Unravelling semi-presidentialism: democracy and government performance in four distinct regime types

Do semi-presidential regimes perform worse than other regime types? Semipresidentialism has become a preferred choice among constitution makers worldwide. The semi-presidential category contains anything but a coherent set of regimes, however. We need to separate between its two subtypes, premierpresidentialism and president-parliamentarism. Following Linz’s argument that presidentialism and semi-presidentialism are less conducive to democracy than parliamentarism a number of studies have empirically analysed the functioning and performance of semi-presidentialism. However, these studies have investigated the performance of semi-presidential subtypes in isolation from other constitutional regimes. By using indicators on regime performance and democracy, the aim of this study is to examine the performance of premier-presidential and presidentparliamentary regimes in relation to parliamentarism and presidentialism. Premierpresidential regimes show performance records on a par with parliamentarism and on some measures even better. President-parliamentary regimes, on the contrary, perform worse than all other regime types on most of our included measures. The results of this novel study provide a strong call to constitution makers to stay away from president-parliamentarism as well as against the idea of thinking about semipresidentialism as a single and coherent type of regime. ARTICLE HISTORY Received 19 January 2017; Accepted 22 May 2017

Do semi-presidential regimes perform worse than other regime types? Semi-presidentialism has become a widespread choice among constitution makers around the world and the study of semi-presidentialism has emerged as a burgeoning research field in the last two decades. Following the warning once raised by Linz 1 that presidentialism and semi-presidentialism are less conducive to democracy than parliamentarism a number of studies have empirically analysed the functioning and performance of semi-presidentialism. 2 Several studies have added to Linz's line of argument about the dangers associated with semi-presidential constitutions. 3 Others have challenged the Linz proposition, stressing the mixed performance of semi-presidential countries 4 as well as the potential for power-sharing and flexible executive relations 5 afforded by this form of government.
Elgie defines semi-presidentialism as a system where the constitution includes both a popularly elected president and a prime minister and cabinet accountable to the parliament. 6 With this inclusive definition there are over 50 countries with some kind of semi-presidential constitution. 7 Particularly considering the diversity in the level of presidential power, semi-presidentialism becomes inadequate as a single explanatory variable, however. We need to separate different forms of semi-presidentialism. Different alternatives exist in the literature 8 and the one that has received the broadest acceptance is Shugart and Carey's subcategorization between premier-presidential and president-parliamentary regimes. 9 Their defining criteria is that in premier-presidential regimes the government can only be dismissed by parliament whereas in president-parliamentary regimes both the president and the parliament have the power of government dismissal. In this article, we take Shugart and Carey's distinction as our point of departure and test their warning that president-parliamentarism is less conducive to democracy than premier-presidentialism. Among scholars there seems to be some support for the perils of president-parliamentarism hypothesis once raised by Shugart and Carey. 10 However, with the notable exception of Elgie,11 there are no large-N studies available where democracy and government performance are actually measured across the two types of semi-presidentialism. Elgie's systematic study offers new findings on a cross-national and crossregional level between the two types of semi-presidentialism, but it does so in isolation from parliamentary and presidential regimes. We strive to fill this existing research gap.
By using indicators on regime performance and democracy from a dataset containing 173 countries, the aim of this study is to examine the performance records of premier-presidential and president-parliamentary regimes in relation to parliamentarism and presidentialism. Our endeavour is guided by Linz's argument on the "perils of presidentialism", 12 and by Shugart and Carey's proposition that president-parliamentary regimes are more perilous to democracy than other regime types.
Our results are a strong call against the idea of thinking about the performance of semi-presidentialism without separating between its two subtypes. We will show that premier-presidential regimes have performance records on a par with parliamentarism and on some measures even better. However, president-parliamentary regimes show performance records worse than all other regime types on most of our included measures, in particular when it comes to democracy.
We start by defining the four separate regime types alongside a theoretical account of the argument about the perils of (semi-)presidentialism from which we derive our propositions for later empirical analysis. The theoretical discussion forms the basis for three hypotheses about the performance of the four regime types with regard to democracy and governance. The empirical section starts with a general and descriptive part where we report on presidential powers and key patterns with regard to performance among the four distinct types of regimes followed by a series of regression analyses of regime type and performance. The conclusions sum up the findings and argument of our study.

Four distinct regime types
The distinction between parliamentarism and presidentialism is quite straightforward. Parliamentarism has an authority structure based on mutual dependence. The prime minister and his or her cabinet is dependent on the consent of the parliament, and parliament in turn is dependent on the prime minister, who is entitled to dissolve parliament and call new elections. The head of state (the president or monarch) upholds mainly ceremonial powers and is not directly elected. Presidentialism, in contrast, is defined by a popularly elected president that selects and directs the cabinet, in which the terms of the president and parliament are fixed. 13 Although some scholars would disagree that there is a single and generally accepted definition of parliamentarism and presidentialism, defining semi-presidentialism has proved an even more contested task. Duverger provided a definition of semi-presidentialism including three criteria: (1) the president is elected by universal suffrage; (2) the president possesses quite considerable powers; and (3) there is also a prime minister and other ministers who possess executive and governmental power and can stay in office only with the consent of the parliament. 14 Since Duverger's founding definition there has been an ongoing debate on how to define and categorize regimes with a dual executive including both a president and prime minister. Especially the second and non-institutional criterion, that "the president possesses quite considerable powers", has been a source of debate and confusion. Different scholars have approached this vague criterion differently and the list of semi-presidential countries has varied extensively from one study 15 to another. 16 In the late 1990s and early 2000s comparative scholars began to increasingly accept the use of strictly constitutional definitions. Elgie came to propose an inclusive version, which has gained academic prominence, stating that "semi-presidentialism is where a constitution includes a popularly elected fixed-term president and a prime minister and cabinet who are collectively responsible to the legislature". 17 Elgie's definition has attracted critique for encompassing too many and disparate countries, which have reduced its comparative value. Elgie has acknowledged this dilemma and recommended not to use semi-presidentialism as an explanatory variable. 18 Siaroff,19 and Cheibub, Elkins and Ginsburg 20 go as far as to argue that the whole category of semi-presidentialism is inadequate. Instead they suggest that scholars should stick to the main distinction between presidentialism and parliamentarism and combine this with measures of presidential powers.
This kind of critique is neither new nor irrelevant but it ignores that semi-presidentialism is unique in terms of origin and survival of the government. In principal-agent terms, the government is at the mercy of two separate agents of the electorate, that is, the president and the parliament. 21 Presidentialism incorporates both separation of origin (popular elections of president and parliament) and separation of survival (government is at the mercy of the president). Semi-presidential systems also have dual elections and thus separation of origin, but the survival of the government is dependent on the maintenance of a parliamentary majority. 22 Thus, the principal-agent structure of semi-presidentialism is distinct from both parliamentarism and presidentialism. 23 However, to take the principal-agent terms seriously, we need to make a distinction between the two subtypes of semi-presidentialism. Shugart and Carey define premierpresidentialism as being where (1) the president is elected by a popular vote for a fixed term in office; (2) the president selects the prime minister who heads the cabinet; but (3) authority to dismiss the cabinet rests exclusively with the parliament, and president-parliamentary systems as being where (1) the president is elected by a popular vote for a fixed term in office; (2) the president appoints and dismisses the prime minister and other cabinet ministers; (3) the prime minister and cabinet ministers are subjected to parliamentary as well as presidential confidence. 24 On theoretical grounds and for comparative reasons, we adhere to the subtypes by Shugart and Carey and will demonstrate their empirical relevance in the subsequent analysis. Figure 1 illustrates the key institutional relations of the four regime types.
The assumed perils of semi-presidentialism: theoretical propositions Some scholars have endorsed semi-presidentialism for its flexibility and power-sharing structure, 25 while others have associated it with institutional conflict 26 and political stalemate. 27 Linz's arguments from the early 1990s have established many of the basic elements of the regime type debate as well as the specific research on semi-presidentialism. Linz 28 raised the often-repeated argument that presidentialism is less conducive to democracy than parliamentarism.
He pointed at four "perilous" factors: (1) the president's fixed term in office, (2) the dual legitimacy, that is, both the president and the parliament rely on a popular mandate, (3) the winner-take-all character of presidential elections, and (4) the risk of personalization of power. Linz claimed that semi-presidentialism shares these features and, moreover, that the responsibility in a semi-presidential system is diffuse and that conflicts are possible and likely. 29 His warning that semi-presidentialism just like presidentialism becomes dependent on the personality and abilities of the president is certainly relevant here, but is insufficient without making the distinction between premier-presidentialism and president-parliamentarism.
The variation in principal-agent relations is expected to affect the performance of the two subtypes. The argument against president-parliamentarism revolves around the strong and independent presidency and the uncertain and dependent position of the government. If the president does not enjoy the support of a parliamentary majority, the dual loyalty of the government to both the president and the parliament is likely to produce conflict and stalemate. 30 Since both the president and the parliament have the power to dismiss the government, each institution may calculate that the best way to maximize influence is to work against rather than with the other institution. Such conflicts over appointments and dismissals are likely to lead to conflict over the regime itself. In Elgie's words, "under president-parliamentarism the president and the legislature have an incentive to act against each other, which means there is little incentive to maintain the status quo and which in turn generates instability that is likely to undermine democratic performance". 32 A key argument in favour of premier-presidentialism over president-parliamentarism is that the former provides the possibility of combining presidential leadership with a government anchored in parliament. Since the president cannot dismiss the government once it has been formed, he or she will have incentives to negotiate with the parliament in order to gain influence over the government and the political process. But again, the arguments presented above concerning the risks and consequences of intra-executive conflict explain why there are few arguments in favour of premier-presidentialism over parliamentarism. 33 Shugart and Carey explicitly warned constitution-makers "to stay away from president-parliamentary designs". 34 They referred to some troubled experiences with this system, exemplified by the cases of Ecuador, Peru, and the German Weimar Republic, and they concluded that "the experience has not been a happy one, in most cases". 35 Case studies on post-Soviet countries where the system has shifted from president-parliamentary to premier-presidential constitutions provide additional support to the negative impact of president-parliamentarism on democracy. There is for instance evidence from Armenia, Georgia, Kyrgyzstan, and Ukraine that reduced presidential powers and a shift to a more balanced semi-presidential system were associated with better democracy records. 36 A general trend among post-Soviet countries is that presidents have used their control over the administration to effectively curb the opposition, thereby directing the trajectory of constitutional developments in their own favour. Such political transitions can be summarized as rapid and deliberate processes of centralizing and concentrating authority in the presidency. This has consisted primarily of establishing strong presidential rule, reformulating relations between the central government and the regional and local administration, maintaining direct control over the media, manipulating elections, reinvigorating a centrally managed party system, and actively excluding organized political opposition.
Previous empirical results concerning democratization and democratic survival suggest that regime-types with weaker presidents are more conducive to democracy than others, although without a clear-cut ranking between all four regime types. It seems evident that the use of different definitions and case selections has thwarted attempts to fully compare results between different studies. Stepan and Skach find parliamentary regimes to be "democratic overachievers" compared to presidential ones but their study excludes OECD countries. 37 In addition, their end-date, 1989, misses many of the newer semi-presidential countries. Cheibub, Przeworski, and Saiegh 38 find that the average life expectancy of a presidential democracy is a mere 24 years, as compared to 74 years of a parliamentary democracy. Shugart and Carey, on the contrary, report diverging democratic performance for different types of presidential regimes, separating between presidentialism and the two semi-presidential subtypes. 39 Their early study, however, does not include the many post-communist countries established after 1991. In a more recent study, Elgie finds better prospects for democratic performance among premier-presidential regimes than among president-parliamentary ones. 40 Elgie's thorough study, however, is limited only to the two subtypes of semi-presidentialism and he does not include parliamentary and presidential regimes.
In general, earlier empirical studies indicate that regime type is correlated with democratic performance and that parliamentary democracies show better records on democracy than presidential ones, and that premier-presidential democracies outperform president-parliamentary ones.
We strongly argue for separating between the two subtypes of semi-presidentialism in order to have more than just a taxonomic container of semi-presidential regimes with little comparative value. The political variation among the countries that meet Elgie's constitutional definition of semi-presidentialism is so great that any effect related to the regime type itself is unlikely to be observed. Even with the sub-categories, however, we should be wary of drawing conclusions on causal effects of the regime type on performance outcomes. Although the approach of our study certainly adheres to the idea that institutions matter, there is indeed the classical problem of endogenous institutional choice. In other words, institutional effects are likely "to be the expression of the preferences that were hardwired into institutional structures at the time they were chosen". 41 Since the aim of our analysis is to reveal general patterns with regard to the four regime types and their performance outcomes, our analysis can only make very modest claims about causal effects. Nevertheless, our theoretically derived hypotheses embrace the assumption that institutions matter and that there is an interplay between various constitutional forms and the political actors who interpret those forms. Drawing on the theoretical arguments in the literature and earlier empirical research, we test the following hypotheses. H1: Parliamentarism performs better than other regime types in terms of democracy and government performance.
H2: Premier-presidentialism performs better than president-parliamentarism and presidentialism in terms of democracy and government performance.
H3: President-parliamentarism performs on a par with, or worse, than presidentialism in terms of democracy and government performance.

Cases, data and measurement
For the classification of countries into parliamentarism, premier-presidentialism, president-parliamentarism and presidentialism we have updated and recoded the typology offered by Bormann and Golder,42 which can be found in the Quality of Government Institute dataset (QoG standard dataset, January 2016 43 ). For updating and recoding we use the constitutional regime classification as of 2011 by Robert Elgie (coded as of May 2016). Using the same definitions as in our study, Elgie has systematically compiled and provided an updated list of parliamentary, semi-presidential (premier-presidential and president-parliamentary) and presidential countries. 44 In the following analyses, we use data from different sources, compiled in the QoG standard dataset. The number of countries under each regime type is reported in Table 1 (the list of countries according to regime type is reported in Table A1).
In Figure 2, we report formal presidential power by regime type. The measure of presidential power is developed by Doyle and Elgie. 45 A number of presidential power indexes are available in the literature 46 and the main advantage of this one is that it is based on 28 of such already existing measures. In addition, Doyle and Elgie have generated their dataset on a larger number of countries with longer time series than other existing ones. The scores are in the range from 0 to 1 in separate time periods following constitutional changes of a country's presidential powers.
The average presidential power scores provide additional support for separating between premier-presidential and president-parliamentary regimes. As expected, the parliamentary countries display the lowest score with a mean of 0.183, followed by the premier-presidential countries (0.253).
There is thus a substantial difference between the parliamentary and premier-presidential regimes on the one side, and the president-parliamentary (0.473) and the presidential regimes (0.465) on the other. In terms of presidential power, premierpresidential regimes are closer to the parliamentary regimes while president-parliamentary regimes come out as even more "presidential" than presidential regimes. This ordinal scale of presidential powers among the four regime types confirms our argument against the idea of thinking about semi-presidentialism as anything else than a Comment: A one-way ANOVA was conducted to determine if presidential power was significantly different for the different regime types. There was a significant difference between groups as determined by one-way ANOVA [F (3,104) = 16.64, p = 0.0000]. A Tukey post-hoc test showed that presidential power was significantly higher in the president-parliamentary group compared to the premier-presidentalism group (0.219 ± 0.052, p = 0.000). However, the difference between premier-presidentalism and parliamentarism was not statistically significant (0.070 ± 0.055, p = 0.584).
taxonomic definition with marginal comparative value at best. Although we cannot do away with within-category variation, the sub-categories of premier-presidentialism and president-parliamentarism help to significantly reduce heterogeneity. To the extent that previous studies have indicated that strong presidential powers are associated with negative democratic performance, this marked difference between the two semi-presidential regime subtypes is in line with our hypotheses. In other words, we have reason to assume that in our subsequent analyses we would find president-parliamentary regimes to be on a par or below the performance records of presidential regimes, whereas premier-presidential regimes would be closer to the parliamentary ones.

Democratic performance
In previous studies, the performance of the semi-presidential subtypes has been assessed in isolation from parliamentary and presidential regimes. 47 We set out to investigate the performance of semi-presidential regimes from a broader comparative perspective, including also parliamentary and presidential regimes. Recall our hypotheses where we expect a general pattern where parliamentary regimes display the best performance (H1) and where premier-presidential regimes outperform president-parliamentary regimes (H2 and H3).
We start out by investigating the performance of the different regime types with regard to democracy. Figure 3 presents mean values on four frequently used indicators of democracy for each regime type. The first indicator is an index combining the two Figure 3. Democratic Performance by Regime Type. Comment: Bars represent means. All variables have been rescaled into a scale ranging from 0 to 1. N = 145. In order to test the statistical significance of the differences, a series of one-way ANOVAs were conducted. All ANOVAS were significant (p < 0.000). A Tukey post hoc test showed that for all democratic performance variables, there is a statistically significant difference between premier-presidentialism and president-parliamentarism (p <0 .01). most frequently used democracy indicators, Freedom House's index of civil liberties and political rights and the Polity IV index. It has been shown that this average index performs better in terms of validity and reliability than its constituent parts. 48 We also use the Polity IV index 49 on its own. In addition, we take advantage of the Economist Intelligence Unit's index of democracy, 50 which is a broader measure covering five dimensions of democracy (electoral process and pluralism, functioning of government, political participation, political culture, and civil liberties). The last indicator is the Executive constraints indicator from Polity IV, which refers to the extent of institutionalized constraints on the decision-making powers of chief executives. Thus, this variable is not a measure of the level of democracy per se but rather an indicator of to what extent the legislative branch has the constitutional means to guard against abuses of executive power, which has recently been demonstrated to have a strong impact on the survival of new democracies. 51 For the sake of comparison, all democracy variables have been rescaled to a scale ranging from 0 to 1.
In general, Figure 3 shows that parliamentary and premier-presidential regimes score higher on all four variables than presidential and president-parliamentary regimes. Moreover, opposite to our expectations, premier-presidential regimes actually outscore parliamentary regimes on three indicators (Polity IV, Economist, and Executive constraints), thus demonstrating the best average performance. What is striking is the difference in democratic performance between president-parliamentary and premier-presidential regimes, where the former shows very low levels of performance, in particular when it comes to the Economist index, which admittedly employs a more maximalist definition of democracy than the other indices. As expected, countries with a pure presidential regime consistently demonstrate lower levels of democracy than parliamentary and premier-presidential regimes. Presidentialism, however, outperforms the president-parliamentary form of semi-presidentialism.

Government performance
We will now turn more explicitly to the output side of the political system. While empirical studies assessing the democratic performance of semi-presidential regimes have been quite frequent, much less attention has been devoted to the question of how different institutional arrangements affect government performance. 52 In the following, we look at four different indicators, representing a broad conception of government performance. First, we use the "Government effectiveness" indicator from the Worldwide Governance Indicators, 53 which is often used as a dependent variable in the literature on "quality of government". The variable captures perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressure, the quality of policy formulation and implementation, and the credibility of the government's commitment to such practices. 54 In relation to government effectiveness, the issue of corruption is highly important. In recent years, the problem of corruption has been argued to be among the most pressing challenges to development in large parts of the world. To measure regime performance in terms of control of corruption we use the "Corruption Perceptions Index (CPI)" 55 from Transparency International. Regime performance is not only about what the regime delivers in terms of economic performance or an effective and impartial public administration but also the absence of undesirable state action. We therefore set out to assess regime performance in terms of respect for human rights, using the "Empowerment rights index" from CIRI Human Rights Data Project. 56 This additive index measures regime performance in terms of rights of foreign and domestic movement, freedom of speech, freedom of assembly and association, workers' rights, electoral self-determination, and freedom of religion. 57 To a certain extent, this variable admittedly captures dimensions of both democratic and government performance. Lastly, we assess regime performance in terms of basic human development, using the "Human development index (HDI)" from the United Nations Development Programme. 58 This frequently used index measures achievements in a country along three basic dimensions (a long and healthy life, knowledge, and standard of living). In the analyses, all variables have been rescaled to a 0-1 scale.
The descriptive data on regime performance presented in Figure 4 show a pattern very similar to the one with democratic performance (Figure 3). In general, parliamentary regimes outperform other regime types on all indicators. Moreover, although the differences between the two types of semi-presidential regimes are not as pronounced as for democratic performance, premier-presidential regimes score consistently higher than president-parliamentary regimes on all indicators. And, as was also the case with democracy, president-parliamentary regimes consistently display the worst performance on all measures of government performance, although it is a very close call when it comes to the extent of corruption, where presidential regimes also perform poorly. In order to test the statistical significance of the differences, a series of one-way ANOVAs were conducted. All ANOVAS are significant (p < 0.01). A Tukey post hoc test shows no significant differences between premier-presidentialism and president-parliamentarism. The most consistent difference is the one between parliamentarism and president-parliamentarism, which is statistically significant on all government performance variables (p <0 .05). There is also a significant difference (p <0 .01) between parliamentarism and presidentialism on all variables with the exception of the empowerment rights index (p =0 .08).

Multivariate analyses of regime type and performance
At face value, the descriptive analysis above demonstrated a clear pattern with regard to the two types of semi-presidential regime. In order to further disentangle the difference in performance between the two types of semi-presidential regime, we now turn to a multivariate analysis, where we also take into account other potential determinants of regime performance. Figure 5 presents the main results from a series of ordinary least squares (OLS) regressions conducted with the intention of further investigating the relationship between regime type and democratic performance. The dependent variables are the same indices that were presented in Figure 3. The figures present the regression coefficients for the two types of semi-presidentialism and pure presidentialism, compared to parliamentarism, which constitutes the reference category. In the models, we include a number of control variables that have been shown to be important determinants of democratic performance. Following Norris' analogous analysis of constitutional systems and democratic performance and Elgie's study of semi-presidential subtypes and democracy, 59 all models include controls for gross domestic product (GDP)/capita based on purchasing power parity from the World Development Indicators, ethnic fractionalization, 60 population size, and dummies for proportional representation, 61 former British colony, the Middle East and North Africa (MENA) region, Western Europe and North America. 62 The data are taken from the QoG standard dataset, January 2016. The full models are presented in Table A2 in the appendix.
When it comes to level of democracy as measured by the combined Freedom House/ Polity index, the only type of regime to perform significantly worse than parliamentary regimes is the president-parliamentary regime type. This is the case also when employing the indices of democracy produced by Polity IV and the Economist. When it comes Figure 5. Regime types and democratic performance (OLS coefficients with 95% confidence intervals). Comment: The graphs are based on the regressions presented in Table A2 (appendix). Parliamentarism is used as the reference category.
to Executive constraints, president-parliamentary and presidential regimes display significantly worse, although the significance level for the latter is lower (p < 0.1). Actually, premier-presidential regimes fare marginally better than parliamentary regimes on all four measures of democracy, although not significantly so. Figure 6 presents the corresponding analyses with the four indicators of government performance as dependent variables, applying the same controls as in the analyses of democratic performance. 63 Just as in the descriptive analysis, the differences in terms of performance are less pronounced. The only model demonstrating significant differences is the one with the empowerment rights index, where presidential-parliamentary regimes perform worse than parliamentary regimes. This is noteworthy, since the empowerment rights index is arguably a measure containing elements of both government and democratic performance.
In general, we find the same pattern as with democratic performance where the premier-presidential regime type comes close to parliamentary regimes, although the former places itself on the negative side with regard to government performance (except for CPI).
We have also conducted the regression analyses using Doyle and Elgie's presidential power index (see Figure 2) as a replacement for the regime dummies used in the regressions presented above. The results are also very similar. Presidential power is negatively related to all four measures of democratic performance (p < 0.05). We find the same relationship between presidential power and our four indicators of government performance; however it is only statistically significant for government effectiveness and HDI. The regression analyses are presented in the appendix (Tables A4 and A5).  Table A3 (appendix). Parliamentarism is used as the reference category.
In terms of conclusions, our argument about the importance of distinguishing between the two types of semi-presidential regimes is strengthened by the data. In general, premier-presidential regimes perform similar to parliamentary regimes, while president-parliamentary regimes display performance records more similar to pure presidentialism, although slightly worse on most indicators. However, since our empirical analyses draw only on cross-sectional data with a limited number of conventional controls, future studies should devote strong efforts towards identify causal relationships.

Conclusions
Do semi-presidential regimes perform worse than other regime types? As we have shown, the answer depends on what subtype of semi-presidential regime we have in mind. Anchored in the assumptions of neo-Madisonian theory, the principal-agent approach employed in previous studies has revealed the normative intentions built into presidential and semi-presidential regimes. Accordingly, scholars have, on a theoretical basis, divided semi-presidentialism into the two subtypes of premier-presidentialism and president-parliamentarism. In this article, we have also demonstrated the empirical relevance of this distinction in relation to parliamentary and presidential regimes.
Guided by the argument once raised by Linz about the "perils of presidentialism", and by Shugart and Carey's proposition that president-parliamentary regimes are less conducive to democracy, we proposed three hypotheses.
Our data do not provide support for the statement that parliamentarism performs better than all other regime types in terms of democracy and government performance (H1). Rather we observed a pattern where premier-presidentialism performs almost as goodand on a few measures even betteras parliamentary regimes. Neither the measures of democracy nor the measures of government performance show significantly better records for parliamentary regimes than for premier-presidential ones. This indicates that a parliamentary constitution with an indirectly elected president does not necessarily go along with better political performance than a premier-presidential one with a popularly elected but weak or medium weak president. Thus, to the extent that we think about semi-presidentialism in terms of premier-presidential regimes, we have good reasons to disregard strong propositions about the "perils of semi-presidentialism". If the most positive accounts about semi-presidentialism are relevant, such as executive flexibility, power-sharing, and a uniting president, those are likely to be identified under the premier-presidential form of government. Thus, our data give no support for general recommendations to avoid dual executives or a popularly elected president with limited powers.
However, the picture certainly looks different with regard to president-parliamentary regimes. While premier-presidential regimes are closer to parliamentary regimes, president-parliamentary regimes display performance records more similar to pure presidentialism, and the regime type performs even worse on most indicators (H2, H3). When it comes to the level of democracy, the only regime type to perform significantly worse than the parliamentary oneon four separate measures and with conventional controlsis the president-parliamentary regime type. The differences in terms of government performance are less pronounced but the pattern is similar to the one we observed for democratic performance. Presidential-parliamentary regimes score significantly worse than parliamentary regimes when it comes to empowerment rights, which is admittedly a measure somewhere in between government and democratic performance. Although there is a tendency of slightly poorer performance by presidentialparliamentary regimes in terms of government performance, and significantly so on only one indicator, our results demonstrate that the type of constitutional system seems to affect democracy more strongly than government performance.
In this article, we have mostly refrained from making claims about mechanisms behind the observed pattern and we are well advised to keep to this position. However, we cannot escape a general comment on the importance of presidential powers. As we have shown, variation in presidential powers follows closely the four regime typesweakest among the parliamentary regimes and strongest among the president-parliamentary regimes. Various case studies on, for example, post-Soviet countries provide empirical support to the negative impact of strong presidential powers on democracy. President-parliamentary constitutions accumulate power in the hands of presidents that are often not very interested in promoting democratic reforms. The outcome has been increased power of already powerful presidentsarguably a straight road to the consolidation of autocracy. Shugart and Carey once warned constitution makers to stay away from the president-parliamentary form of government. Our findings confirm the relevance of their decisive recommendation.
The empirical analysis is limited to the extent that it draws on cross-sectional data only and with a limited set of controls, and we recognize the need for further and more sophisticated analyses in future research. We should also be wary of drawing far-reaching conclusions on causal effects of regime types on performance and outcomes. We adhere to the notion that institutions matter although our analysis provides no robust answers on independent institutional influence. Where there appears to be an institutional effect, it might be just a reflection of the preferences of the actors involved in the original process of institutional choice. And we certainly acknowledge that there are many factors affecting democratic and government performanceeconomic, social and political. Yet, we have revealed a general pattern with regard to the four regime types on performance. Based on our findings, we claim that democratic performance is likely to be better with a parliamentary or premier-presidential form of government.
Finally, we claim that discussions about the pros and cons of semi-presidentialism should include the distinction between its sub-categories and consider dimensions of presidential power. By itself, semi-presidentialism should not be used as a discrete explanatory variable.