The problems of semi-presidentialism

In semi-presidential systems, the president is directly elected and serves for a fixed term, while the prime minister and cabinet are collectively responsible solely to the legislature. The academic consensus against semi-presidentialism is profound. For example, Linz states that: ‘In view of some of the experiences with this type of system it seems dubious to argue that in and by itself [semi-presidentialism] can generate democratic stability’ (1994: 55). Valenzuela (Reference Valenzuela2004: 17) argues that semi-presidentialism ‘may not solve some of the inherent problems of presidentialism, and indeed could make them worse by reifying the conflict between two state powers and personalizing them in the figure of the president and the prime minister’. Stepan and Suleiman recommend against semi-presidentialism arguing that it ‘is a more risk-prone system than the modern parliamentarism that has evolved in Europe other than France after World War II’ (Stepan and Suleiman, Reference Stepan, Suleiman, Chehabi and Stepan1995: 412).

In this article, there is no attempt to compare the performance of countries with semi-presidential constitutions relative to those with parliamentary and/or presidential constitutions. Thus, we do not aim to contribute to the more general debate as to whether parliamentarism should be chosen ahead of both semi-presidentialism and presidentialism as the standard wisdom suggests. However, we seek to contribute to the literature by assessing the extent to which the problems of semi-presidentialism apply to different semi-presidential regimes. There are three main arguments against semi-presidentialism.

Sample, hypotheses, variables, model specification, and findings

In this article, semi-presidentialism is defined as the situation where there is a directly elected president and a prime minister and cabinet who are responsible to the legislature. This is now a common way of defining this type of system (see, for example, Elgie, Reference Elgie2005; Skach, Reference Skach2005; Shugart, Reference Shugart2005, Reference Shugart, Rhodes, Binder and Rockman2006). It should be noted that this definition makes no reference to the powers of the president. Therefore, a country like Ireland with a figurehead president, but nonetheless a directly elected president, should be classed as semi-presidential as well as a country like Russia with a very powerful directly elected president. In one sense, such a definition seems counterintuitive. However, there are at least two reasons for adopting this definition. Firstly, a definition that includes reference to the powers of the president leads to a problem of selection bias. If the list of semi-presidential countries includes only those countries with at least moderately powerful countries, then it is hardly surprising that semi-presidentialism is associated with the standard problems of presidentialization. By defining semi-presidentialism without reference to the powers of the president, we avoid any problem of selection bias. Certainly, it means that we should not operationalize semi-presidentialism as a discrete explanatory variable. However, it also means that we can explore the effects of variation within semi-presidentialism. To what extent does this variation matter? We hypothesize that it does matter and that semi-presidential countries with stronger presidents are likely to perform worse than those with weaker presidents. Secondly, it must be acknowledged that some countries choose to have directly elected presidents with very few powers. These countries choose to operate in a parliamentary-like manner, but, for whatever reason, they also choose to directly elect their president. This is a discrete constitutional choice and it is a choice that is different from a parliamentary system with an indirectly elected president or a monarch. Are there benefits to combining a directly elected and weak president and a strictly parliamentary system? We hypothesize that such countries may perform better than those that choose to combine a directly elected and powerful president with a government that is responsible to the legislature.

To identify a semi-presidential country on the basis of our definition, it is necessary simply to read the country's constitution. This means that there is very little room for the list of semi-presidential countries to vary from one writer to the next. There is no need to make a call as how powerful a country's president must be in order for it to be classed as semi-presidential. Instead, it is simply necessary to identify those countries that have both a directly elected president and a prime minister and cabinet that are responsible to the legislature. That said, there are still some judgment calls to be made. For example, we exclude countries such as South Korea, where the legislature has to consent to the individual appointment of the prime minister, rather like the case of cabinet nominations in the US, and where, once appointed, the legislature has no means to dismiss the government. All the same, we ensure that our findings are not sensitive to case selection on the basis of these and other definitional issues. Table 1 provides a list of countries with semi-presidential constitutions as of June 2008.

Table 1.

Countries with semi-presidential constitutions, 2008

Plainly, the list of semi-presidential countries in Table 1 is very diverse in terms of their democratic status. It includes some countries that are unequivocally autocratic, such as Chad, other countries that are unequivocally consolidated, such as Austria, and others that have started the process of democratization but are not yet fully consolidated, such as Madagascar. For the purposes of this article, it is assumed that the purportedly negative consequences of semi-presidential performance have the chance to ‘kick in’ only when a country has reached a certain point of democratization. In other words, in an autocracy the institutional effects of semi-presidentialism have no room to have a negative influence on democratic performance because the political system is so tightly controlled and democratic performance is already so poor. By the same token, it is also assumed that the negative consequences of semi-presidentialism have no impact on a country that is fully consolidated. In these countries, semi-presidentialism may have some or other effect on policy outcomes, but the quality of democracy cannot be impaired when democracy is ‘the only game in town’. Overall, it is assumed that the negative consequences of semi-presidentialism will be observable only in countries that remain incompletely consolidated and yet are not autocratic.

To identify this set of countries within the category of countries with semi-presidential constitutions, we use the measures of democracy provided by the Polity IV project.Footnote ¹ The methodology adopted by Polity is widely used in comparative analysis. The Polity project scores countries on a scale from −10 (complete autocracy) to +10 (complete democracy). This scale cannot be used as a continuous measure of democracy.Footnote ² However, there are various ways in which the scale can be adapted to the research project at hand. The authors of the Polity project make a distinction between three categories of countries. Those scoring from −10 to −6 inclusive are autocracies; those from −5 to +5 are anocracies; and those from +6 to +10 are democracies. Anocracies are transitional regimes that are ‘about three times more likely to experience major reversions to autocracy than democracies’ (Marshall and Gurr, Reference Marshall and Robert Gurr2003). Therefore, a key feature of anocracies is that they are relatively unstable and that instability is often associated with a shift to autocracy. Other writers have operationalized the Polity scores somewhat differently. For example, Przeworski et al. (Reference Przeworski, Alvarez, Cheibub and Limongi2000) adopted a dichotomous distinction between autocracies (−10 to 0 inclusive) and democracies (+1 to +10).

In this article, we assume that countries scoring 0 or below are insufficiently democratic for semi-presidentialism to have a negative effect on democratic performance. Equally, we assume that countries scoring +10 are fully consolidated and that again the institutional effects of semi-presidentialism will not affect democratic performance. However, within this range the supposed perils of semi-presidentialism may have an impact. Table 2 identifies all countries with semi-presidential constitutions that have been placed in the range +1 to +9 by Polity. We include countries, such as the Comoros, Congo-Brazzaville, Moldova, and Weimar Germany, which were semi-presidential for a period when they scored within that range, but have since abandoned semi-presidentialism. Each year that a country was semi-presidential and scored in the range +1 to +9 is one observation. There are 393 observations in total. This is a very unbalanced sample: 36 countries are observed across a period of 86 years. The number of years for which countries are observed ranges from one (Belarus) to 42 (France). The first country to be observed is Germany from 1919 to 1932, while Timor-Leste does not enter the dataset until 2002.

Table 2.

Semi-presidential case selection and power of the president

Given that we cannot use the Polity measures as a continuous variable, we dichotomize the dependent variable to measure the performance of democracy. Thus, we distinguish between two categories of countries – those that score in the range +1 to +5 inclusive (anocracies) and those that score in the range +6 to +9 (democracies). We assume that the three purported disadvantages of semi-presidentialism will be associated with poor democratic performance and, hence, with anocracies rather than democracies. In our sample, 32.7% of the observations are anocracies.

We have three explanatory variables. The first explanatory variable is the power of the president. We hypothesize that powerful presidents will be associated with anocracies rather than democracies. To measure the power of presidents, we use the scale proposed by Siaroff (Reference Siaroff2003). He identifies nine constitutional indicators of presidential power. He gives a value of 1 if the constitution includes the indicator and 0 otherwise. He then measures the powers of presidents cross-nationally within a range of 0 to 9.Footnote ³ The scores for the countries in our sample are given in Table 2. In a small number of cases, we had to measure the power of presidents ourselves because Siaroff did not code them. These scores are provided in Table 3. For our sample, the minimum score in Siaroff's schema is 1 because one of his indicators is the direct election of the president and all of the countries in our sample will score 1 for this indicator. Overall, our sample is skewed towards semi-presidential countries with strong presidents: the mean is 5.9 along a range of 1 to 9.

Table 3.

Presidential powers in semi-presidential countries not measured by Siaroff

Notes: PE = popularly elected; CE = concurrent presidential and legislative elections; AP = discretionary appointments powers; CM = chairs cabinet meetings; VT = veto power; EDP = long-term emergency or decree powers; FP = central role in foreign policy; GF = central role in government formation; DL = ability to dissolve the legislature

Scores based on the indicators in Siaroff (Reference Siaroff2003).

The second explanatory variable is cohabitation. We hypothesize that cohabitation will be associated with anocracies rather than democracies. To identify periods of cohabitation, we began by consulting www.worldstatesmen.org. This is a very thorough and reliable data source. It provides the names and terms of office of all presidents and prime ministers. It also records their party affiliation. We identified all cases when the party affiliation of the two executive actors was specifically identified and when it was different. We excluded cases where either the president or the prime minister was classed as non-partisan. We then consulted secondary sources to confirm whether the cases where the party affiliation of the president and the prime minister were different were examples of coalition government – namely where the president and prime minister were from different parties but where the president's party was represented in government – or cohabitation – the situation where the president and prime minister were from opposing parties and where the president's party was not represented in government. Cohabitation is quite rare in our sample: it accounts for only 9.4% of 393 observations.

The third explanatory variable is divided minority government. We hypothesize that divided minority government will be associated with anocracies rather than democracies. We identified periods of divided minority government by consulting the World Bank's Database of Political Institutions (DPI).Footnote ⁴ This dataset has an entry called ‘Majority’. The DPI codebook states that this entry records ‘the fraction of seats held by the government’. It is calculated by dividing the number of government seats by the total number of seats in the main house of the legislature. When the score for ‘Majority’ was below 50% in a given year, we coded the case as a period of minority government. The DPI database only goes back to 1975. This range covers most of our examples. For pre-1975 cases, we use secondary sources to determine whether or not there was minority government. In our sample, minority government occurs in 21.1% of our observations.

There are six control variables: wealth, population, legislative fractionalization, and ethnic fractionalization, as well as dummies for Europe and the post-Cold War era. We follow the literature in taking the natural logarithm of GDP per capita and population as the first two control variables. More wealthy countries are expected to be associated with democracies as are those with smaller populations. We take data for GDP per capita in 1990 Geary-Khamis dollars and population from the Total Economy Database of the University of Gröningen Growth and Development Centre (http://www.ggdc.net/dseries/totecon.shtml). For missing values and pre-1950 figures, we use Angus Maddison's data set of Historical Statistics for the World Economy (http://www.ggdc.net/Maddison/). In effect, this data set is the precursor of the Total Economy Database and there is a very high correlation between the two. For Timor-Leste's population we used the World Bank's World Development Indicators. There is a wide variation on each of both the GDP and population measures. The third control variable is legislative fractionalization. The more fragmented the legislature, the more difficult it is to sustain democracy. Therefore, the greater the fractionalization, the more likely a country is to be an anocracy. For this measure, we use the effective number of political parties. We rely mainly on Michael Gallagher's data set (http://www.tcd.ie/Political_Science/staff/michael_gallagher/ElSystems/Docts/ElectionIndices.pdf). We also use Sarah Birch's (Reference Birch2003) calculations for early Central and East European elections. We calculated any missing values ourselves. The mean figure for ENPP is 3.6 across our 393 observations. Finally, we included Alesina et al.'s Reference Alesina, Devleeschauwer, Easterly, Kurlat and Wacziarg2002 figures for ethnic fractionalization, except for Timor-Leste where we used the figure for linguistic fractionalization. Semi-presidentialism has become much more widespread in the aftermath of the Cold War: only 28% of the observations predate 1990. Also, 56% of the observations are for European countries.

We did not opt for full fixed-effects estimation because of the radically imbalanced nature of our panel. However, we do test for country effects by dropping one by one all of the countries that appear for over ten years in the dataset. Our model is a pooled logit with NeweyWest standard errors to correct for heteroskedasticity and autocorrelation. We present four models in Table 4. The first is our basic model and the other three test its robustness to sampling and measurement issues. It is worth mentioning the performance of the controls included in all equations. As the literature would predict, the wealth variable is always a significant predictor of democracy. In contrast to previous studies, a larger population is associated with democracy. This effect may be due to the fact that micro-states are already excluded from the Polity data set. Unsurprisingly, the effective number of parties is associated with anocracy. However, ethnic fractionalization is associated with democracy, although its coefficient is not always significant. It seems likely that in our analyses much of the effect of ethnic fractionalization is channelled through the party system. The post-Cold War era is always positively and significantly associated with democracy. The results for the European dummy are inconsistent across the equations.

Table 4.

Logit analyses of semi-presidentialism and democratic performance

Notes: Cell entries are coefficients with Newey-West standard errors in parentheses set for a maximum lag of one. *Denotes p < 0.10. **Denotes p < 0.05. ***Denotes p < 0.01. In model 2, cohabitation perfectly predicts the dependent variable and has been dropped from the equation.

We will now discuss our semi-presidential variables equation by equation. As predicted, presidential power is bad for democracy. The presidential power coefficient is in the right direction and is highly significant. Logit coefficients are difficult to interpret directly. However, we can compare the probabilities for different values of the index of presidential power. Taking an average European post-Cold War observation in terms of GDP, population, party system, and ethnic fragmentation, without divided majority government or cohabitation and the maximum level of presidential power, then the probability of democracy is 0.51. Holding all else equal but reducing presidential power to the minimum the probability of democracy doubles to approach certainty −0.98. These figures are clearly substantively, as well as statistically, significant.

By contrast, the cohabitation variable is significantly in the wrong direction. The minority government hypothesis is insignificant and in the wrong direction.

We tested the robustness of these conclusions to the inclusion of countries in our sample by rerunning the model, while excluding one-by-one all 20 country cases that were observed for at least ten years. None of these exclusions made a substantive difference to the models, in terms of significance level and/or sign of coefficient, except for Sri Lanka. In model 2, we can see that the exclusion of Sri Lanka does not modify our conclusion regarding presidential power. Neither does the significantly positive result for divided minority government provide a basis for re-evaluating this variable. Nonetheless, model 2 is quite revealing in respect of cohabitation. Sri Lanka 2003 is the only observation of cohabitation in an anocracy. As we noted already, cohabitation is very rare in semi-presidential regimes. It is even rarer outside Europe: Niger, Sri Lanka, and Mongolia are the only non-European countries to have experienced cohabitation, for one, two, and three years respectively. In short, we suspect our striking finding on cohabitation is more likely a result of its rarity than it operating according to a logic, which is radically different to that outlined in the qualitative literature.

We also investigate the robustness of our models to two measurement issues. Sudden shifts in the Polity rating of countries may affect our conclusions. For example, according to our measure, described above, a democracy that underwent a sudden transition from democracy to autocracy, without going through even one year of anocracy would simply drop out of the dataset and no decline in democratic performance would be registered. Conversely, a country might move directly from anocracy to consolidated democracy without an increase being noted. There are seven cases of such sudden regime change in our sample (Belarus 1994, Austria 1932 and 1945, Peru 1991, Niger 1995, Germany 1932, and Finland 1943). We recoded each of these to record a shift to anocracy or democracy, as appropriate, before the country exits the dataset. As model 3 shows, this did not affect our conclusions.

Finally, we recode Siaroff's scale of presidential power. The literature tends to conceptualize presidential power under semi-presidentialism in terms of three ranked categories: strongly, moderately, and weakly presidential. We coded all countries with scores of over five as strongly presidential, those with scores of four or five as moderately presidential, and all those below four as weakly presidential. The vast majority of observations are strongly presidential, 22% are balanced, while only 10% are weakly presidential. The results in model 4 support our previous conclusions. Both moderately and weakly presidential systems outperform strongly presidential systems and have statistically significant coefficients. The dummy for weak presidents is much bigger than that for moderately powerful presidents. Taking the same values as in our illustration of the presidential power index, the probability of democracy in a weakly presidential system is 0.99. In a balanced system it is 0.86, and in strongly presidential systems it is 0.68.