 −0.23 0.41 0.37
1995–1999 0.19 0.15 −0.13 −0.19 0.28 0.23
2000–2005 0.20 0.16 −0.08 −0.20 0.26 0.21
All years 0.21 0.11 −0.15 −0.26 0.45 0.31
No. of 74 69 38 34 67 64
countries
Source: Own calculations based on the data described in the text.
Notes: The figures report NRA and RRA averages for the full sample, autocracies, and democracies in different
subperiods. The number of countries refers to the total number in each category in the 1955–2005 period and changes
over time due to entry and exit.
Although these statistics demonstrate that the average NRA and RRA values are
much higher in democracies than in autocracies, they say nothing about the potential
causal effect of democratization on agricultural policies. To obtain further insight on this
issue, we examine the NRA and RRA values of countries in the data set that have
experienced a political transition from autocracy to democracy. Specifically, figures 1 and
2 present the average NRA and RRA values in the predemocratization and
postdemocratization periods for 23 countries that have experienced permanent
democratization, as defined by Papaiannou and Siourounis (2008). 4
5
Figure 1. Average NRA and the timing of political reforms.
.
30
Nominal rate of assistance
0 10
-10 20
-10 -5 0 5 10
Years relative to democratisation
NRA 3-year moving average
average predemocratisation average postdemocratisation
Source: Own calculations based on data from the World Bank’s agricultural distortions database
and the Polity IV database
Figure 2. Average RRA and the timing of political reforms.
20
Relative rate of assistance
-10 0-20 10
-10 -5 0 5 10
Years relative to democratisation
RRA 3-year moving average
average predemocratisation average postdemocratisation
Source: Own calculations based on data from the World Bank’s agricultural distortions database
and the Polity IV database.
6
The figures reveal interesting patterns. First, both the average NRA and RRA
values are relatively stable during the decade prior to a democratic transition,
approximately −5 percent for NRA and approximately −13 percent for RRA. Second, the
average NRA and RRA values are significantly higher following democratic reform. The
average NRA in the decade after democratization is 13 percent, or 18 percent higher than
in the decade before democratization. For the RRA, the increase is 16 percent (from −13
percent to +3 percent). Third, the figures suggest that there is both an immediate effect at
about the time of democratization and an additional increase approximately five years
later.
In summary, these descriptive statistics indicate interesting correlations between
agricultural policies and political regimes, both across countries and over time. In the
remainder of this paper, we use econometric methods to analyze whether there is a causal
relationship.
<>II. CONCEPTUAL ISSUES AND LITERATURE
There is a substantial body of literature on how political reforms influence
government policies (Mulligan et al. 2004). However, theory does not provide a simple
prediction of how democratic reforms affect agricultural protection. In democracies, the
distribution of political power is typically more equal than the distributions of income and
wealth. Consequently, median voter models predict that democracies tend to redistribute
from the rich to the poor, and this effect is stronger with greater income inequality
because the middle class has a greater incentive to form coalitions with the poor (Alesina
and Rodrik 1994; Persson and Tabellini 1994). Similarly, democratic regimes may lead to
economic policy reforms if these reforms create more winners than losers (Giavazzi and
Tabellini 2005).
7
Empirical studies have attempted to test this prediction using data on democracy
and economic liberalization. An area that has attracted substantial interest is trade policy.
Overall, the existing literature suggests a positive impact of democracy on economic
(trade) liberalization (e.g., Banerji and Ghanem 1997; Milner and Kubota 2005; Giavazzi
and Tabellini 2005; Eichengreen and Leblang 2008; Giuliano et al. 2011). Some studies,
however, have argued that this effect is not generally true but depends on a country’s
resource endowment (e.g., O’Rourke and Taylor 2007; Kono 2006).
There are several methodological critiques of these studies, such as the problem of
spurious correlation between democracy and economic reforms (Eichengreen and Leblang
2008) or the existence of potential feedback effects (Giavazzi and Tabellini 2005; Milner
and Mukherjee 2009). An additional problem is that most existing studies have examined
the relationship between democracy and trade policy using aggregate trade indices, such
as the trade to GDP ratio or the Sachs and Warner (1995) openness index (e.g., Giavazzi
and Tabellini 2005; Milner and Kubota 2005; Persson 2005; Eichengreen and Leblang
2008; Tavares 2007). Studies have only rarely employed direct indicators of trade policy,
such as tariffs. Moreover, aggregated trade policy indicators may be misleading because
different (and possibly offsetting) effects may occur at a disaggregated level (Anderson
and Martin 2006). Thus, an examination of disaggregated policies, such as agricultural
and food policies, could yield additional insights.
Empirical studies have estimated the impact of political institutions on agricultural
policies. Lindert (1991) was the first to document a positive impact of democracy on
agricultural protection. Beghin and Kherallah (1994) examine the impact of different
political systems (no-party, one-party, dominant party, and multiparty systems) on
agricultural protection. They find that political institutions are important and that their
effect is nonmonotonic: protection peaks with dominant party systems and then becomes
8
nonincreasing despite further democratization. A nonmonotonic relationship between
democracy and protection is also found by Swinnen et al. (2000), who uses the Gastil
index of political rights. Specifically, they demonstrate that moving from low to medium
levels of political rights reduces protection, but any further increase in democratization
does not necessarily result in substantial effects on agricultural protection. However, this
nonlinear behavior runs in the opposite direction of that found by Beghin and Kherallah.
Olper (2001) finds that the level of democracy per se does not seem to matter, but the
quality of institutions that protect and enforce property rights is important.
Although these studies highlight a number of interesting aspects, they should be
interpreted with caution. The studies all have potential problems of reverse causality and
omitted variable bias because they rely predominantly on the between-country variation in
the data. Their data sets do not allow for the exploitation of time series variation. To date,
the only study to investigate the relationship between democracy and agricultural
protection by employing a long time series is the study by Swinnen et al. (2001), which
examines agricultural protection patterns in Belgium between 1877 and 1990 and uses
detailed indicators of political reforms. Their paper demonstrates that only those political
reforms that generate a significant shift in the political balance toward agricultural
interests (e.g., the extension of voting rights to small farmers in the early 20th century)
induce an increase in agricultural protection. This result provides a logical interpretation
of the democracy-protection nonlinearity discussed above and highlights the importance
of drawing inferences from autocratic-democratic regime changes to improve
understanding of the impact of democratization on agricultural protection.
An additional problem is that the absence of representative information on the
preferences of autocratic rulers complicates predictions of the effect of democratization.
The insulation of decision makers means that they can follow their private preferences to a
9
large extent when selecting policies. However, this argument has little predictive power in
the absence of information on autocrats’ preferences. The preference of rulers is a key
variable, but there are major data and measurement problems. For example, quantitative
data exist on ideologies, but these data are limited to democracies. 5,6
Assuming that rulers’ preferences are randomly distributed, 7 the median voter
model predicts that the impact of democratization is conditional on the structure of the
economy. The share of farmers (or the rural population) in the economy differs
significantly between rich and poor countries. The factors that make it difficult for farmers
to organize politically in poor countries (such as their large number and substantial
geographic dispersion; see Olson 1965) render them potentially powerful in electoral
settings because they represent a large share of the votes (Bates and Block 2010;
Varshney 1995). Therefore, ceteris paribus, one would expect that democratization is
more likely to benefit farmers in poor countries.
In our data set, the vast majority of transitions from autocracy to democracy occur
in poorer countries with a large number of farmers. 8 In fact, the average share of
agriculture in total employment at the time of political transition is 65 percent, whereas
the average share for all countries and time periods in the data set is 25 percent. This
finding implies that the measured effect of democratization on agricultural policies in our
data set should be in favor of farmers (i.e., a positive impact on NRA and RRA) because of
the structural “bias� of political reforms. The move from autocracy to democracy
primarily occurs in countries where farmers constitute the majority of the population, and
the median voter model predicts that this situation should induce a profarmer policy
effect.
10
<>III. EMPIRICAL METHODOLOGY
To address the problems of omitted variable bias and reverse causation in the
analysis of the effect of political institutions on policies and to make use of both cross-
country variations and time variations in the data, we use a D-in-D strategy, as in recent
studies (e.g., Giavazzi and Tabellini 2005; Rodrik and Wacziarg 2005). To analyze the
robustness of our results, we combine the standard D-in-D approach with semiparametric
matching methods, as in Blundell et al. (2004) and Persson and Tabellini (2008).
Following Giavazzi and Tabellini (2005), we define regime changes as a
“treatment� experienced by some countries but not by others. Then, we estimate the effect
of the treatment through a D-in-D regression. In this way, we are able to exploit both the
time series and cross-sectional variation in the data. We refer to countries that experience
a regime change in the observed period as treated countries and to countries that do not
experience a regime change as control countries. In the regressions, we compare
agricultural policies in the treated countries before and after the treatment with
agricultural policies in the control countries over the same period.
More formally, we run panel regressions 9 with the following specification:
(1) yit = βDit + Ï?X it + α i + θ t + ε it ,
where yit denotes our measure of interest, namely, agricultural policies measured by NRA
and RRA; αi and θt are country and year fixed effects, respectively; Xit is a set of control
variables; and Dit is a dummy variable that takes the value one for democracy and zero
otherwise (see section IV). The parameter β is the D-in-D estimate of the regime change
effect. It is obtained by comparing the average protection after a regime change, minus
protection before the transition in the treated countries, to the change in protection in the
control countries over the same period. Here, the control countries are those that do not
11
experience a transition into or out of democracy—that is, those that have either Dit = 1 or
Dit = 0 over the entire sample period.
Estimates obtained from the standard D-in-D procedure are based on several
restrictive assumptions (see Abadie 2005; Persson and Tabellini 2008). First, it is assumed
that, absent any regime change, the average growth in protection in the treated countries
should be the same as in the control countries. 10 Second, the estimates do not take into
account the (potential) heterogeneity of regime change effects on agricultural policies. 11
Finally, the estimates may suffer from omitted variable bias due to time-varying (country-
specific) covariates correlated with both democracy and policies.
To address the latter problem, in addition to the traditional controls, we include in
our specifications several time-varying, country-specific variables. Furthermore, given our
specific concern for (omitted) time-varying factors, we add continent-year interaction
effects in some specifications. This process takes into account that changes in agricultural
policies may be due to general developments in geographical clusters. Finally, we check
the robustness of our results by running dynamic panel models.
To circumvent the heterogeneity of regime change effects, the existing literature
interacts the political reform dummy with other characteristics of reforms, such as specific
electoral rules or forms of government implemented by the new democracy (see Persson
2005; Olper and Raimondi Forthcoming). However, the problem with this approach is that
the potential interactions or nonlinearities are too numerous compared with the regime
transitions in the data. Therefore, we use semiparametric methods to address these
problems; that is, we combine a D-in-D methodology with a propensity score matching
method, following the approach discussed by Smith and Todd (2005) and Abadie (2005)
and applied by Blundell et al. (2004) and Persson and Tabellini (2008). This method has
two main advantages over the standard D-in-D estimator. First, it ensures that the
12
pretreatment characteristics that are thought to determine the outcome variable are
balanced between the treated and untreated countries. Thus, this method relaxes the strong
identifying restriction of the standard approach (Abadie 2005). Second, it relaxes linearity
assumptions by allowing for heterogeneous impacts of democratic transitions on
agricultural policies.
Our matching cum D-in-D strategy is implemented in two steps. First, to avoid
confounding the effect of political regime transition with that of factors that determine this
shift and because we cannot observe what would have happened if a democratic country
had remained an autocracy, an estimate of the counterfactual is constructed. Conditional
on the number of observable characteristics, the probability of regime change is calculated
for each country (i.e., the propensity score). Based on this estimate, the next step involves
an evaluation of the difference in the evolution of agricultural policies between countries
with and without a regime change. Because matching relies on comparing countries with
similar propensity score values, the inferences are not distorted by counterfactuals that
differ substantially from the treated observations.
The average estimated effect of regime transitions that we compute (the so-called
average treatment on treated, ATT) can be presented as follows:
(2) ATT =
1
I i
(
∑ ai − ∑ j wij aij)
,
where I stands for the number of treated observations within the common support; ai is the
difference between the average level of agricultural protection before and after the
transition in the treated country i; aij is the difference between the average level of
agricultural protection in the control country j over the periods before and after the
transition in the treated country with which it is matched; and wij (wij>0 and ∑j wij=1) are
weights based on the propensity score that depend on the matching estimator (Sianesi
13
2001). We use Epanechnikov kernel and Gaussian kernel estimators (Fan 1992; Heckman
et al. 1998).
<>IV. POLITICAL REFORM INDICATORS AND CONTROL VARIABLES
To study how a regime transition toward democracy affects agricultural and food
policies, we need data on democratization episodes. Unfortunately, although various
democracy data sets exist, none of these data sets provides a specific coding of regime
transitions. Therefore, we follow the same strategy as recent studies that have investigated
similar questions at the aggregate level by relying on the Polity2 index from the Polity IV
data (Marshall and Jaggers 2007). 12 The composite Polity2 index assigns a value ranging
from −10 to +10 to each country and year, with higher values associated with better
democracies on the basis of several institutional characteristics, such as the openness of
elections or constraints on the executive branch. Following Persson (2005) and Giavazzi
and Tabellini (2005), we code a country as “democratic� in each year that the Polity2
index is strictly positive, setting a binary indicator called democracy to one (zero
otherwise). A reform into (or out of) democracy occurs in a country-year when this
democracy indicator switches from zero to one (from one to zero).
A potential shortcoming of this definition of political reform is that being near any
particular divide may differ from being far from the divide. 13 Indeed, the threshold of zero
for Polity2 corresponds to a generous definition of democracy. However, as emphasized
by Persson and Tabellini (2008) and others, this definition has the important advantage
that many large changes in the Polity2 score are clustered around zero, an important
property given our identification strategy based on the within-country variation in the
data. Consequently, using a higher threshold for the definition of democracy has the
14
shortcoming of including very (small) gradual changes that are only poorly related to
significant regime changes in democratic transitions. 14
Applying these criteria to our 74-country data set, we obtain 67 regime changes, of
which 42 are transitions into democracy and 25 are transitions into autocracy (see table
S.1). The distribution of these reforms is uniform over time (53 percent before 1985) but
not across continents: approximately 50 percent of the reforms are in Africa, 28 percent
are in Asia, and 18 percent are in Latin America.
To avoid the use of very brief reform episodes, we introduce the criterion that the
dependent variable must be observed for at least four years before and after each regime
transition. Under this rule, the effective number of reform episodes decreases to
approximately 40. As a robustness check, we relax this criterion to only two years of
observable outcomes; this period includes nearly all of the reform episodes reported in
table S.1.
To check the robustness of our results, we use a distinct definition of regime
transitions. Specifically, we use the recently developed data set by Papaioannou and
Siourounis (2008) to define regime changes. These data are based on a more complex
procedure than that applied above. Specifically, to identify the precise timing of each
regime change, Papaioannou and Siourounis (2008) rely not only on the Polity2 index and
the Freedom House democracy index but also historical evidence derived from numerous
political archives and election databases. Using this procedure, the authors identify “full�
or “partial� democratization episodes. However, because their analysis focuses solely on
permanent democratization, the use of their coding applies to a lower number of transition
episodes (23) in our data (see table S.1).
15
<>Control variables
In the empirical specification, we include additional controls that are likely to
affect agricultural and food policies, as suggested by many previous studies (e.g.,
Anderson 1995; Beghin and Kherallah 1994; Swinnen et al. 2000; Olper 2007).
Specifically, our basic D-in-D specification always includes the following structural
controls: the level of development, measured by the log of real per capita GDP; the share
of agricultural employment in total employment; the log of agricultural land per capita;
and the log of total population. All of these variables are computed from World Bank
(WDI), FAO, or national statistics.
We also test the robustness of our findings by controlling for several other (macro)
covariates, such as different indicators of (aggregate) openness, government expenditures,
and economic and political crises (wars and conflicts). Openness indicators (the trade to
GDP ratio and the Sachs and Warner (1995) index) and government expenditures are
obtained from Wacziarg and Welch (2008) 15 and the Penn World Table, respectively. War
and conflict year dummies are based on the UCDP/PRIO Armed Conflict Dataset Version
4-2008 (see Gleditsch et al. 2002).
For our matching strategy, we use a limited number of covariates that are likely to
influence both regime change and agricultural and food policies. As discussed previously,
a shift in agricultural policy may require political reforms of sufficient size (Swinnen et
al. 2001). Therefore, in our model, we include a variable, initial polity2, that takes the
value of our democracy index at the beginning of the sample. This variable is included to
take into account that countries with Polity2 values close to zero are more likely to have a
political regime change.
To control for the fact that the sample period varies in length across countries and
that the length of the sample may be correlated with the probability of changes in the
16
political regime, we include the variable length of sample (measured in years). This
variable is designed to account for the possibility that democratization may require time to
have an impact. Furthermore, to control for the fact that changes in both agricultural
policy and political regime may be related to economic development, we include the
variable relative gdp, which measures each country’s per capita income at the beginning
of the sample relative to U.S. per capita income in the same year. Finally, to control for
the possibility that the change in political regime may be related to the occurrence of
conflicts (both domestic and international), we include the variable conflict years, which
measures the share of conflict years over the total length of the period for which policy
data are available.
<>V. ESTIMATION RESULTS
Table 2 reports D-in-D econometric results with the NRA and RRA as dependent
variables. Columns 1 and 5 report “unconditional� democracy effects by adding only the
level of development to the vector of covariates X to control for the well-known positive
correlations between per capita GDP and both democracy and agricultural protection. In
columns 2–4 and 6–8, we analyze the democratization effect using regressions controlling
for both the standard determinants of agricultural protection and macroeconomic and trade
policy. In all regressions, the standard errors are clustered at the country level, allowing
for arbitrary, country-specific serial correlation (see Bertrand et al. 2004). 16 Because the
fixed effects and other covariates are correlated, we only report the fixed effects results
(Mundlak 1978; Mundlak and Larson 1992). 17
17
TABLE 2. Effect of Democratic Reforms on Agricultural Protection
Estimation D-in-D regressions
Regression (1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable NRA NRA NRA NRA RRA RRA RRA RRA
Democratic reform 18.560 16.272 13.997 13.274 13.183 11.206 9.506 10.378
(.001) (.001) (.003) (.005) (.006) (.016) (.036) (.016)
Log GDP per capita 32.919 48.717 42.461 45.935 34.518 39.014 35.076 41.540
(.011) (.000) (.002) (.001) (.004) (.008) (.005) (.000)
Employment share −88.857 −94.180 −61.086 −65.281
(.107) (.082) (.324) (.276)
Land per capita −2.392 −2.484 −1.180 −1.292
(.097) (.125) (.371) (.348)
Log population −28.825 −32.349 −3.249 −16.952
(.410) (.340) (.931) (.622)
Trade policy reform (Sachs-Warner)
(.002) (.000)
Trade openness −0.065 −0.053
(.278) (.390)
Government consumption −0.213 0.616
(.583) (.176)
Treatment All All All All All All All All
Time fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Continental trends No Yes Yes Yes No Yes Yes Yes
Countries 74 74 74 72 69 69 69 67
Observations 2,664 2,664 2,565 2,502 2,394 2,394 2,314 2,253
2
R (within) 0.184 0.323 0.338 0.359 0.230 0.339 0.351 0.387
Source: Own calculations based on the data described in the text.
Note: p values based on clustered standard errors at the country level in parentheses. Year and country fixed
effects as well as interaction effects between continents (Africa, Asia, and Latin America) and year
dummies are included as indicated. The democracy variable is based on the Polity2 index (see text).
18
All specifications yield positive estimates of the democracy coefficient. The
significance varies between the 1 percent and 5 percent levels. The magnitude of the
democracy variable in column 1 suggests that a transition from autocracy to democracy
induces a strong effect: the NRA increases, on average, by 18.6 percentage points. In
column 2, we add a set of continent-year interaction effects to control for both differences
in regional protection dynamics and the nonstationary nature of the democracy dummy. 18
Although their inclusion slightly reduces the democracy coefficient, it remains significant
at the 1 percent level. Columns 3 and 4 test the robustness of our findings by including a
set of covariates normally found to be significant determinants of agricultural protection
(in column 3) and the share of government consumption expenditures in GDP 19 and two
different openness variables: the trade to GDP ratio and a trade policy reform index based
on Sachs and Warner (1995) (in column 4). The democracy effect is still estimated with
strong precision (p < .01). The magnitude of the estimated effect is very similar in both
equations and slightly lower than in columns 1–2. The effect on NRA is now
approximately 14 percentage points. These results suggest that the positive effect of a
regime change on the NRA is very robust. The estimated coefficients of the other variables
are consistent with expectations from the agricultural protection literature. 20
Columns 5–8 are analogous using the RRA as the dependent variable. The results
are similar, but the sizes of the effects and the precision of the estimates are somewhat
smaller. The magnitude of the estimated effect of reforms into democracy on the RRA is
10–13 percentage points, depending on the model. The small difference (4 percentage
points) between the NRA and RRA regressions suggests that the bulk of the democracy
effect comes from changes in agricultural policies. 21
An interesting hypothetical question is what the level of agricultural protection
would be if all countries were democracies. 22 In our sample, autocracies are only present
19
in Africa and Asia at the end of the data period. The issue is most relevant for Africa
because 10 out of 22 countries were still autocracies, whereas only 3 out of 11 were still
autocracies in Asia. A simple prediction based on average effects (an increase from 14
percent to 18 percent for NRA) and the use of 2000 as the base year (the year for which we
have the largest recent country sample) yields the following: with an average NRA of −15
percent in Africa in 2000, ceteris paribus, a hypothetical democratization wave would
induce a reduction in the average level of taxation of 6 to 8 percentage points, resulting in
an average NRA of −7 percent to −9 percent and effectively halving agricultural taxation.
In Asia, the average effect of hypothetical democratization is smaller because fewer
autocracies remain. The effect depends on whether a simple average or a weighted
average is used; China is one of three remaining autocracies in the data set. The simple
average effect on NRA is an increase of between 4 and 5 percentage points for Asia,
whereas the population weighted average protection effect is an increase of 7 to 9
percentage points (from a weighted average NRA of approximately 9 percent).
<>VI. EXTENSIONS AND ROBUSTNESS CHECKS
To further test whether our results capture a causal effect of democratization on
agricultural policies, we run several extensions of the model and robustness checks.
Specifically, in this section, we analyze how the results are affected by considering or
using (a) disaggregated commodities, (b) different indicators to capture the timing of
political reforms, (c) alternative estimation models (matching, dynamic panels, feedback
effects), (d) alternative definitions of regime changes, and (e) additional indicators of
economic and political crises. For brevity, some of these additional regression results are
reported in the supplemental appendix.
20
<>Disaggregated commodities
The results reported in table 2 are based on an aggregated measure of protection.
However, various sectors are taxed and subsidized differently for a number of reasons,
including differences in demand and supply conditions and because these sectors are
characterized by different market structures (e.g., small vs. large farms), which influences
rent-seeking behavior. The heterogeneous nature of agricultural protection may also cause
an aggregation bias in measures such as the NRA (see Aksoy 2005). To investigate
potential heterogeneity in the political reform effects across different groups of
commodities, table 3 reports regression results by separating importing and exporting
sectors (columns 1 and 2) and four commodity groups (columns 3–7). 23
TABLE 3. Effect of Democratic Reforms on Agricultural Protection at the Sector Level
Estimation D-in-D regressions
Products/sectors Import Export Grains Livestock Oilseeds Tropical
sectors sectors and tubers products crops
(1) (3) (4) (5) (6) (7)
Dependent variable NRA NRA NRA NRA NRA NRA
Democracy 17.4 6.8 17.4 5.1 31.0 7.1
(.007) (.026) (.000) (.632) (.050) (.367)
Treatment All All All All All All
Controls Yes Yes Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes Yes Yes
Continental trends Yes Yes Yes Yes Yes Yes
R2 (within) 0.239 0.146 0.228 0.279 0.391 0.424
Country-sectors 519 440 269 238 80 112
Observations 13,278 9,558 8,932 6,920 2,510 3,869
Source: Own calculations based on the data described in the text.
Notes: p values based on robust standard errors clustered at country-sectors in parentheses. Additional
controls include the log of per capita GDP, the log of population, agricultural employment share, land per
capita, and product value shares. All regressions include time and country fixed effects and interactions
between years and continental dummies (for African, Asian, and Latin American countries).
21
The disaggregated regressions demonstrate that democratization increases the NRA
for all subsectors, but the magnitude of the estimated effect differs: it is higher
(approximately 17.4 percentage points) for import-competing sectors than for exporting
sectors (6.8 percentage points). Similarly, the democratization effect is positive for the
four different product groups, but it is much higher for grains and tubers (17 percent) and
oilseeds (31 percent) than for livestock products (5 percent) and tropical crops (7 percent).
<>Timing of political reforms
A potential shortcoming of our findings is that we have constrained the
democratization effect to be monotonic (Papaioannou and Siourounis 2008). Relaxing this
assumption could yield additional insights into the dynamics of this effect. Following
Giavazzi and Tabellini (2005) and Wacziarg and Welch (2008), we investigate these
issues by studying the timing of the reform effects. To do so, we replace the variable
democracy with three nonoverlapping dummies: a dummy equal to one in the three years
preceding the regime change (3 years before), a dummy equal to one in the year of the
reform and in the three following years (years 0–3), and a dummy equal to one from the
fourth year after the regime change and onward (years 4 and after). The 3 years before
dummy aims to account for potential positive changes in agricultural protection before the
democratic transition. For example, it is possible that an autocratic government may
implement protectionist policies to gain legitimacy and remain in power.
Table 4 presents the results for the NRA and RRA. The estimated effect of the 3
years before dummy is negative, except in column (2), but it is never significant. This
finding suggests that agricultural policies do not change prior to democratization. Thus,
our results do not support the hypothesis that the anticipation of the democratization
process is reflected in changes in agricultural protection.
22
TABLE 4. Timing of Political and Agricultural Policy Reforms
Estimation D-in-D regressions
Regression (1) (2) (3) (4)
Dependent variable NRA NRA RRA RRA
3 years before democratic reform −1.371 0.954 −5.505 −4.100
(.640) (.744) (.047) (.145)
(.736) (.776) (.131) (.222)
years 0–3 after democratic reform 6.962 8.478 1.156 1.841
(.028) (.008) (.696) (.544)
(.191) (.067) (.775) (.637)
years 4 or more after democratic 16.360 19.455 13.629 13.675
reform
(.000) (.000) (.000) (.000)
(.019) (.004) (.032) (.034)
Time fixed effects Yes Yes Yes Yes
Country fixed effects Yes Yes Yes Yes
Continent-year dummies No Yes No Yes
Controls Yes Yes Yes Yes
Observations 2,565 2,565 2,314 2,314
Number of countries 74 74 69 69
R2 (within) 0.246 0.342 0.269 0.356
Source: Own calculations based on the data described in the text.
Note: p values in parentheses based on robust and clustered standard errors, respectively. Controls include
log per capita GDP, employment share, land per capita, log of population, and year and country fixed
effects included in every regression.
The estimated coefficient of the variable years 4 and after, which captures the
long-term effect of regime change, is positive and strongly significant for both the NRA
and RRA. The estimated values are similar to the values in columns 4 and 8 of table 2. The
results imply a long-run democratization effect of approximately 16–19 percentage points
for NRA and 13 percentage points for RRA. 24 For both the NRA and RRA, the short-term
effect, captured by the year 0–3 dummy, is always positive but is smaller in magnitude
23
than the long-term effect. For the RRA, in particular, the short-term effect is small and not
significant.
The results in table 4 are consistent with the descriptive evidence reported in
figures 1 and 2. After a democratization episode, there is an immediate increase in
agricultural protection. Then, policies appear to be stable for some years. After a few
years of democracy, we observe an additional increase in agricultural protection. Thus, it
appears that it takes time for democratization to fully exert its influence on agricultural
policy.
<>Matching
We now use a semiparametric analysis (i.e., a matching approach) to at least partly
relax the strong identifying assumptions in the D-in-D approach. 25 The results of the
matching procedure are presented in table 5. Given that the estimates are less efficient and
less precise owing to fewer usable observations, the matching results are consistent with
the results obtained from the standard D-in-D method. The effect of a transition to
democracy is strongly positive and statistically significant and is of the same order of
magnitude. As in the D-in-D regressions, the effect of democracy is larger on the NRA
than on the RRA.
TABLE 5. Robustness Check: Matching Estimates of the Democratization Effect
NRA RRA
(1) (2) (3) (4)
Growth in 14.63 13.95 9.72 9.25
agric.
Protection
Std. error (.062) .076) (.087) (.109)
lower bound
Std. error (.070) (.085) (.107) (.152)
upper bound
Estimation Matching Kernel Matching Matching Kernel Matching
technique Epanechnikov Kernel Epanechnikov Kernel
Gaussian Gaussian
24
No. of treated 10 10 5 5
countries
No. of control 10 10 7 7
countries
No. of controls 79 100 32 35
with
repetitions
Source: Own calculations based on the data described in the text.
Notes: p values in parentheses. In the upper row, they are estimated assuming independent observations,
whereas in the lower row, they are estimated assuming perfect correlations of repeated observations in
control countries.
<>Dynamic panel methods
Next, we estimate the effect of democratization on protection using dynamic panel
models. Specifically, we employ a dynamic D-in-D regression to control for the well-
known persistence of agricultural protection. However, because the lagged dependent
variables in a fixed effects specification are mechanically correlated with the error term
for N > T, we also use a first difference Generalized Method of Moments (GMM)
estimator (see Arellano and Bond 1991). The inclusion of a lagged protection variable on
the right-hand side may help to attenuate omitted variables bias because it captures
accumulated (unobserved) factors that affect actual protection.
To reduce bias due to the contemporaneous presence of both fixed effects and the
lagged dependent variable, we do not include countries for which fewer than 20 years of
data are available in the dynamic D-in-D regressions. In addition, to render the regressions
more comparable across dynamic estimators, the dynamic D-in-D specification does not
include the continental-year interaction terms used in the static D-in-D regressions. 26
The results of these additional regressions are reported in table 6. They are
consistent with our previous findings. The democratization dummies are consistently
estimated with strong precision (p < .01). As expected, the magnitudes of the estimated
effects are lower than with the static model because we are now capturing only the short-
25
term effect of democratization on agricultural protection. Moreover, the magnitudes of the
democratization effects in the GMM first-difference regressions are even higher than
those using the least-squares dynamic estimator.
TABLE 6. Robustness Check: Dynamic Panel Model of the Effect of Democratization on
Policy Reforms
Estimator D-in-D regression with T GMM difference
> 20 years
Regression (1) (2) (3) (4)
Dependent variable NRA RRA NRA RRA
Democratic reform 4.972 4.002 7.798 5.213
(.001) (.002) (.000) (.000)
Lagged NRA (RRA) 0.771 0.775 0.852 0.878
(.000) (.000) (.000) (.000)
Log per capita GDP 13.655 11.308 50.915 47.153
(.000) (.001) (.002) (.001)
Controls Yes Yes Yes Yes
Number of countries 61 55 74 68
Observations 2,364 2,151 2,439 2,194
R2 (within) 0.695 0.707
No. of GMM Instruments 52 52
Hansen test for over-id. (p value) .858 .920
AR2 test (p value) .279 .605
Source: Own calculations based on the data described in the text.
Note: p values based on clustered standard errors in parentheses. Controls include log per capita GDP,
employment share, land per capita, log of population, and year fixed effects included in every regression.
GMM first difference based on xtabond2 in Stata, with instrument lag structure (2 4) and collapse option to
control for instrument proliferation and using forward orthogonal deviations instead of first differencing (see
Arellano and Bover 1995).
<>Feedback effects
To further assess the problem of potential simultaneity bias, we regress the Polity2
democracy index on the level of protection in period t−1. Specifically, we estimate the
following democracy regression:
(3) d it = αd it −1 + φNRAit −1 + X it −1 β + µ t + δ i + ε it ,
26
where dit is the Polity2 democracy index of country i in period t. The lagged value of this
variable on the right-hand side is included to capture the persistence of democracy. The
parameter NRAit−1 is the lagged value of the protection level in agriculture. Other
covariates are included in the vector Xit−1. The parameters μt and δi denote full sets of year
and country fixed effects, respectively, and εit is an error term capturing all other omitted
factors.
For the same reason given above, the model in equation 3 is estimated using D-in-
D 27 and first-difference GMM estimators. Moreover, because democracy is a persistent
variable, we run a system GMM regression (see Arellano and Bover 1995). We find that
the lagged protection coefficient is always insignificant in these additional regressions
(see the results in table S.4 in the supplemental appendix). Thus, these results suggest that
there is no feedback effect of agricultural protection on the transition to democracy.
<>Definition of regime change
The evidence presented thus far has been obtained from approximately 40 political
reform episodes based on the Polity2 index. We have checked whether our results are
driven by the specific definition of our political reform variable. We have employed three
alternative approaches: (a) defining a democracy variable using all of the 67 reform
episodes from Polity2; (b) using the data from Papaioannou and Siourounis (2008) and
including only 23 (permanent) democratization episodes; and (c) only considering
permanent transitions from Polity2, namely, those that lasted at least eight years. These
democratization dummies differ in terms of not only the number of regime transitions
considered but also the timing of the reform episodes (see table S.1).
The results are presented in table S.5 in the supplemental appendix. The results
remain robust using these different measures of democratic transitions. Moreover, the
additional regression results suggest that permanent transitions are most important. In line
27
with the dynamic results discussed above, temporary democratization episodes (i.e., in
countries that revert to dictatorships after a brief democratization episode) have a
significantly lower effect on agricultural protection.
<>Economic and political crises
As noted in the recent political economy literature, the implemented policies may
be related to both economic and political (in)stability (see North et al. 2009; Besley and
Persson 2009 among others). Therefore, we complement our earlier specifications with
three variables designed to capture the effect of economic and political crises. An
economic crisis is measured with a dummy equal to one for every year that the real GDP
per capita growth rate from the Penn World Table is negative (zero otherwise). A political
crisis is measured with two dummies equal to one in every year a country is involved in a
domestic war or international conflict (zero otherwise). All three variables are used in the
regressions with several lags. 28 The effect of democratic reform on policy outcomes is
very robust to the inclusion of these additional covariates (see table S.6).
<>VII. CONCLUSIONS
In this paper, we investigate how democratization affects agricultural and food
policies. On the basis of the unique data set collected by the World Bank, we empirically
analyze the impact of political regime transitions on agricultural taxation and
subsidization.
We find a significant positive (negative) effect of a democratic transition on
agricultural protection (taxation). The transition to democracy increases agricultural
protection by 10 to 18 percentage points, depending on the indicator and the model
employed. This measured effect primarily reflects changes in poor countries, where the
vast majority of the transitions from autocracy to democracy occurred and where farmers
28
constitute a large share of the population. In the data set we used, the average share of
agriculture in total employment at the time of the transition from autocracy to democracy
was 65 percent (whereas the average share for all countries and time periods was 25
percent).
Our results are consistent with the predictions of the median voter model
suggesting that the impact of democratization is conditional on the structure of the
economy, which determines the share of votes of farmers among all voters. The median
voter model predicts that in poor countries where a large share of the population is
involved in farming, democratic reforms induce a profarmer policy effect. The factors that
make it difficult for farmers to organize politically in poor countries (such as their large
number and substantial geographic dispersion) render them potentially powerful in
electoral settings. Thus, our results suggest that democratization has benefited farmers in
poor countries.
We also find that the short-term effects are smaller than the long-term effects. The
effect of democratization on agricultural policies is strongest four to five years after a
change in political regime. This finding suggests that time is needed to arrive at a new
equilibrium in economic and political institutions.
An important question related to an empirical analysis such as ours is whether the
relationship that we document is causal. We cannot rule out the possibility of spurious
correlation due to various shocks that may have occurred over the past 50 years. We ran a
number of extensions of the model and robustness checks to account for this possibility to
the greatest extent possible. Our tests demonstrate that the results are robust to using
different levels of commodity aggregation, different indicators to capture the timing of
political reforms, alternative estimation methods, alternative definitions of regime
changes, and additional variables.
29
NOTES
1
This includes implicit trade taxes related to government intervention on the
domestic market for foreign currency and support for public agricultural research
(Anderson et al. 2009).
2
Note that the heterogeneous nature of agricultural protection in both developing
and developed countries may cause an aggregation bias in measures such as the NRA (see
Aksoy 2005). To attenuate this potential aggregation bias, in the empirical analysis
presented below, we also work with data at the commodity level.
3
Specifically, RRA = 100[(1 + NRAag/100)/(1 + NRAnonag/100) − 1], where NRAag
is the nominal assistance to agriculture, and NRAnonag is the nominal assistance to
nonagricultural sectors. Note that because of the computational complexity of this index,
the NRA to nonagricultural sectors is only based on distortion owing to tariff protection at
the border.
4
See section III for more detail on the definition and measures of
“democratization.� Depending on the time period covered and the year of
democratization, the average at each point in time in figures 1 and 2 is based on different
samples of countries.
5
Olper’s (2007) study of a cross-section of countries found that, on average, right-
wing governments are more protectionist with respect to agriculture than left-wing
governments. Furthermore, although left-wing governments support agriculture to a lesser
extent, they tend to support farmers more in unequal societies. This finding is consistent
with qualitative evidence from Bates (1983), who argues that socialist rulers in Africa tax
farmers (by imposing low commodity prices), and from Tracy (1989), who found that
30
right-wing governments in Europe (such as those dominated by Catholic and conservative
parties) tend to support farm interests and protectionism.
6
There are other problems in empirically assessing the impact of rulers’
preferences. First, applying a simple left-wing/right-wing model to agricultural policy is
not straightforward because increases to food costs through agricultural protection hurts
both urban workers (left-wing interests) and industrial capitalists (right-wing interests).
Thus, rulers who support either labor or capital should oppose agricultural protection, as
they did historically in Europe (Kindleberger 1975; Schonhardt-Bailey 1998; Findlay and
O’Rourke 2007). Second, economic development may change rulers’ preferences. As
their economies developed, Communist autocracies shifted from taxing to subsidizing
agriculture, as was the case in democracies. Communist dictators of poor countries, such
as Stalin in Russia, Mao in China, and Hoxha in Albania, heavily taxed agriculture.
However, farmers were subsidized at higher incomes, such as in the Soviet Union under
Brezhnev and in most Eastern European Communist countries in the 1970s and 1980s
(Swinnen and Rozelle 2009). Third, rulers’ preferences are not restricted to ideology; they
may also reflect regional interests. Bates and Block (2010) show that the regional
backgrounds of leaders in Africa significantly affected their policy preferences. Leaders
who drew political support from cities and semiarid regions (as in Tanzania and Ghana)
seized a major portion of revenues generated by the export of cash crops (coffee and
cocoa), whereas in countries where leaders came from regions where cash crops were
important sources of income (such as in Kenya and Ivory Coast), they imposed few taxes
on coffee and cocoa exports.
7
Olper (2007) finds more variation in policy choices, ceteris paribus, under
dictatorial regimes than under democracies. This result is consistent with the argument
31
that dictatorial leaders are less constrained in setting policies and that government
responses to pressure from interest groups are stronger in democracies.
8
Of the 42 democratic transitions (see table S.1 and the discussion below)
included in the data set, only five occurred in countries that are currently members of the
OECD (Spain, Portugal, Mexico, South Korea, and Turkey), and these transitions
occurred at times when they had considerably lower incomes than at present.
9
For a discussion of the relationships among various estimators in the context of
panel data, see Mundlak (1978) and Mundlak and Larson (1992).
10
This restriction is partially addressed by adding several covariates in the vector
Xi,t, to increase the similarity between treated and control countries.
11
See Ashenfelter (1978) and Ashenfelter and Card (1985) for a general
discussion of this subject.
12
Polity IV has a longer time series and therefore includes more usable political
reforms than other existing democracy indices. For example, in addition to its
shortcomings due to classification bias (see Papaioannou and Siourounis 2008), the use of
the Freedom House data strongly limits the number of usable transitions because the
information only begins in 1972. For a critical discussion of democracy indices, see
Munck and Verkuilen (2002).
13
We thank an anonymous referee for focusing our attention on this issue.
14
It is important to note that the use of the “continuous� Polity2 index, instead of a
discrete index, does not affect our qualitative conclusions; a higher Polity2 score increases
the level of agricultural protection. These additional results are not reported to conserve
space, but they may be obtained from the authors upon request.
32
15
The Sachs and Warner index, based on the recent update by Wacziarg and
Welch (2008), is equal to one when a country is considered open and zero otherwise on
the basis on the following criteria: an aggregate tariff rate greater than 40 percent, a
nontariff barrier covering more than 40 percent of trade, a black market exchange rate of
less than 20 percent relative to the official exchange rate, and a state monopoly in major
exports.
16
An alternative means of correcting for the potential problem of inconsistent
standard errors would be to follow a residual-aggregation procedure, as suggested by
Bertrand et al. (2004). In our case, where we consider approximately 40 reform episodes,
this could be problematic because the power of this procedure is quite low and diminishes
rapidly with sample size.
17
Hausman tests also confirm this correlation. Please note that in the random
effects model the key result (i.e., the effect of democratization on the NRA and RRA) is
positive and strongly significant and is virtually identical in magnitude to the results
reported in table 2 (additional results are available upon request).
18
As emphasized by Papaioannou and Siourounis (2008), the democracy indicator
behaves as a trend because countries that switch to democracy seldom revert to autocracy.
19
We use total government consumption instead of government spending owing to
data limitations. For our broad country sample and the 1955–2005 time period, this is the
most widely available measure of government spending.
20
See de Gorter and Swinnen (2002) for a survey. A positive impact of GDP per
capita is in line with the so-called development paradox. A negative impact of agricultural
employment is in line with Olson’s (small) interest group story and the reduced per capita
33
tax costs of subsidizing a declining sector. A negative impact of land per capita is in
accordance with the notion that countries with a comparative advantage in agriculture are
less protected (Anderson 1995; Swinnen 1994). Moreover, this variable may capture
collective action problems due to the heterogeneity of the farm group. This latter
interpretation draws on the observation that countries with more abundant land tend to
consistently have a more unequal distribution of land (Olper 2007).
21
This is consistent with the fact that running a regression using the nominal rate
of assistance to nonagricultural products, NRAnonag (i.e., the denominator of the RRA), as
the dependent variable means that the democracy reform dummy is never significant,
irrespective of specification. These additional results are available from the authors upon
request.
22
We thank a referee for this suggestion.
23
The compositions of the groups are as follows: grains and tubers (e.g., rice,
wheat, maize, cassava, barley, sorghum, millet, and oats), oilseeds (e.g., soybean,
groundnut, palm oil, rapeseed, sunflower, and sesame), livestock products (e.g., pigment,
milk, beef, poultry, egg, sheep meat, and wool), and tropical crops (e.g., sugar, cotton,
coconut, coffee, rubber, tea, and cocoa).
24
Not surprisingly, the magnitudes of these reform effects are similar to those in
the regressions that consider only permanent reforms. See the regressions in columns 3
and 6 in table S.5 in the supplemental appendix.
25
Table S.2 in the supplemental appendix presents the coefficients of the Probit
models that were used to calculate propensity scores. Although our model is not ideal for
the prediction of shifts toward democracy, the selected covariates provide some
34
explanation for a regime change (pseudo R2 equal to 0.23−0.24). Table S.3 in the
supplemental appendix compares the distribution of observed covariates between the
countries in the treatment and control groups. The matching performed well in terms of
removing significant differences between the treatment and control countries, although the
treatment and control groups were not particularly different prior to matching. Matching
reduces the difference in means for several variables, such as the dummy for Africa,
relative GDP, and conflict years.
26
Adding these continent-year interaction terms in the GMM equations induces a
strong increase in the number of instruments, rendering it difficult, if not impossible, to
have fewer instruments than groups and thus to respect the “rule of the thumb� when
running GMM models (see Roodman 2009). However, note that cross-country differences
in protection dynamics are now largely subsumed in the autoregressive coefficient.
27
We use a sample that excludes countries for which fewer than 20 years of data
are available to reduce bias resulting from the contemporaneous presence of both fixed
effects and the lagged dependent variable.
28
As correctly noted by a referee, these variables can serve as imperfect proxies, at
best, for shocks in policy or world markets or variations in world prices. Nevertheless,
they seem to be the best proxies available. Note that by using a dynamic panel model, as
in table 6, we implicitly account for potential spurious correlations between
democratization and protection due to (unobserved) policy shocks.
35
REFERENCES
Abadie, Alberto. 2005. “Semiparametric Difference-in-Difference Estimators.�
Review of Economic Studies 72: 1–19.
Aksoy, M.A. 2005. “Global Agricultural Trade Policies.� In Global Agricultural
Trade and Developing Countries, eds. M.A. Aksoy and J. Beghin, 37–53.
Washington, DC: The World Bank.
Alesina, Alberto, and Dani Rodrik. 1994. “Distributive Politics and Economic
Growth.� The Quarterly Journal of Economics 109(2): 465–490.
Anderson, Kym, and Ernesto Valenzuela. 2008. Estimates of Distortions to
Agricultural Incentives, 1955 to 2007. Washington, DC: World Bank.
www.worldbank.org/agdistortions.
Anderson, Kym, and Will Martin. 2006. Agricultural Trade Reform and the Doha
Development Agenda. London and Washington, DC: Palgrave Macmillan and
the World Bank.
Anderson, Kym, and Yujiro Hayami. 1986. The Political Economy of Agricultural
Protection: East Asia in International Perspective. London: Allen and Unwin.
Anderson, Kym, Johanna Corser, Damiano Sandri, and Ernesto Valenzuela. 2009.
“Agricultural distortions patterns since the 1950s: what needs explaining?�
Agricultural Distortions Working Paper 90.
Anderson, Kym. 1995. “Lobbying incentives and the pattern of protection in rich and
poor countries.� Economic Development and Cultural Change 43(2): 401–23.
36
Anderson, Kym. 2009. Distortions to Agricultural Incentives: A Global
Perspective,1955–2007. London and Washington, DC: Palgrave Macmillan
and the World Bank.
Arellano, Manuel and Olympia Bover. 1995. “Another look at instrumental variables
estimation of error-component models� Journal of Econometrics 68: 29–51.
Arellano, Manuel and Stephen Bond. 1991. “Some Tests of Specification for Panel
Data: Monte Carlo Evidence and an Application to Employment Equations.�
Review of Economic Studies, 58(2): 277–297.
Ashenfelter, Orley, and David Card. 1985. “Using the Longitudinal Structure of
Earnings to Estimate the Effects of Training Programs.� Review of Economics
and Statistics 67: 648–660.
Ashenfelter, Orley. 1978. “Estimating the Effect of Training Programs on Earnings.�
Review of Economics and Statistics 60: 47–57.
Banerji, Arup, and Hafez Ghanem. 1997. “Does the type of political regime matter for
trade and labor market policies?� World Bank Economic Review 11: 171–194.
Bates, Robert, H., and Steven Block. 2010. “Political Institutions and Agricultural
Trade Interventions in Africa�, American Journal of Agricultural Economics
93(2): 317–323.
Bates, Robert. H. 1983. “Patterns of Market Intervention in Agrarian Africa.� Food
Policy 8(4): 297–304.
Beghin, John C., and Mylene Kherallah. 1994. “Political Institutions and International
Patterns of Agricultural Protection.� Review of Economics and Statistics,
LXXVI: 482–489.
37
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. “How much
should we trust difference-in-difference estimates?� The Quarterly Journal of
Economics 119: 249–275.
Besley, Timothy and Torsten Persson. 2009. “The Origins of State Capacity:
Property Rights, Taxation and Politics.� American Economic Review 99:
1218–1244.
Blundell, Richard, Monica Costa Dias, Costas Meghir, and John Van Reenen. 2004.
“Evaluating the Employment Impact of a Mandatory Job Search Assistance
Program.� Journal of the European Economic Association 2: 596–606.
de Gorter, Harry and Johan F.M. Swinnen. 2002. “Political Economy of Agricultural
Policies.� In The Handbook of Agricultural Economics, Volume 2, eds. Bruce
Gardner and Gordon Rausser, 2073–2123. Amsterdam: Elsevier.
Eichengreen, Barry, and David Leblang. 2008. “Democracy and Globalization.�
Economics & Politics 20: 289–334.
Fan, Jianqing. 1992. “Local Linear Regression Smoothers and their Minimax
Efficiencies.� The Annals of Statistics 21: 196–216.
Findlay, Ronald, and Kevin O’Rourke. 2007. Power and Plenty: Trade, War and the
World Economy in the Second Millennium. Princeton, NJ: Princeton
University Press.
Giavazzi, Francesco, and Guido Tabellini. 2005. “Economic and political
liberalization.� Journal of Monetary Economics 52: 1297–1330.
Giuliano, Paola, Prachi Mishra and Antonio Spilimbergo. 2011. “Democracy and
Reforms: Evidence from a New Dataset.� CARF F-Series CARF-F-247,
University of Tokyo.
38
Gleditsch, Nils, P., Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg, and
Havard Strand. 2002. “Armed Conflict 1946–2001: A New Dataset.� Journal
of Peace Research 39 (5): 615–637.
Heckman, James, Hidehiko Ichimura, and Petra E. Todd. 1998. “Matching as an
Econometric Evaluation Estimator.� Review of Economic Studies 65 (2): 261–
294.
Kindelberger, Charles P. 1975. “The Rise of Free Trade in Western Europe, 1820–
1875.� The Journal of Economic History 35 (1): 20–55.
Kono, Daniel Y. 2006. “Optimal Obfuscation: Democracy and Trade Policy
Transparency.� American Political Science Review 100 (3): 369–384.
Krueger, Anne O., Maurice Schiff and Alberto Valdes. 1988. “Agricultural Incentives
in Developing Countries: Measuring the Effect of Sectoral and Economy-wide
Policies.� World Bank Economic Review 2 (3): 255–72.
Lindert, Peter H. 1991. “Historical Patterns of Agricultural Policy.� Agriculture and
the State. Growth, Employment, and Poverty in Developing Countries, ed.
Timmer C. Peter, 29–83. Ithaca, NY: Cornell University Press.
Marshall, Monty G., and Keith Jaggers. 2007. Polity IV Project: Dataset Users’
Manual. Arlington: Polity IV Project.
Milner, Helen V., and Bumba Mukherjee. 2009. “Democratization and Economic
Globalization.� Annual Review of Political Science 12: 161–181.
Milner, Helen V., and Keiko Kubota. 2005. “Why the Move to Free Trade?
Democracy and Trade Policy in the Developing Countries.� International
Organization 59: 107–143.
39
Mulligan, Casey B., Richard Gil, and Xavier Sala-i-Martin. 2004. “Do Democracies
have Different Public Policies than Non Democracies.� Journal of Economic
Perspective 18 (1): 51–74.
Munck, Gerardo L., and Jay Verkuilen. 2002. “Conceptualizing and Measuring
Democracy: Evaluating Alternative Indices.� Comparative Political Studies
35: 5–34.
Mundlak, Yair, and Larson, Donald F. 1992. “On the Transmission of World
Agricultural Prices.� World Bank Economic Review 6 (3): 399–422.
Mundlak, Yair. 1978. “On the Pooling of Time-Series and Cross-Section Data.�
Econometrica 46: 69–86.
North, Douglas, John J. Wallis and Barry R. Weingast. 2009. Violence and Social
Orders: A Conceptual Framework for Interpreting Recorded Human History.
New York: Cambridge University Press.
O’Rourke, Kevin H., and Alan, M. Taylor. 2007. “Democracy and Protection.� In The
New Comparative Economic History: Essays in Honor of Jeffrey G.
Williamson, eds. Timothy Hatton, Kevin H. O’Rourke, and Alan M. Taylor,
193–216. Cambridge, MA: MIT Press.
Olper, Alessandro, and Valentina Raimondi. Forthcoming. “Electoral Rules, Forms of
Government and Redistributive Policy: Evidence from Agriculture and Food
Policies.� Journal of Comparative Economics.
Olper, Alessandro. 2001. “Determinants of Agricultural Protection: The Role of
Democracy and Institutional Setting.� Journal of Agricultural Economics
52(2): 75–92.
40
Olper, Alessandro. 2007. “Land Inequality, Government Ideology and Agricultural
Protection.� Food Policy 32: 67–83.
Olson, Mancur. 1965. The Logic of Collective Action. Cambridge MA: Harvard
University Press.
Papaioannou, Elias, and Gregorios Siourounis. 2008. “Democratization and Growth.�
The Economic Journal 118: 1520–1551.
Persson, Torsten, and Guido Tabellini. 1994. “Is Inequality Harmful for Growth?,�
American Economic Review 84 (3): 600–621.
Persson, Torsten, and Guido Tabellini. 2008. “The growth effect of democracy: Is it
heterogeneous and how can it be estimated?� In Institutions and Economic
Performance, ed. E. Helpman, 544–584. Harvard University Press.
Persson, Torsten. 2005. “Forms of Democracy, Policy and Economic Development.�
Working Paper, Institute for International Economic Studies, Stockholm
University.
Rodrik, Dani, and Romain Wacziarg. 2005. “Do democratic transitions produce bad
economic outcomes?�American Economic Review 95 (2): 50–55.
Roodman David. 2009. “A Note on the Theme of Too Many Instruments.� Oxford
Bulletin of Economics and Statistics 71 (1): 135–158.
Sachs, Jeffrey, and Andrew Warner. 1995. “Economic Reform and the Process of
Global Integration.� Brookings Papers on Economic Activity 1: 1–95.
Schonhardt-Bailey, Cheryl. 1998. “Interests, Ideology and Politics: Agricultural Trade
Policy in Nineteenth Century Britain and Germany.� In Free Trade and Its
Reception, 1815–1960, ed. Andrew Marrison, 63–81. London: Routledge.
41
Sianesi, Barbara. 2001. “Implementing Propensity Score Matching Estimators with
STATA.� Paper presented at the UK Stata Users Group, VII Meeting, London,
May 2001.
Smith, Jeffrey. A., and Petra E.Todd. 2005. “Does matching overcome LaLonde's
critique of nonexperimental estimators?� Journal of Econometrics 125: 305–
353.
Swinnen, Johan F. M. 1994. “A positive Theory of Agricultural Protection.,�
American Journal of Agricultural Economics 76 (1): 1–14.
Swinnen, Johan F. M. 2010. “The Political Economy of Agricultural and Food
Policies: Recent Contributions, New Insights, and Areas for Further
Research.� Applied Economic Perspectives and Policy 32 (1): 33–58.
Swinnen, Johan F. M., Anurag N. Banerjee, and Harry de Gorter. 2001. “Economic
Development, Institutional Change, and the Political Economy of Agricultural
Protection: An Econometric Study of Belgium Since the 19th Century.,�
Agricultural Economics 26 (1): 25–43.
Swinnen, Johan F. M., Harry de Gorter, Gordon, C, Rausser, and Anurag, N.
Banerjee. 2000. “The Political Economy of Public Research Investment and
Commodity Policies in Agriculture: An Empirical Study.� Agricultural
Economics 22: 111–122.
Swinnen, Johan F.M., and Scott Rozelle. 2006. From Marx and Mao to the Market:
The Economics and Politics of Agricultural Transition. Oxford University
Press.
42
Swinnen, Johan, and Scott, Rozelle. 2009. “Governance Structures and Resource
Policy Reform: Insights from Agricultural Transition.,� Annual Review of
Resource Economics 1 (1): 33–54.
Tavares, Samia C. 2007. “Democracy and Trade Liberalization.� Working Paper,
Rochester Institute of Technology, February.
Tracy, Michael. 1989. Government and Agriculture in Western Europe 1880–1988,
3rd ed. New York: Harvester Wheatsheaf.
Varshney, Ashutosh. 1995. Democracy, Development and the Countryside.
Cambridge: Cambridge University Press.
Wacziarg, Romain, and Karen H. Welch. 2008. “Trade Liberalization and Growth:
New Evidence.� The World Bank Economic Review 22 (2): 187–231.
43
POLITICAL REFORMS AND PUBLIC POLICIES:
EVIDENCE FROM AGRICULTURAL AND FOOD POLICIES
SUPPLEMENTAL APPENDIX
By ALESSANDRO OLPER, JAN FAÅ?KOWSKI, AND JO SWINNEN*
Table A.I. Country sample and democratic (autocratic) reforms
Years coverage Democratic reforms: Polity 2 Democratic reforms: P&S
Country Start End Into Out
1 Argentina 1960 2005 1973; 1983 1976 Full democratization 1983
2 Australia 1955 2005 Always democracy Always democracy
3 Austria 1956 2005 Always democracy Always democracy
4 Bangladesh 1974 2004 1991 Partial democratization 1991
5 Benin 1970 2005 1991 Full democratization 1991
6 Brazil 1966 2005 1985 Partial democratization 1985
7 Bulgaria 1992 2005 Always democracy Always democracy
8 Burkina Faso 1970 2005 1977 1980 Always autocracy
9 Cameroon 1961 2005 Always autocracy Always autocracy
10 Canada 1961 2005 Always democracy Always democracy
11 Chad 1970 2005 Always autocracy Always autocracy
Full dem. 1990 (dem. 1960-
12 Chile 1960 2005 1989 1973 1972)
13 China 1981 2005 Always autocracy Always autocracy
14 Colombia 1960 2005 Always democracy Always democracy
15 Cote d'Ivoire 1961 2005 2000 2002 Always autocracy
16 Czech Republic 1992 2005 Always democracy Always democracy
17 Denmark 1956 2005 Always democracy Always democracy
Dominican Full democratization 1978
18 Republic 1955 2005 1978
19 Ecuador 1970 2003 1968; 1979 1970 Full democratization 1979
20 Egypt 1955 2005 Always autocracy Always autocracy
21 Estonia 1992 2005 Always democracy Always democracy
22 Ethiopia 1981 2005 1994 Partial democratization 1995
23 Finland 1956 2005 Always democracy Always democracy
24 France 1956 2005 Always democracy Always democracy
25 Germany 1955 2005 Always democracy Always democracy
1970; 1979; Full democratization 1996
26 Ghana 1960 2004 1996 1972; 1981
27 Hungary 1992 2005 Always democracy Always democracy
28 India 1960 2005 Always democracy Always democracy
29 Indonesia 1970 2005 1999 Partial democratization 1999
30 Ireland 1956 2005 Always democracy Always democracy
31 Italy 1956 2005 Always democracy Always democracy
32 Japan 1955 2005 Always democracy Always democracy
*
Alessandro Olper: University of Milano and LICOS, alessandro.olper@unimi.it; Jan Fałkowski: University of
Warsaw and CEAPS, jfalkowski@wne.uw.edu.pl; Jo Swinnen: Catholic University of Leuven and LICOS,
Jo.Swinnen@econ.kuleuven.be. Corresponding author: Alessandro Olper, Dipartimento di Economia e Politica
Agraria, Agro-alimentare e Ambientale, Facoltà di Agraria - Via Celoria 2 - 20133 MILANO. Tel. +39–02–
50316481.
33 Kenya 1966 2001 2002 1966 Always autocracy
34 Korea South 1955 2005 1963; 1987 1972 Full democratization 1988
35 Latvia 1992 2005 Always democracy Always democracy
36 Lithuania 1992 2005 Always democracy Always democracy
37 Madagascar 1960 2005 1991 Partial democratization 1993
38 Malaysia 1960 2005 Always democracy Always intermediate
39 Mali 1970 2005 1992 Full democratization 1992
40 Mexico 1979 2005 1994 Full democratization 1997
41 Morocco 1961 2004 Always autocracy Always autocracy
42 Mozambique 1975 2005 1994 Partial democratization 1994
43 Netherlands 1956 2005 Always democracy Always democracy
44 New Zealand 1955 2005 Always democracy Always democracy
45 Nicaragua 1991 2004 1990 Partial democratization 1990
46 Nigeria 1961 2004 1979; 1999 1966; 1984 Partial democratization 1999
47 Norway 1956 2005 Always democracy Always democracy
1969; 1977; Borderline democratization
48 Pakistan 1962 2005 1972; 1988 1999 1988
49 Philippines 1962 2005 1987 1972 Full democratization 1987
50 Poland 1992 2005 Always democracy Always democracy
51 Portugal 1956 2005 1975 Full democratization 1976
52 Romania 1992 2005 Always democracy Always democracy
53 South Africa 1961 2005 Always democracy Full democratization 1994
54 Russia 1992 2005 Always democracy Partial democratization 1993
55 Senegal 1961 2005 2000 Full democratization 2000
56 Slovakia 1992 2005 Always democracy Always democracy
57 Slovenia 1992 2005 Always democracy Always democracy
58 Spain 1955 2005 1976 Full democratization 1978
59 Srilanka 1955 2004 Always democracy Always democracy
1958; 1970; Always autocracy
60 Sudan 1958 2004 1965; 1986 1989
61 Sweden 1956 2005 Always democracy Always democracy
62 Switzerland 1956 2005 Always democracy Always democracy
63 Taiwan 1955 2002 1992 N.A.
64 Tanzania 1976 2004 2000 Partial democratization 1995
65 Thailand 1978 2004 1974; 1978 1976 Full democratization 1992
66 Togo 1970 2005 Always autocracy Always autocracy
67 Turkey 1961 2005 1973; 1983 1971; 1980 Partial democratization 1983
68 Uganda 1961 2004 1980 1966; 1985 Always autocracy
69 UK 1956 2005 Always democracy Always democracy
70 Ukraine 1992 2005 Always democracy Partial democratization 1994
71 USA 1955 2005 Always democracy Always democracy
72 Vietnam 1986 2005 Always autocracy Always autocracy
73 Zambia 1964 2005 1991 1968 Partial democratization 1991
74 Zimbabwe 1970 2005 1987 Reverse transition 1987
Notes: The table reports country and years coverage (columns 1-3); the classification of democratic (autocratic)
reform episodes and political regimes using Polity2 (columns 4-5) with Bold numbers in columns 4 and 5
referring to reform episodes that satisfy the criteria requested and thus are those used in the estimation of the
democracy effect (see text). Columns 6 reports democratization episodes following Papaioannou and Siourounis
(P&S, 2008). (See text).
Table S.2 Estimates of the propensity score
Transitions to democracy
RRA NRA
Initial polity2 0.07 0.09
(0.77) (1.03)
Relative GDP 25.17 35.77
(1.84)* (1.73)*
Sample length 0.01 -0.06
(0.13) (1.75)*
Conflict years 3.30 0.85
(1.77)* (0.84)
Constant -0.38 2.17
(0.36) (1.72)*
Observations 33 38
Pseudo R2 0.23 0.24
Notes: t-values in parentheses: ***p < .01; **p < .05; *p < .10.
Table S.3 Transitions to democracies: balancing properties
RRA NRA
Variable Mean t-test Mean t-test
Sample Treated Control t-value p > |t| Treated Control t-value p > |t|
Relative GDP Unmatched 0.065 0.026 1.19 0.242 0.062 0.022 1.47 0.150
Matched 0.031 0.035 -0.58 0.577 0.021 0.023 -0.29 0.776
Initial polity2 Unmatched -5.538 -5.714 0.13 0.899 -5.464 -6.000 0.46 0.645
Matched -4.000 -5.694 0.90 0.395 -5.700 -6.342 0.51 0.619
Sample length Unmatched 33.23 32.00 0.24 0.813 33.79 36.90 -0.80 0.426
Matched 35.80 42.29 -0.84 0.423 37.00 37.18 -0.04 0.967
Conflict years Unmatched 0.246 0.127 1.01 0.322 0.229 0.202 0.26 0.799
Matched 0.089 0.091 -0.02 0.985 0.176 0.189 -0.12 0.905
Latin America Unmatched 0.230 0.000 1.40 0.170 0.214 0.000 1.61 0.117
Matched 0.400 0.000 1.63 0.141 0.100 0.000 1.00 0.331
Asia Unmatched 0.269 0.285 -0.08 0.933 0.250 0.200 0.31 0.757
Matched 0.000 0.000 . . 0.100 0.206 -0.63 0.535
Africa Unmatched 0.423 0.714 -1.37 0.182 0.464 0.800 -1.87 0.070
Matched 0.600 1.000 -1.63 0.141 0.800 0.794 0.03 0.974
Notes: Matching is based on the estimates reported in table S.2.
Table S.4 Robustness check: Feedback effect from policy reforms to democracy
Estimator D-in-D with T > 20 years GMM difference GMM system
Regression (1) (2) (3) (4) (5) (6) (5) (6)
Dependent variable Polity2 Polity2 Polity2 Polity2 Polity2 Polity2 Polity2 Polity2
Lagged Polity2 0.884 0.879 0.885 0.879 0.858 0.886 0.875 0.879
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Lagged NRA 0.002 0.001 -0.003 0.001
(0.035) (0.247) (0.323) (0.467)
Lagged RRA 0.001 0.001 0.002 0.001
(0.416) (0.550) (0.690) (0.505)
Lagged per capita GDP -0.177 -0.039 -0.110 -0.003 -1.315 -1.700 0.062 0.055
(0.473) (0.900) (0.671) (0.991) (0.156) (0.114) (0.476) (0.511)
Controls No Yes No Yes Yes Yes Yes Yes
Countries 61 61 55 55 74 69 74 69
Observations 2492 2416 2281 2219 2530 2297 2606 2371
R square (Within) 0.830 0.828 0.821 0.821
GMM Instruments # 51 51 53 53
Hansen test for over-identification (p-value ) 0.366 0.411 0.348 0.646
AR2 test (p-value ) 0.069 0.143 0.054 0.130
Note: p-value based on clustered standard errors in parentheses. Controls: one year lag of log per capita GDP,
employment share, land per capita, log of population included as indicated; year fixed effects included in every
regression. GMM system based on xtabond2 in Stata, with instrument lag structure (3 4), and collapse option to
control for instruments proliferation.
Table S.5 Robustness check: Alternative definitions of democratic reform
Estimation Difference-in-Difference Regressions
Regression (1) (2) (3) (4) (5) (6)
Dependent variable NRA NRA NRA RRA RRA RRA
Reform definition
All reforms based on Polity2 11.918 9.025
(0.005) (0.024)
Papaioannu&Siourounis 15.209 10.165
(0.003) (0.050)
Only permanent reforms 17.293 11.892
(0.004) (0.039)
Other covariates Yes Yes Yes Yes Yes Yes
Continental trends Yes Yes Yes Yes Yes Yes
Countries 74 74 74 69 69 69
Observations 2606 2606 2606 2352 2352 2352
2
R (within) 0.339 0.341 0.339 0.355 0.358 0.353
Note: p-value based on clustered standard errors in parentheses. Controls: log per capita GDP, employment
share, land per capita, log of population and lagged economic and political crisis as in Table 4, plus country and
years fixed effects, included in every regression. The first reform definition considers all 66 democratic
(autocratic) reforms based on the Polity2 index. The second one is based on Papaioannou and Siourounis (2008)
database. Finally, the third definition is again based on the Polity IV database but considers only permanent
transitions. (See text for further details).
Table S.6 Robustness check: Economic and political crisis
Estimation Difference in Difference estimates
Regression (1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable NRA NRA NRA NRA RRA RRA RRA RRA
Democratic reform 14.255 14.089 14.078 14.339 9.633 9.628 9.745 9.865
(0.003) (0.003) (0.003) (0.003) (0.032) (0.034) (0.033) (0.030)
Lagged crisis_1 (grgdppc<0) 2.545 2.412 1.594 1.439
(0.108) (0.133) (0.340) (0.394)
Lagged crisis_2 2.690 2.561 2.392 2.217
(0.028) (0.036) (0.078) (0.101)
Lagged crisis_3 2.156 2.114 1.683 1.557
(0.090) (0.090) (0.217) (0.240)
Lagged war_1 4.693 3.144 3.447 1.832
(0.168) (0.329) (0.273) (0.531)
Lagged war_2 0.371 -0.961 0.955 -0.573
(0.859) (0.644) (0.641) (0.779)
Lagged war_3 -0.318 -1.041 1.054 -0.549
(0.910) (0.702) (0.729) (0.850)
Lagged conflict_1 1.822 1.238 1.876 1.443
(0.336) (0.462) (0.315) (0.397)
Lagged conflict_2 1.891 1.652 1.823 1.633
(0.221) (0.274) (0.271) (0.333)
Lagged conflict_3 1.475 1.725 2.911 2.977
(0.478) (0.379) (0.200) (0.163)
Other covariates Yes Yes Yes Yes Yes Yes Yes Yes
Continental trends Yes Yes Yes Yes Yes Yes Yes Yes
Countries 74 74 74 74 69 69 69 69
Observations 2565 2565 2565 2565 2314 2314 2314 2314
R square 0.341 0.339 0.340 0.343 0.353 0.352 0.354 0.356
Note: p-value based on clustered standard errors at country level in parentheses. Controls: log per capita GDP,
employment share, land per capita and log of population plus country and years fixed effects, included in every
regression. Democracy variable is based on the Polity2 index.