The channels of the impact of democracy

The effect of democracy on institutional outcomes is theoretically ambiguous.Footnote ¹ For instance, North (Reference North1990) and North and Weingast (Reference North and Weingast1989) suggest that democracy should result in safer property rights, because it constrains the action of policy makers. In contrast, David Ricardo and Karl Marx viewed democracy and universal suffrage as a threat on property rights, because of the incentive for poorer voters to expropriate the rich. Berggren and Bjørnskov (Reference Berggren and Bjørnskov2013) report findings suggesting that democracy allows religiosity to be mediated by the political process and deteriorate property rights in democracies. Przeworski and Limongi (Reference Przeworski and Limongi1993), Alesina and Rodrik (Reference Alesina and Rodrik1994), and Persson and Tabellini (Reference Persson and Tabellini1994) provide modern versions of the argument. Empirical evidence, in general, points to a positive association between democracy and the safety of property rights (Clague et al., Reference Clague, Keefer, Knack and Olson1996).

Moreover, democracy is expected to limit institutional dysfunctions such as corruption. This is the case, for instance, when voters can sanction incumbents convicted of corruption (Ferraz and Finan, Reference Ferraz and Finan2008), which, in turn, acts as a disciplining device in a principal-agent framework (Aidt et al., Reference Aidt, Dutta and Sena2008; Ferejohn, Reference Ferejohn1986). Here again, the impact of democracy on corruption is not a priori univocal. Democracy may create incentives to engage in corruption to fund electoral campaigns. It may also affect the incentive to try and influence decision makers (Campos and Giovannoni, Reference Campos and Giovannoni2017). Partisan loyalty may incite voting for corrupt incumbents (Rose-Ackerman, Reference Rose-Ackerman2013).

Finally, democracy may deliver more political stability and reduce the likelihood of internal and external conflicts. Democratic countries are better able to find compromises on how to share income shocks (Rodrik, Reference Rodrik1999). Conversely, such shocks create unrest in non-democratic countries, because the dominant group will try to impose the burden of adjustment on the minority (Dutt and Mobarak, Reference Dutt and Mobarak2016; Henisz, Reference Henisz2004). Democracy may also reduce the likelihood of inter-country conflicts, because a majority of citizens would vote against going to war (Kant, Reference Kant1795; Maoz and Russett, Reference Maoz and Russett1993). It will reduce the likelihood of internal conflicts by being less repressive and more inclusive (Gleditsch et al., Reference Gleditsch, Hegre, Strand, Midlarsky and Arbor2009; Green, Reference Green2018). However, it also constrains the possibilities of government repression, which is favorable to rebellion (Collier and Rohner, Reference Collier and Rohner2008).

The timing of the effect

A couple of contributions focus on the timing of the effect of transitions. Mohtadi and Roe (Reference Mohtadi and Roe2003) provide the most specific theoretical discussion of the change in corruption after a democratic transition and suggest that corruption will initially grow before decreasing as democracy matures. Giavazzi and Tabellini (Reference Giavazzi and Tabellini2005) do not support this result, as they report evidence that corruption decreases after political liberalizations. The impact on property rights is less clear.Footnote ² Given that their period of study is only 15 years long, they cannot track the change in corruption and property rights beyond 4 years after democratizations.

Acemoglu and Robinson (Reference Acemoglu and Robinson2001) focus on redistribution policy and fiscal stability. At the time of democratic transitions, an unequal income distribution would prompt redistribution. Governments will adjust redistribution over the business cycle to avoid unrest. Fiscal policy is therefore volatile. As democracy matures, income inequality should be reduced and fiscal policy become less volatile.

The role of the age of democracy seems confirmed by the empirical studies of Clague et al. (Reference Clague, Keefer, Knack and Olson1996), Treisman (Reference Treisman2000), and Henderson and Kuncoro (Reference Henderson and Kuncoro2011). Moreover, Weingast (Reference Weingast1997) argues that democratic stability becomes self-enforcing if citizens have overcome the coordination problem that would otherwise prevent them from sanctioning officials that transgress formal institutions. Persson and Tabellini (Reference Persson and Tabellini2009) build a model where the citizens of a country receive a warm glow from fighting for democracy that increases while the country is a democracy and depreciates when it is not. Their empirical analysis confirms that older democracies tend to be more stable and grow faster. It, however, cannot be used to determine how long it takes for democracies to become stable after a transition.

If one assumes that democratizations can affect growth because they affect institutional outcomes, then the literature on growth indirectly provides an upper bound on the time that democratizations take to affect institutional quality. Hausmann Ricardo and Rodrik (Reference Hausmann Ricardo and Rodrik2005) find that growth accelerations are associated with a change of political regime in a window of 5 years. Rodrik and Wacziarg (Reference Rodrik and Wacziarg2005) observe that new democracies, defined as countries that have been democratic for less than 5 years, experience faster growth. Papaioannou and Siourounis (Reference Papaioannou and Siourounis2008b) provide the most detailed analysis of the effect over time of democratization on growth. They observe that the growth rate of democratizing countries is already significantly larger in the period ranging from the first to the third year after the transition. The effect remains significant during the following 3 years and beyond the 7th year after the transition.

However, Rodrik (Reference Rodrik1999) and Acemoglu et al. (Reference Acemoglu, Naidu, Restrepo and Robinson2019) report evidence suggesting that longer horizons should not be ruled out. Rodrik (Reference Rodrik1999) observes that the quality of institutions is a good predictor of differences in average growth rates between the periods 1960–1975 and 1975–1989. Acemoglu et al. (Reference Acemoglu, Naidu, Restrepo and Robinson2019) compare per capita GDP respectively 5 years and 25–30 years after democratization with its value during the year of the democratization and find that it is statistically significantly larger after 25 years. Most of their estimates also suggest that GDP per capita is larger 5 years after the transition, although the magnitude and statistical significance of the difference are lower. Madsen et al. (Reference Madsen, Raschky and Skali2015) regress per capita income on the lagged level of the Polity2 democracy index over the period 1820–2000. They observe that the coefficient of democracy is significantly positive. Because they consider 10-year periods, one may contend that 10 years is an upper limit on the time that democracy takes to affect income, hence institutional outcomes.

Overall, the available evidence suggests that although a couple of years might be necessary for democratic transitions to affect economic outcomes, a period of 20 years is not unrealistic a priori.

Before moving on, a caveat must be made. The effect of democracy may be limited for two reasons. First, outcomes may move slowly, either because they are deeply rooted cultural norms (Roland, Reference Roland2004), or because the incentive for individuals to behave honestly is reduced by systemic obstacles (Tirole, Reference Tirole1996). Second, de facto democratization may be insubstantial if not accompanied by changes in the de facto distribution of power. The elite may compensate for lost political power after democratization by greater investment in de facto power, preventing changes in policies and economic institutions (Acemoglu and Robinson, Reference Acemoglu and Robinson2008). Those contentions echo Hayek's (Reference Hayek2012: chapter 1) argument that institutions are the result of an evolutionary process rather than of an intentional design and they give superior strength and are therefore stable. Bjørnskov (Reference Bjørnskov2020) illustrates that possibility by showing that military dictatorships introduce new constitutions before democratization, in particular to introduce presidential democracies that will allow the military to keep more power.

Empirical method

We apply to institutional outcomes the method initiated by Rodrik and Wacziarg (Reference Rodrik and Wacziarg2005) and used by Papaioannou and Siourounis (Reference Papaioannou and Siourounis2008b) and Acemoglu et al. (Reference Acemoglu, Naidu, Restrepo and Robinson2019) to study growth, which uses a panel of countries and defines episodes of democratization. The following regression equation summarizes it:

(1)$$Inst_{i, \;t}-Inst_{i, \;t-1} = \alpha \;Inst_{i, \;t-1} + \sum\limits_{\,j = 1}^5 {\beta _j\cdot D_{i, \;t}^j } + \gamma A_{i, \;t} + \Gamma {X}^{\prime}_{i, \;t} + \varphi _i + \eta _t + \varepsilon _{i, \;t}$$

where Inst _i,t is a measure of country i's institutional quality in year t; $D_{i, \;t}^j$ is a series of five dummy variables signaling a democratic transition; A _i,t is a dummy variable set to one from the year of a change to autocracy; ${X}^{\prime}_{i, t}$ is a vector of time-variant control variables; φ _i is a country fixed effect; η _t is a year fixed effect; α is a coefficient; β _j is a coefficient; Γ is a vector of coefficients; and ɛ _i,t is the error term.

We control for the lagged value of institutional quality to control for convergence. Country and year fixed effects control for unobserved time-invariant country characteristics and time varying common shocks.

The variables of interest are the five $D_{i, t}^j$ dummies that capture the timing of democratic transitions. They follow Papaioannou and Siourounis's (Reference Papaioannou and Siourounis2008a, Reference Papaioannou and Siourounis2008b) coding. $D_{i, t}^1$ and $D_{i, t}^2$ capture the change in the dependent variable in the years preceding the transition. Specifically, $D_{i, t}^1$ is set equal to one in the fifth, fourth, and third pre-democratization years, while $D_{i, t}^2$ is set to one in the second and first pre-democratization years and the transition year. $D_{i, t}^3$ is set to one during the first, second, and third years after the transition. $D_{i, t}^4$ is set to one during the fourth, fifth, and sixth post-transition years. Finally, $D_{i, t}^5$ equals one from the seventh year after the transition onward. All dummies are equal to zero elsewhere. They are also set to zero if the democratic transition was reversed within 5 years. Figure 1 summarizes the definition of these dummies.

Figure 1.

Definition of democratic transition dummies.

This coding for the timing of the transition enables us to capture anticipation effects and unrest in the lead-up to the transition through $D_{i, t}^1$ and $D_{i, t}^2$.Footnote ³ The other three variables capture the aftermath of the transition, from the short run, with $D_{i, t}^3$, to the long run, with $D_{i, t}^5$. The implicit base period is the non-democratic years.

The above regression constitutes a difference-in-difference model, where countries that have undergone a transition are the treated group, while non-reforming countries serve as the control group (Papaioannou and Siourounis, Reference Papaioannou and Siourounis2008a, Reference Papaioannou and Siourounis2008b). Thanks to the inclusion of country and year fixed effects, coefficients β_j measure the change in institutional quality between two years. The change is allowed to differ across the five periods over which variables $D_{i, \;t}^j$ are defined.

For the method to give causal estimates, transitions must be exogenous. That assumption is backed by the fact that revolutions are to a large extent unpredictable, as Kuran (Reference Kuran1989, Reference Kuran1991) argues. Bueno de Mesquita (Reference Bueno de Mesquita2010) provides a model of regime changes that produces multiple equilibria. It implies that transitions can only be loosely related to fundamentals. One concern may be that democratic transitions are driven by economic development, as Papaioannou and Siourounis (Reference Papaioannou and Siourounis2008a) observe over the 1960–2005 period or Boix (Reference Boix2011) over the 1800–2000 period. Bjørnskov (Reference Bjørnskov2018) also reports that improvements in press freedom follow improvements in economic freedom whereas Berggren and Bjørnskov (Reference Berggren and Bjørnskov2019) observe that reforms undermine citizens' support for democracy. However, Boix (Reference Boix2011) also finds that the relationship becomes weaker during the post-war period, which covers our period of study. Moreover, Treisman (Reference Treisman2015) shows that structural factors matter in the medium run (10–20 years) but can be switched off by the contingencies of leadership. Finally, Paldam (Reference Paldam2020) studies the events perceived to have triggered regime changes and concludes that they are largely exogenous in the perspective of development. The possibility of a reverse causality driven by economic development is therefore weak.

We test the assumption that countries that undergo a transition do not differ from the others before the transition by checking that the coefficients of dummy variables $D_{i, t}^1$ and $D_{i, \;t}^2$ are statistically insignificant. This finding would signal that the countries that underwent a transition followed the same trend as, and were not different from, other countries before the transition.Footnote ⁴

Data

Indicators of institutional outcomes

The dependent variable must be time-variant and available over a long enough time span, and its variations over time must be meaningful. The ICRG index, published yearly since 1984 by the Political Risk Services Group, fulfills these requirements. It is based on experts' subjective evaluations and computed as a weighted average of 12 individual political risk indicators spanning all the dimensions of a country's institutional framework.Footnote ⁵ Among those individual indicators, democratic accountability is directly related to democratic transitions. Regressing an index containing that indicator on an indicator of democratization would be tautological. We therefore computed a ‘democratic accountability-free’ ICRG index as the sum of the 11 other basic components. We refer to it as the ICRG₁₁ index, because it is computed on 11 components out of 12. In our sample it ranges from 13.25 to 91, with a mean of 61.85 and a standard deviation of 14.52.Footnote ⁶

It could be argued that simply summing components of the ICRG₁₁ index is an arbitrary aggregation. To address that criticism, we proceed in two ways. First, we apply principal component analysis to the 11 dimensions of the ICRG₁₁ index and use the first principal component as dependent variable.Footnote ⁷ Accordingly, we allow the weight of each indicator to be determined endogenously. Second, we use each component of the ICRG₁₁ index in turn as the dependent variable. The results with the first principal component are reported in Table 1. The results with the separate components of ICRG₁₁ are discussed in detail in section 5.

Table 1.

Impact of democratic transitions on overall institutional quality: baseline results

All regressions include country and time fixed effects. Standard errors are heteroskedastic-consistent and clustered by country and year.

***Significant at 1%, **significant at 5%, *significant at 10%.

A first look at the data

Figure 2 describes the change in the ICRG₁₁ index relative to the world average around democratic transitions. The index is constructed as follows. First, we remove year fixed effects from the ICRG₁₁. Specifically, for each year, we subtract from each country's score the world average of ICRG₁₁ scores in that year. Second, we focus on countries having experienced a transition, and compute the average of the transformed ICRG₁₁ scores five, four, etc. years before the transition, at the transition year, and one, two, etc. years after the transition. Stopping at the 8th year after the transition, we obtain 14 numbers corresponding to the transformed ICRG₁₁ scores around the year of a democratic transition. Finally, for the sake of clarity we subtract from each of the 14 numbers the one corresponding to the transition year to obtain a normalized index.

Figure 2.

Change in the transformed ICRG₁₁ index around democratic transitions.

Figure 2 shows a clear difference between the average transformed ICRG₁₁ before and after transitions. The index follows no particular trend prior to the transition. This is noteworthy, because it suggests that countries where a transition occurs are not different from other countries during the 5 years leading to the transition. It, therefore, lends credence to a causal interpretation of the statistical estimates that will be reported in the next sections.

The ICRG₁₁ index improves and trends upward after the transition. Moreover, the difference between the index in the transition year and in the following years is always statistically significant beyond the 5% level.Footnote ¹² As a result, 5 years after the transition, the index is on average 4.48 points higher than in the transition year. It is 5.19 points higher than in the year immediately preceding the transition. Such an improvement would not bring a risky country to Western European standards, but it is non-negligible. It amounts to a third of the standard deviation of the ICRG₁₁ index in our sample. It is of the same magnitude as the differences between the ICRG₁₁ indices of Iran and India, between the indices for Nigeria and Ivory Coast, or between the indices for Mexico and Costa Rica, at the end of our period of study.Footnote ¹³

Baseline estimates

Table 1 reports the results of the estimations of model 1 using ordinary least squares (OLS), with heteroskedastic-consistent standard errors clustered by country and year. Although the ICRG₁₁ index is theoretically bounded by 0 and 94, none of the observations in our sample hits the bounds. Hence, OLS gives the same results as truncated-variable estimation methods. Regressions 1.1 and 1.2 use the ICRG₁₁ index as their dependent variable, while regressions 1.3 and 1.4 use the principal component of the 11 sub-indices of the ICRG index, PC. In both cases, we first estimate model 1 without controlling for autocratic transitions before controlling for the autocratic transition dummy.

In Table 1, the adjusted R ² rounds to 18%, which is reasonably high, and the F-test rejects the hypothesis that all coefficients are jointly zero in all regressions. The coefficient of the lagged value of the dependent variable is negative and significant at the 1% level, which is in line with the notion that countries with better initial institutional quality find it more difficult to improve it further.

In no regression are the coefficients of dummy variables D ¹ and D ², which cover the years leading to the transition, statistically significant. As pointed out previously, this finding is important, because it confirms that countries where a transition occurs are statistically indistinguishable from other countries during the 5 years leading to the transition, and therefore supports a causal interpretation of the coefficients.

The key result, however, pertains to the coefficients of the dummy variables capturing democratic transitions and is robust across the four regressions reported in Table 1. The first dummy that appears significant in all regressions is dummy D ³, which is equal to one during the first, second, and third years after the transition. Its coefficient is significant beyond the 5% level in all regressions, which suggests that the effect of democratic transitions can appear over a fairly short term.

The coefficient of dummy variables D ⁴ and D ⁵ is statistically insignificant at standard levels of significance in all regressions, suggesting that the bulk of the improvement in institutional quality is obtained within the 3 years following the transition.

We observe a negative impact of autocratic transitions on institutional outcomes. The coefficient of dummy variable A is statistically significant at the 5% level of confidence in both regressions 1.2 and 1.4.

To sum up our baseline results, democratic transitions have a positive and significant effect on institutional outcomes. The estimates suggest that the bulk of the improvement occurs during the 3 years following the transition, and we find no anticipation effect. To go beyond averages, we study in online Appendix A2 specific democratic transitions that took place in Bangladesh, Senegal, Hungary, and Nicaragua and neatly illustrate baseline results (Méon and Sekkat, Reference Méon and Sekkat2021).

The finding that the coefficients of the two dummy variables capturing the effect of democratic transitions in the medium- and long-run, D ⁴ and D ⁵, although positive are statistically insignificant suggests that the trajectories of countries that have democratized becomes more heterogenous in the long run. Although some countries may keep on reaping the benefits of their democratic transition, others may backslide. The case of Hungary may illustrate that point. Although it improved the quality of its institutions in the aftermath of its transition, it has slid back in the longer run under Viktor Orban's leadership. At the same time, its ICRG₁₁ index decreased after reaching a peak in the late 90s.Footnote ¹⁴

The results hold qualitatively both when the ICRG₁₁ and the principal component of the 11 sub-indices are used. The results are, therefore, not driven by the way in which the information contained in the components of the ICRG index is aggregated.

Quantitatively, the ICRG₁₁ index improves by a little less than one point per year during the 3 years following the transition. During the same years, the principal component improves by about 1.2 points per year, resulting in an improvement of nearly 3.6 points 3 years after the transition. Although not negligible, the effect is not massive if one bears in mind that the ICRG₁₁ has a mean of 62.18 and a standard deviation of 13.68. Over the first 3 years of the democratic transition, the index therefore increases by around a little less than a third of its standard deviation. This finding is in line with Hayek's (Reference Hayek2012: chapter 1) argument that institutions are stable or Acemoglu and Robinson's (Reference Acemoglu and Robinson2008) contention that former elites can invest in de facto power.

Robustness checks

We checked the robustness of our results to the definition of democratization episodes, and therefore selected democratization episodes using three basic datasets separately. Specifically, we first only considered jumps in the PolityIV index, then only jumps in the Freedom House status. We finally used the classification of democratic transitions by Acemoglu et al. (Reference Acemoglu, Naidu, Restrepo and Robinson2019), which includes transitions that do not meet the 5-year stability condition. Our results survived those robustness checks and are therefore robust to alternative definitions of democratic transitions.Footnote ¹⁵

To make sure that our results are not specific to the way in which we coded periods preceding and following transitions, we also considered codings imposing respectively less and more structure on the estimated relationship.Footnote ¹⁶ We first distinguished the period after the transition from the period preceding it. The dummy capturing the after-transition period was found to bear a positive and significant sign. Second, we also defined one dummy variable for each year, ranging from 5 years before the transition to 6 years after the transition, to avoid pooling years that may be different. None of the pre-transition dummy variables were significant, again suggesting that transitions were not anticipated. The post-transition dummies showed that the bulk of the effect appears within 4 years after the transition. The two robustness checks therefore confirmed our results.

Our results so far rest on specifications where time-invariant country characteristics are controlled for using country fixed effects, while changes common to all countries in a given year are controlled for using year fixed effects. We therefore added time-variant control variables: GDP per capita, openness to trade, secondary enrolment, the ratio of government consumption to GDP (all from the World Development Indicators database), press freedom (from Freedom House's 2014 historical dataset), and a series of time-invariant regional dummy variables.Footnote ¹⁷

Although the relationship is complex and unclear over our period of study, GDP per capita has been found to correlate with democracy (Boix, Reference Boix2011). Friedman (Reference Friedman1962) emphasized the role of economic openness in fostering democracy, arguing that the diffusion of liberal norms may exert pressure on autocrats to expand political rights. Education is one of the requisites of democracy according to Lipset (Reference Lipset1959). Brunetti and Weder (Reference Brunetti and Weder2003) argue that press freedom is a check on corruption. GDP, education, and a free press have in addition all been found to correlate with at least one dimension of institutional quality (see, e.g. early contributions by Brunetti and Weder (Reference Brunetti and Weder2003), Treisman (Reference Treisman2000), and Wei (Reference Wei2000)). The results obtained when controlling for those variables confirm that democratic transitions are followed by an improvement in institutional quality, in line with the findings of the baseline estimations.

Finally, because our period of study includes the end of the Cold War, our results might be affected by the inclusion of former socialist countries in the sample. The prospect of joining and receiving support from the European Union may have resulted in the expectation of what Aidt et al. (Reference Aidt, Albornoz and Gassebner2018) refer to as a ‘golden hello’ putting some of those countries on a specific trajectory around their transition to democracy. We, therefore, ran specific regressions where former socialist countries were dropped from the sample. Previous findings were unchanged.Footnote ¹⁸

Distinguishing types of transitions

We run three pairs of regressions distinguishing respectively full and partial transitions, transitions out of a military dictatorship and out of a civil dictatorship, and transitions that also resulted in a transition out of communism or not.Footnote ¹⁹

In baseline regressions, a country is coded as having undergone a transition even if its PolityIV score is only marginally positive or if Freedom House classifies it as ‘partly free’. Such countries are more democratic than before but Epstein et al. (Reference Epstein, Bates, Goldstone, Kristensen and O'Halloran2006) point out that partial democracies often act in a way that differs from fully democratic countries.

To deal with this issue, we again followed Papaioannou and Siourounis (Reference Papaioannou and Siourounis2008b) and coded as partial democratic transitions those resulting in the Freedom House index remaining ‘partly free’ or in the PolityIV index remaining below seven points. By contrast, full democratic transitions are transitions that prompted the Freedom House index to be ‘free’ and the PolityIV index to exceed seven points. Our dataset contains 13 full and 41 partial transitions. Accordingly, the two series of D^j are constructed in the same way as in baseline regressions but distinguish partial and full transitions. We then included the two series of dummies in our regressions.

The results pertaining to full and partial democratizations are the same across regressions.Footnote ²⁰ First, we see no evidence of anticipation effects for either of the two types of transitions, as the coefficients of D ¹ and D ² are statistically insignificant. Second, the effects of both partial and full democratic transitions are positive, appear entirely in the 3 years that follow the transition and are of similar magnitude. Specifically, the coefficients of D ³ are positive and significant at the 5% level in both regressions. The effects of full and partial democratizations are of a similar magnitude, slightly above one point of the ICRG₁₁ index. Finally, all the other dummy variables capturing post-democratization periods are insignificant at standard levels of statistical significance.

Acemoglu et al. (Reference Acemoglu, Ticchi and Vindigni2010) argue that the transition out of military regimes may be more difficult, because of the threat that a stronger military poses to the new regime. We therefore distinguished transitions out of military regimes from other transitions. We use Cheibub et al.'s (Reference Cheibub, Gandhi and Vreeland2010) dataset updated by Bjørnskov and Rode (Reference Bjørnskov and Rode2020a), who define a regime as military if the effective head is or was a member of the military by profession. In our dataset, there are 20 transitions out of military regimes and 34 other transitions. We then coded one series of D^j dummies for transitions out of a military regime, and another series for transitions out of other dictatorships.

The results show that transitions from non-military regimes result in an improvement of the ICRG₁₁ index in the second 3-year period following the transition, as the coefficient of Dummy D ⁴ is positive and statistically significant beyond the 5% level.Footnote ²¹ The order of magnitude of the coefficient is also the same in the two regressions. We also observe an improvement of the ICRG₁₁ index after transitions out of military regimes, as D ³ is significant at the 5% level. An interesting feature of transitions out of military regimes is that they tend to be preceded by a deterioration of the ICRG₁₁ index, as D ¹ exhibits a negative coefficient statistically significant at the 5% level.

Finally, we distinguished transitions that were accompanied by a move out of communism from other transitions, because implementing political and economic reforms at the same time may be more difficult than simply reforming the political system. To take that possibility into account, we again coded two series of D^j dummy variables, one for transitions out of dictatorship only, and another for transitions out of both dictatorship and communism. The data come from Cheibub et al. (Reference Cheibub, Gandhi and Vreeland2010). It classifies a regime as communist if the country leader is the head of the Communist Party. Our dataset features 14 transitions out of communism and 40 other transitions.

The results of the regressions are reported in the last two columns of Table A12 in the online Appendix. They show that transitions out of dictatorships that already had a market economy result in an improvement of the ICRG₁₁ index within 3 years after the transition, as the coefficient of Dummy D ³ is positive and statistically significant beyond the 5% level in both regressions and has the same magnitude. Conversely, results for transitions out of and communism resulted in a deterioration of the ICRG₁₁ index, as the coefficient of dummy variable D ⁴ is negative and statistically significant at the 5% level.

Components of the ICRG index

The overall change in the index may hide differences between specific components. We therefore separately estimated model 1 using each component of the ICRG₁₁ index as dependent variable.Footnote ²² The results show that the components of the ICRG₁₁ index can be classified into three groups.

The first group features indices whose change around democratic transitions is in line with that of the ICRG₁₁ index. It consists of the corruption, law and order, and military in politics indices. Specifically, the results obtained with those indices show no sign of anticipation effects, as the coefficients of dummy variables D ¹ and D ² are insignificant at standard levels of significance. Conversely, the coefficient of variable D ³ is positive and statistically significant. Finally, the coefficients of variables D ⁴ and D ⁵ are insignificant. The military in politics index behaves slightly differently from the other two indices in that group, because it keeps on improving after the immediate aftermath of transitions. Specifically, variables D ⁴ and D ⁵ exhibit positive and statistically significant coefficients when that variable is the dependent variable.

The second group of components of the ICRG₁₁ consists of the external conflict and internal conflict indices, which both improve after the transition. The difference between the behavior of the external conflict index and the ICRG₁₁ index is that it starts improving in the period that spans the transition and is captured by variable D ², whose coefficient is positive and statistically significant at the 1% level. It keeps on improving in subsequent periods. The internal conflict index improves after transitions but deteriorates in the first pre-transition period coded by D ¹.

Finally, a third group contains sub-indices that deteriorate either before or after transitions. In that group, bureaucratic quality and ethnic tensions only show signs of deterioration in the first pre-transition period and no significant change thereafter. The investment profile index deteriorates only in the period that spans the transition. Religious tensions and socioeconomic conditions deteriorate both before and after transitions. Government stability only deteriorates in the immediate aftermath of transitions.