Scholarly article on topic 'Turnout Under Semipresidentialism: First- and Second-Order Elections to National-Level Institutions'

Turnout Under Semipresidentialism: First- and Second-Order Elections to National-Level Institutions Academic research paper on "Political Science"

0
0
Share paper
Academic journal
Comparative Political Studies
OECD Field of science
Keywords
{""}

Academic research paper on topic "Turnout Under Semipresidentialism: First- and Second-Order Elections to National-Level Institutions"

Comparative Political Studies

http://cps.sagepub.com/

Turnout Under Semipresidentialism: First- and Second-Order Elections to National-Level Institutions

Robert Elgie and Christine Fauvelle-Aymar Comparative Political Studies 2012 45: 1598 DOI: 10.1177/0010414012463903

The online version of this article can be found at: http://cps.sagepub.com/content/45/12/1598

Published by: ®SAGE

http://www.sagepublications.com

Additional services and information for Comparative Political Studies can be found at: Email Alerts: http://cps.sagepub.com/cgi/alerts Subscriptions: http://cps.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav

>> Version of Record - Nov 26, 2012 What is This?

Turnout Under Semipresidentialism: First- and Second-Order Elections to NationalLevel Institutions

Comparative Political Studies 45(12) 1598-1623 © The Author(s) 2012 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0010414012463903 http://cps.sagepub.com

Robert Elgie1

and Christine Fauvelle-Aymar2

Abstract

We test the effect of the importance of elections by focusing on turnout at presidential and legislative elections in countries with semipresidential constitutions. These countries have two potentially first-order elections, but they vary considerably in the powers that are granted to their presidents and prime ministers. We hypothesize that turnout at legislative elections will vary inversely with presidential power and that above a certain degree of presidential power turnout at presidential elections will be higher than turnout at legislative elections. We test these hypotheses on the set of electoral democracies from 1989 to 2010 inclusive. Controlling for standard social, political, and institutional variables, we find strong support for our propositions. The findings show that the importance of the electoral contest is a determining factor for turnout at legislative elections and that the relative importance of presidential and legislative elections can be understood as a continuous scale, and not merely dichotomously.

Keywords

turnout, first- and second-order elections, semipresidentialism, presidential power, legislative elections, presidential elections

'Dublin City University, Dublin, Ireland Université de Tours,Tours, France

Corresponding Author:

Robert Elgie, School of Law and Government, Dublin City University, Dublin 9, Ireland Email: robert.elgie@dcu.ie

Why do people turn out to vote at elections? This is an important issue. Turnout is often taken as an indicator of the democratic health of a polity. Falling turnout is usually interpreted as a sign of a failing democracy (Lutz & Marsh, 2007, p. 549). Turnout can also be seen as a sign of the importance of elections to voters. There is a general assumption that, all else equal, the less important the election the lower the turnout (Reif & Schmitt, 1980). The standard way of identifying the effect of such second-order elections is in a multilevel context. Turnout at elections to national-level institutions is likely to be greater than turnout at elections to subnational or supranational institutions. This proposition has received strong empirical support (Hix & Marsh, 2011; Marsh, 1998). In addition, the U.S. experience clearly shows that there is higher turnout at congressional elections in years when the elections are held concurrently with the presidential election than at midterm congressional elections. Thus, even though Congress is a "first-order" national-level institution, when we control for the effect of synchronized elections voters treat congressional elections as if they were less important, second-order elections.

In this article, we test the effect of the importance of elections to voters by examining turnout in countries with two potentially decisive national-level institutions, but where regular midterm elections are absent. We do so by focusing on differential turnout at national-level elections in countries with semi-presidential constitutions. These countries have in common both a directly elected fixed-term president and a prime minister and cabinet that are collectively responsible to the legislature. Thus, they have two national-level elections. However, these countries vary considerably in the powers that are granted to their presidents and prime ministers. Some countries have strong presidents and weak prime ministers, such as France and Russia. Other countries have weak presidents and strong prime ministers, such as Austria and Bulgaria. In a previous study, Dettrey and Schwindt-Bayer (2009) demonstrated that presidential power had no effect on turnout at presidential elections. In this article, we focus on legislative elections under semipresidentialism. Tavits (2009) argued that voter fatigue under semipresidentialism led to lower turnout at legislative elections in semipresidential countries relative to parliamentary countries, even controlling for presidential power. In this article, we hypothesize that under semipresidentialism turnout at legislative elections varies as a function of presidential power. In countries where presidents are strong, legislative elections become second order. We test this hypothesis on the set of semipresi-dential electoral democracies from 1989 to 2010 inclusive. Controlling for standard social, political, and institutional variables, we find strong support for our proposition. We also show that above a certain degree of presidential power, turnout at presidential elections is higher than at legislative elections.

These results have implications for the study of turnout generally, emphasizing the importance of elections as a determining factor for turnout to national-level institutions. They also open up a new perspective on second-order elections, showing that the relative importance of elections can be understood as a continuous scale, and not merely, as is usual, dichotomously. Finally, the result also has implications for the study of semipresidentialism, revealing that the introduction of this constitutional arrangement can lead to differential turnout at presidential and legislative elections.

The Importance of Elections,

Voter Turnout, and Semipresidentialism

Considerable political and academic attention has been paid to the issue of voter turnout. In the context of consolidated democracies where turnout seems to be declining over time (Blais, 2007, p. 624), decision makers have taken steps to try to reverse this general trend, introducing initiatives such as e-voting and simplifying the procedures for electoral registration. For political actors, the decline in turnout is taken as an indicator of social and political malaise and reforms to increase turnout are part of a plan to reengage citizens with the political process. For their part, academics have identified various social, political, and institutional determinants of turnout. In a meta-analysis of 83 aggregate-level empirical studies, Geys (2006, p. 653) found that "a number of (theoretically important) variables are significantly related to turnout rates." Apart from the puzzles inherent in the study of turnout, so many academics have been motivated to examine this topic because, similar to the desire of political decision makers, there is a desire to understand a phenomenon that is intrinsic to modern political life and the functioning of contemporary democracies.

Just as turnout has been taken as an indicator of the general health of a democracy, it has also been interpreted as a sign of the importance of particular elections to voters. Here the standard distinction is between first- and second-order elections (Reif & Schmitt, 1980). First-order elections are those to institutions that are most salient to citizens, invariably national-level representative institutions such as legislatures. They are elections that "decide who is in power and what policies are pursued" (Schmitt, 2005, p. 651). Second-order elections are those to less salient institutions, such as those at the subnational or supranational level. These elections "are perceived to be less important, because there is less at stake" (Schmitt, 2005, p. 651). Reif and Schmitt (1980) hypothesized that second-order elections would have various consequences relative to their first-order counterparts. One such

consequence was that the level of participation would be lower at elections to second-order institutions. This proposition has now been tested many times and, in general, has been found to be robust. For example, Hix (1998) confirms that turnout at elections to the European Parliament (EP) is lower than it is for elections to national parliaments. Koepke and Ringe's (2006) study of EP elections found less support for some of the other consequences of second-order elections but confirmed the proposition in relation to turnout. Lijphart (1997, p. 6) cites evidence to suggest that turnout is lower at local elections than at national elections. Equally, Rallings and Thrasher (2007) provide figures demonstrating that there is a large "turnout gap" between local and national elections in the United Kingdom with, as expected, turnout being much lower in the former.

As the above examples indicate, the standard application of the second-order model is a multilevel comparison of voter turnout. However, this model can also be applied to presidential systems where there are two decisive national-level elections. For example, in the United States the presidency and Congress are both powerful institutions that help to determine what policies are pursued. That is to say, they are both first-order institutions. Here, though, it is commonly observed that there is a higher level of turnout at congressional elections in years when the elections are held concurrently with the presidential election than at midterm congressional elections (Blais, 2007, p. 626; Kuenzi & Lambright, 2007). More generally, a study of congressional elections in Latin America, Fornos, Power, and Garand (2004) found that turnout was about 5% higher when legislative elections were held concurrently with presidential elections. Thus, even though the standard application of the second-order model is based on a comparison of turnout at elections to nationallevel institutions with elections to subnational or supranational institutions, this model may also be applied to a comparison of turnout at elections to solely national-level institutions. Voters consider elections to some nationallevel institutions to be more important than to others.

To test the relative importance of national-level elections as a determining factor in voter turnout, we focus on countries with semipresidential constitutions. Semipresidentialism is the situation where a constitution makes provision for both a directly elected president and a prime minister and cabinet that are collectively responsible to the legislature (Elgie, 1999). Even though, by definition, semipresidential countries share basic constitutional features, the key observation about semipresidentialism is that it is not a unimodal category. Semipresidential countries vary considerably in the powers that their constitutions grant to the president, the prime minister, and the legislature. This variation can be captured in different ways. Shugart and Carey (1992)

distinguished between president-parliamentary and premier-presidential forms of semipresidentialism. In the former, the government is responsible to both the president and the legislature, whereas in the latter it is responsible solely to the legislature. Metcalf (2000) reworked Shugart and Carey's method of measuring presidential power indicators to allow cross-national measures of presidential power to be constructed. Siaroff (2003) undertook a similar exercise, providing scores for a wide range of presidential, parliamentary, and semipresidential countries. Costa Lobo and Amorim Neto (2010), Krouwel (2003), and Van Cranenburgh (2008) have all conducted equivalent exercises. Whichever method is adopted, the "critical characteristic of semipresidential constitutions is that they do not locate the power to control the government as unambiguously with either president or assembly as presidential or parliamentary constitutions do" (Schleiter & Morgan-Jones, 2010, p. 1418).

In this context, semipresidential countries provide a good test of the effect of the perceived importance of national-level elections on turnout because, as in presidential systems, there are elections to two potentially decisive firstorder institutions, the presidency and the legislature. We test three hypotheses. The first two relate to presidential power and turnout at presidential and legislative elections separately. The third relates to presidential power and turnout at presidential elections relative to legislative elections.

The first hypothesis relates to the relationship between presidential power and turnout at presidential elections. In a recent article, Dettrey and Schwindt-Bayer (2009) demonstrated that variation in presidential power did not affect the level of turnout at presidential elections. Their study pooled presidential and semipresidential democracies. If they are correct, then we would not expect presidential power to be associated with turnout at presidential elections even if we restrict the sample solely to semipresidential countries.

The second hypothesis relates to the relationship between presidential power and turnout at legislative elections. Under semipresidentialism, we expect variation in presidential power to create legislative elections of differing importance. When the president is powerful, presidential elections are the principal electoral contest. Presidential candidates present competing programs that cover all aspects of foreign and domestic policy. Voters and parties expect the winning candidate to implement their program. To do so, the president needs the support of the legislature. Voters are aware of this requirement. As a result, we would expect them to return a legislature that is favorable to the president. With the support of the legislature, the president can appoint a loyal prime minister and a government tasked with implementing the election program. So with the presidential election being the primary electoral contest, we would expect parties to devote fewer resources to the legislative election

and/or voters to be less interested in it, thus decreasing turnout at legislative elections. By the same token, when presidents are weak legislative elections are the principal electoral contest. Here, the prime minister is the main political leader. The prime minister takes power by virtue of the support of the legislature. With the legislative election being the predominant electoral contest, parties and/or voters devote more attention to it, thus increasing turnout at legislative elections. Thus, we hypothesize that, all else equal, turnout is likely to be lower at legislative elections in countries with strong presidencies than in those with weak presidencies. To put it another way, in countries where presidents are strong, legislative elections become second order.

The third hypothesis relates to presidential power and turnout at presidential elections relative to legislative elections. Given we expect presidential power to affect the level of turnout at legislative elections, we might also expect this relationship to affect the level of turnout at presidential and legislative elections relative to each other. In particular, if turnout at legislative elections decreases with presidential power but turnout at presidential elections stays the same, then we would expect turnout at presidential elections to be greater than turnout at legislative elections after a certain level of presidential power. As research on turnout generally focuses on one or other set of elections in isolation, the issue of the relative value of turnout has never been examined. Thus, whereas the existing literature suggests that a particular type of election is either first-order or second-order, we expect legislative elections to become second-order relative to presidential elections as presidential power increases. This suggests that we should understand the value of elections on the basis of a continuous scale rather than dichotomously.

In this article, we bring together two previously separate literatures. Although there is an abundant literature on turnout generally and the distinction between turnout at first- and second-order elections in particular (Clark & Rohrschneider, 2009; Marsh, 1998; Schmitt, 2005), this work has not been applied systematically to the relative importance of national-level elections. Also, although certain studies have incorporated semipresidential countries in more general studies of turnout (Dettrey & Schwindt-Bayer, 2009; Pacek, Pop-Elches, & Tucker, 2009; Tavits, 2009), to our knowledge there is no study that applies the logic of first- and second-order elections to the context of legislative elections under semipresidentialism. Equally, although there is now a growing literature on semipresidentialism, this work has tended to focus on the effect of this regime type on democratic stability (Elgie, 2011; Elgie & Schleiter, 2011; Moestrup, 1999; Skach, 2005) and on relations among the president, the prime minister, and the legislature (Protsyk, 2006; Schleiter & Morgan-Jones, 2010; Sedelius & Ekman, 2010). In this article,

we focus on the differential effect on turnout of presidential power under semipresidentialism. In a recent article, Tavits (2009, p. 47) found that turnout at legislative elections in semipresidential countries was 7% lower than at the same elections in countries with an indirectly elected president. She also found that the effect of direct versus indirect elections survives even when presidential powers are controlled for (p. 48). She attributed these findings to the effect of voter fatigue in countries with both presidential and legislative elections. In this article, we argue that under semipresidentialism presidential power can have the effect of generating first- and second-order elections. Such a study is important because it speaks to the debate about the quality of contemporary democracies. If we can provide a richer explanation of voter turnout, then we are better placed to address the normative questions that are raised about the effects of apparently declining turnout over time. In the next section, we outline the research design that allows us to test our propositions about the effect on turnout of the perceived importance of legislative elections under semipresidentialism.

Case Selection, Model Specification,Variables

To test for the effect of the relative importance of national-level elections on turnout, we compiled a new data set that includes data for all democratic semipresidential countries. We identify the list of countries with semipresidential constitutions from Elgie (2011, p. 24). We identify democracies by referring to Freedom House's list of Electoral Democracies.1 This list is available from 1989 to 2010 inclusive. Therefore, our sample includes all elections in semipresidential countries that Freedom House classes as electoral democracies between these dates.2 For countries that were already classed as electoral democracies in 1989 but where no election was held in that year, we include turnout for the most recent election prior to that time. For example, the observations for Finland begin with the 1988 presidential election. If a country is first classed as an electoral democracy at some point after 1989, we include only the elections from this point onward. So we include Mali from 1992. Similarly, if a country loses the status of an electoral democracy, then we cease to include it in the data set.3

Our dependent variable is voter turnout. If there is a two-round electoral system, such as in the vast majority of our sample of countries for the presidential elections, we record the turnout figure at the first round of the election. This is because competition at the second round is usually restricted to a smaller number of candidates, which may artificially increase or decrease the turnout figure depending on the context of the election. In the literature, there are two

general ways of measuring turnout. It can be calculated either as a percentage of the registered voters in a country or as a percentage of the voting-age population. As Geys (2006) notes, there is no decisive argument in favor of one or the other of these alternatives. The choice between the two is mainly a function of data availability. In our case, we measure turnout as a percentage of registered voters. This is because most of our turnout data were gathered from national electoral commission websites that do not have any statistics on voting-age population. We preferred, where possible, to use official sources for turnout rather than rely on the International Institute for Democracy and Electoral Assistance (IDEA) website, even though this source provides data for both indicators of turnout and even though it is commonly used in turnout studies.4 We chose to do so because of the unreliability of the information recorded in the IDEA website. For example, the IDEA website records turnout at the 2010 Croatian presidential election as 50.13% of registered voters, whereas we record a figure of 43.96%. This is because the IDEA database records the official second-round figure rather than the first-round figure.5 There is also a discrepancy between the IDEA figure for the 2000 Croatian presidential election (60.88%) and the official first-round turnout figure (62.98%).6 Where information is not available on national electoral commission websites, we rely on country-specific or region-specific sources that specify first-round turnout figures.7 Overall, we were able to gain information for 254 elections held in 39 countries.8 The data set is unbalanced because all the countries are not included for all the period 1989-2010 inclusive and because some countries have held more elections than others. There are 137 parliamentary elections and 117 presidential elections. The average turnout rate of our sample is 66.50% (with no significant difference between average legislative and presidential turnout rates), and there is a high standard error (14.26).

In terms of explanatory variables, we include a dummy variable (Legis) that is equal to 1 for legislative elections and 0 for presidential elections.

To identify the likely importance of an election to voters, we use the Siaroff (2003) index. This index of presidential power is based on nine indicators, including whether the president chairs cabinet meetings, whether or not the president can veto legislation passed by parliament, whether or not the president can invoke emergency powers and so on. Siaroff records a score of 1 if the president in a particular country enjoys the power in question and 0 otherwise. The individual scores are summed. Thus, countries emerge with a presidential power score somewhere in a range from 0 to 9. If a country was not included in the Siaroff study, then we calculated the value of the Siaroff index. Given our sample includes only countries with a popularly elected president and given one of the Siaroff indicators is whether or not the president is directly elected, the lowest possible Siaroff value in our

Table 1. Value of the Siaroff Index in the Sample of Countries

Value of the Siaroff index 1 2 3 4 5 6 7 8

Number of countries 3 4 4 4 5 11 12 2

sample is 1. This score is recorded for Austria, Iceland, and Slovenia. The highest Siaroff value in our sample is 8. This score is recorded for Mozambique and Sao Tomé and Príncipe. Table 1 below indicates the number of countries at each level of the Siaroff index.9

We include an interaction variable (LegSia) that multiplies the legislative election dummy variable by the Siaroff index. This allows us to test our main theoretical hypotheses. More precisely, we estimate the following equation,

Turnout = a + P Legis + y Siaroff + 5 LegSia + ^ X + s

where Legis is the legislative dummy variable, Siaroff is the value of the Siaroff index, LegSia is the legislative dummy multiplied by the Siaroff index, and X is a vector of control variables.

Using this estimation, we can directly assess the relationship between presidential power and turnout at presidential elections by looking at the coefficient associated with the Siaroff variable (y). As explained before, we do not expect the estimated coefficient to be significant.

As we use an interaction term, it is not possible to assess directly the relation between turnout at legislative elections and presidential power. Instead, this relationship is measured by the sum of two coefficients, the estimated coefficients associated with Siaroff (y) and LegSia (<5).10 We expect this sum to be negative as turnout at legislative elections is expected to decrease with presidential power.

This equation also allows us to compare the level of turnout at presidential elections relative to legislative elections. To do so, we calculate the sum (p+ 5 Siaroff) for each value of the Siaroff index. A negative and significant sum would suggest that turnout is higher at presidential elections than legislative elections.11

In his meta-analysis of the determinants of turnout, Geys (2006) distinguishes among socioeconomic, institutional, and political factors. Here, we follow his categorization and include each of the three factors as control variables. A detailed presentation of the theoretical justifications for these determinants is beyond the scope of this article. They are introduced merely to avoid omitted-variable bias. We do not claim to be providing any new insights in relation to these variables. We merely record the expected effect of each variable on turnout.

The principal socioeconomic factor identified by Geys (2006) is the size of the country's population. We include a variable for the natural log of a country's population (lnPop).12 The main argument advanced to justify the introduction of this variable is related to the Downs's (1957) model of rational voting. The probability of casting a decisive ballot is inversely related to the size of the constituency. Thus, we would expect a negative relationship between turnout and the size of the population.

The sample of countries included in our data set also requires the introduction of certain socioeconomic control variables that are not generally identified in Geys (2006) metastudy. Our sample contains countries with very different levels of economic development. Therefore, we include a variable that records the natural log of the level of GDP per capita (lnGDPc).13 According to Blais, Massicotte, and Dobrzynska (2003), turnout is expected to be higher in more developed countries. This is because development is associated with a higher level of education and a broader range of social concerns, thus generating more competitive elections. Moreover, even though all of the countries in our sample have passed a certain threshold of democracy, there remains considerable variation in the overall level of democracy. Therefore, we also include the level of democracy in a country as a control variable. There is no particular theoretical justification concerning the direction of the relationship (if any) between turnout and level of democracy. However, as democracy is higher in more developed countries, one might expect the level of democracy to have a positive impact on turnout. Our measure of democracy (Demo) is a dummy variable that distinguishes between the electoral democracies that are classed as free by Freedom House and those that are classed as only partly free.14 The correlation between democracy and development means that we do not introduce both of these control variables simultaneously.

We include two institutional control variables. The first captures the nature of the electoral system, which is probably one of the most important factors influencing turnout. The lower level of disproportionality between votes and seats in proportional representation (PR) electoral systems constitutes a clear incentive for voters to turn out (Geys, 2006). Thus, we would expect turnout to be higher, ceteris paribus, in legislative elections with PR.

The second institutional variable captures whether or not there are concurrent presidential and legislative elections. Concurrent elections are likely to boost turnout (Geys, 2006). In our sample, only a small number of countries hold or have held concurrent elections. They have been held regularly in Mozambique, Namibia, and Peru. In Romania, concurrent elections were held until 2004 inclusive. In other countries, the occurrence of synchronized presidential and legislative elections varies depending on the particular electoral

calendar and political circumstances. There are 41 concurrent elections in our data set. To control for this particular institutional context, a dummy variable (Synch) is introduced in the estimations. We expect its estimated coefficient to be positive.

Another institutional variable that is often introduced in turnout analysis is compulsory voting. However, there is only one country in our sample, Peru, where voting is compulsory. The introduction of this variable would be equivalent to introducing a dummy variable for Peru and, therefore, is not meaningful for the purposes of this study.

We also include two political control variables. The first captures the level of electoral competition. When the outcome of an election is expected to be close, then the level of turnout is likely to be higher. This factor has long been recognized as a decisive factor that determines turnout. The theoretical explanations refer either to Downs's (1957) model of rational voting or to the extra resources that political parties deploy in the case of close elections (Cox & Munger, 1989). Consistent with the standard operationalization of this variable, we include an ex-post (based on electoral results) measure of competitiveness. The Closeness variable is defined as the difference between the score of the two first lists (or candidates) multiplied by -1 (to generate an indicator whose value increases with the degree of closeness). The estimated coefficient associated with this variable is expected to be positive.

The second political variable captures the level of party competition. As Geys (2006) clearly explains, various theoretical arguments have been developed concerning the relationship between turnout and party fragmentation, and there is no standard expectation. Some expect a negative relationship (turnout decreasing with the number of parties), whereas others expect a positive relationship (turnout increases with the number of parties). The results of empirical studies present conflicting results (Geys, 2006). Thus, we have no explicit expectation about the effect of this variable. To capture the level of party competition, we include the effective number of parliamentary parties (ENPP). We record the figures calculated by Michael Gallagher.15 If the data were not available in the Gallagher database, we calculated the value of the ENPP index using electoral results.

Finally, we introduce two additional control variables that were not identified in Geys's (2006) analysis. First, as our study concerns semipresidential countries, it seems logical to take account of cohabitation, which is a specific situation that can occur only in this type of regime. Cohabitation is defined as the situation where the president and the prime minister are from opposing political parties or coalitions and where the president's party is not represented in the government (Samuels & Shugart, 2010). Thus, periods of

cohabitation record the situation where a partisan president is isolated within the executive. Cohabitation usually occurs when the president's party loses the legislative election and is forced out of government, or when a president is elected from a party that is not in government and the legislature is not dissolved giving no chance for the president's party to be returned to government. Our database contains 42 cases of cohabitation. There is no literature about the effect of cohabitation on turnout.16 However, we might have reason to believe that when an election takes place during a period of cohabitation, the stakes are likely to be high since these elections will decide whether or not cohabitation ends. If so, then we would expect turnout to be higher at elections that occur during a period of cohabitation. Thus, the coefficient associated with the Cohab dummy variable is expected to be positive.

The final control variable captures the fact that our data set includes founding elections. Founding elections are the first elections held in states that have democratized for the first time, such as the presidential election held in Congo-Brazzaville in 1992 or the legislative election held in Namibia in 1989, or in newly created states such as the presidential election in East Timor in 2002. It is generally found that turnout is higher at founding elections (Fornos et al., 2004; Kostadinova & Power, 2007). Thus, we expect the coefficient associated with the Founding dummy variable to be positive.

The case selection generates a panel data set, containing observations about different countries and different elections over time. Such a data set cannot be properly estimated using ordinary least squares (OLS) regression since it will violate the assumption of homoscedasticity and uncorrelated error terms (Wooldridge, 2010). It is surprising that, despite the inadequacy of the OLS method, most of the comparative literature on turnout has used this method of estimation. For instance, of the 13 empirical studies identified by Geys (2006) that are based on a panel data set, only 2 use estimation techniques that are appropriate for panel data.17 Only recently have techniques adapted to a panel data set started to be used in this literature (e.g., Endersby & Krieckhaus, 2008; Fornos et al., 2004; Tavits, 2009) and that the issue of economic robustness has been discussed.

With panel data, two main estimation techniques can be employed depending on how the cross-sectional heterogeneity is modeled, the fixed effects model (or least squares dummy variable) or the random effects model (Wooldridge, 2010). The choice between these two estimation techniques is not always straightforward. For the purposes of this article, there are two reasons for adopting the random effects method of estimation. First, the Hausman test indicates that the random effects model is appropriate.18 The second argument is related to the characteristics of the regressors included in

the estimation. When there is an explanatory variable that is time invariant within groups, the fixed effect estimators cannot be computed since the timeinvariant regressors would be perfectly collinear with the fixed effect for that country (Wooldridge, 2010). In our case, using the fixed effect model would prevent us from introducing certain time-invariant variables, such as the characteristics of the electoral system. Therefore, we estimate the model using the random effects technique.19

Another issue with panel data sets is the possibility of heterogeneity across time. "The two-way error components model allows for specific time effects accounting for unobserved factors assumed to affect all individuals in a similar way at a given point in time" (Baltagi, Matyas, & Sevestre, 2008, p. 64). Our data set includes concurrent presidential and legislative elections, namely, two country observations that occur at exactly the same date. This prevents us from using a two-way random effects model. An alternative way of accounting for the possibility of time effects is to introduce a trend variable. In our case, this is well suited to the data set because of the temporal evolution of the turnout variable. There has been a general decrease in turnout over time (Blais, 2000; Gray & Caul, 2000). For this reason, a trend variable is particularly appropriate in a comparative study of turnout.20 It is surprising, though, that to our knowledge it has never been introduced in any such study. Thus, we employ a random effects model and include a trend variable.21

Results and Robustness Tests

Table 2 presents the results of our estimation. They show that the estimated coefficient associated with the Siaroff variable is negative but not significant. In other words, this result confirms the Dettrey and Schwindt-Bayer (2009) finding. Turnout at presidential elections does not depend on the level of presidential power. Voters seem to be just as inclined to turn out and vote at the election for a less powerful president as for a more powerful president.

The results also confirm our main theoretical hypothesis concerning turnout at legislative elections. The sum y + 3 = -0.64 -1.34= -1.98 is negative and highly significant (Model 1).22 This confirms that the higher the presidential power, the lower the turnout at legislative elections.

As explained above, the estimated equation also allows us to compare the level of turnout at both types of election. Table 3 gives the value of the sum (P + 3 Siaroff) for each value of the Siaroff index, the value of its standard error, and the Student t. For instance, in countries with a Siaroff value of 4, turnout at legislative elections is lower than turnout at presidential elections. The estimated coefficient is negative and highly significant. In fact, the

Table 2. The Importance of Presidential and Legislative Elections Under Semipresidentialism

(1) (2) (3)

InPop 0.208 (0.15)

InGDPc 0.0391 (0.03)

Synch 10.42*** 9.765*** 9.389***

(4.05) (3.92) (3.61)

PR 5.329** 5.524** 5.830**

(2.46) (2.56) (2.53)

Closeness 0.136*** 0.126*** 0.125***

(3.55) (3.40) (3.15)

ENPP -1.066** -1.003* -1.074*

(-1.99) (-1.93) (-1.95)

Cohab -0.0603 (-0.04)

Founding 1.826 (0.76)

Trend -0.590*** -0.603*** -0.596***

(-5.58) (-6.20) (-5.69)

Legis 1.866 1.894 1.660

(0.58) (0.59) (0.48)

Siaroff -0.640 -0.402 -0.404

(-0.74) (-0.54) (-0.56)

LegSia -1.338*** -1.369*** -1.360**

(-2.61) (-2.69) (-2.46)

Constant 78.80*** 81.93*** 82.07***

(3.42) (15.63) (15.72)

N 252 255 255

R2 between .0995 .0661

R2 within .293 .288

R2 overall .183 .173

x2 89.30 86.05 73.86

(I) = random effect model; (2) = robust standard errors adjusted for clustering; (3) = generalized estimating estimation. T statistics in parentheses. *p < .I. **p < .05. ***p < .01.

results in Table 3 show that turnout at legislative elections is always lower than turnout at presidential elections when the Siaroff value is higher than 3. By contrast, in countries with a weak president (a Siaroff value of 3 or lower),

Table 3. Turnout Differential and Presidential Power

Siaroff value Coeff. SE Student t

1 0.53 2.78 0.19

2 -0.81 2.40 -0.34

3 -2.15 2.09 -1.03

4 -3.49 1.86 -1.88

5 -4.82 1.75 -2.75

6 -6.16 1.80 -3.43

7 -7.50 1.97 -3.80

8 -8.84 2.26 -3.92

the estimated coefficient is still negative but not significant. In other words, in these latter countries there is no difference between turnout at presidential and legislative elections. Thus, turnout at legislative elections relative to presidential elections varies as a function of presidential power.

The results for the control variables are globally consistent with our expectations, except for the socioeconomic determinants of turnout. As regards the latter, we expected to find a negative relationship between turnout and population and a positive relationship between turnout and economic development. However, our study did not demonstrate any significant relationship between turnout and either of these two variables. To check the robustness of our result, we ran an OLS estimation. This is because Blais's study on economic development and many of the studies that include a population variable are based on the results of an OLS model. When we did so, we found highly significant negative coefficients for both variables. However, as noted above, it is inappropriate to use OLS to estimate a panel data set. Therefore, although we do not return the standard results for the socioeconomic determinants of turnout in our estimations, we conclude that this is because of the estimation techniques used in previous studies.23

The results concerning the institutional determinants of turnout are as expected. Using an electoral system based on PR boosts turnout. Equally, concurrent elections also have a positive and highly significant impact on turnout. As regards political factors, the coefficient associated with the closeness variable is positive (turnout increases with the closeness of the race) and highly significant. As noted previously, the relationship between turnout and the number of parliamentary parties was theoretically undetermined. Our results suggest the existence of a negative relationship. The higher the number of party (or the degree of fractionalization), the lower the rate of turnout.

The two last control variables introduced in the estimations were not significant. When elections occur during a period of cohabitation, there is no impact on turnout. Equally, founding elections have no impact. This last result can probably be explained by the fact that a trend variable is introduced in the estimation and thus trumps the effect of founding elections. As can be seen, the coefficient associated with this trend variable is negative and highly significant. Turnout has declined over the period. The founding election variable also captures this situation. It is certainly the case that many studies have shown that turnout is higher at the first democratic elections than at the following elections. When the trend variable is excluded from the estimation, the coefficient associated with the Founding variable becomes positive and highly significant whereas the other results remain unchanged.

To assess the robustness of our estimations, we ran various tests (see Table 2). Model 2 presents a reduced model that includes only those control variables that were found to be significant. This estimation also presents robust standard errors. In terms of additional estimation techniques, whereas a number of recent studies have used methods that are appropriate for panel data, there is no agreement as to which models should be employed. Endersby and Krieckhaus (2008) and Dettrey and Schwindt-Bayer (2009) estimate an OLS model with robust standard errors adjusted for country-clustered standard errors. However, the asymptotic properties of the estimator are not verified when the number of clusters is small and the estimated coefficient can then be biased and/or inefficient (Wooldridge, 2003). The problem is that in most studies of turnout the number of clusters (i.e., countries) is indeed quite small. Nevertheless, we used this method of estimation, and it did not change our results (Model 2). The exclusion of the nonsignificant variables in Model 1 did not lead to any change in the other estimated coefficients. We also estimated the model on a sample excluding all the synchronized elections and the results were not changed at all. In their study, Fornos et al. (2004) used a generalized estimating equation (Liang & Zeger, 1986). This estimation technique aims to take grouped data into account even in the case of a small number of clusters (and thus does not suffer from the problem outlined in the previous technique). Again, when we use this method, we return equivalent results to those presented before (Model 3). Overall, we are confident that our results are not sensitive to the particular estimation technique that is employed.

Finally, we checked for the presence of outliers and influential observations. When we compute Cook's distance, which takes into account the leverage and/or the large residual of a data point, we identify 23 observations whose Cook's distance is above the conventional cutoff value for this test (4/n). Running the estimation without these observations does not change the

results. The exclusion of the observations with an absolute value of the stu-dentized residual above 2 also does not change the results. Similarly, the exclusion of the observations with a leverage (measured by the diagonal elements of the hat matrix) above the usual cutoff value does change one aspect of the results.24 Of the six observations with a leverage above the cutoff value, two stand out. They are the data points for the legislative election in Russia in 1993 and Poland in 1991. These two observations have the highest ENPP value in the sample. The exclusion of these two observations leads to the estimated coefficient associated to the ENPP variable becoming insignificant. This result raises some doubts about the previously reported ENPP result, suggesting that there might not be a relationship between turnout and the number of political parties. However, it should be stressed that we do not aim to make a contribution to the literature on turnout in this regard and that the exclusion of these two observations does not change the results regarding our main explanatory variable.

Discussion

By definition, semipresidential countries share certain institutional characteristics. In practice, though, semipresidential countries vary considerably in the prominence they give to these institutions. In some the president is little more than a figurehead and the prime minister is the focus of decision making. In others, the president is the foremost political figure and the prime minister is subordinate. In this article, we have demonstrated that semipresi-dential countries can vary in another way as well. Even though all semi-presidential countries hold elections to two national-level institutions, they can differ in terms of the importance that is placed on these elections. In semipresidential countries the importance of legislative elections varies inversely with the level of presidential power. Currently, countries as diverse as the Czech Republic, Italy, Moldova, and Nepal are considering introducing the direct election of the president and adopting a semipresidential constitution. This article has shown that, depending on the power of the president in these countries, turnout at legislative elections may vary as a consequence of any such reform. And yet, when debating the choice of semipresidential-ism, this issue is never mentioned. We are not suggesting that the likelihood of differential turnout is an argument either for or against semipresidential-ism. We are saying, though, that for a more informed constitutional choice to be made, citizens in these countries and others citizens should be made aware that differential turnout is a potential consequence of the introduction of a semipresidential system.

The results also speak to the classical distinction between first- and second-order elections. This distinction implies a binary categorization with some elections being important and other elections being unimportant. The results here clearly show that the importance of elections has to be considered along a continuum. In countries with a low level of presidential power, voters are faced with two first-order elections. However, as presidential power increases, legislative elections lose their first-order status little by little and end up as second-order elections. Previously, Van der Eijk, Franklin, and Marsh (1996) advanced a hypothesis of this sort. In the conclusion to their study, they noted, "Some second-order elections are evidently more second-order in character than others, which leads to the idea that some first-order elections may also be more second-order in character than others" (p. 162). To our knowledge, this hypothesis has never been empirically tested. Our results demonstrate the validity of their intuition.

We have demonstrated that in semipresidential countries with powerful presidents, turnout is higher at presidential elections than legislative elections. However, we are not in a position to conclude whether this is because instrumental voters perceive the presidential election to be more important than the legislative election and, therefore, are more likely to turn out to vote at the former relative to the latter, or because parties consider the presidential election to be more important than the legislative election and, consequently, deploy more campaign and communication resources to that election, thus generating more voter interest. There is anecdotal evidence to suggest that voters respond to the value of the election rather than the mobilization of resources. For example, the Romanian president has a Siaroff value of 5, implying that legislative elections are second order. Indeed, at the 2008 legislative election turnout was 39.2%, whereas at the 2009 presidential election it was 54.4%. However, official figures show that the PDL party spent 14.9 million lei on President Basescu's reelection campaign and 25.9 million lei on the party's legislative election campaign.25 Therefore, the level of turnout was higher at the presidential election, even though parties devoted many more resources to the legislative election. That said, campaign expenditure limits are more permissive at legislative elections. By itself, this will have artificially inflated the level of spending there relative to presidential elections. Clearly, more research is now needed to determine the extent to which the level of turnout varies as a function of whether voters perceive particular elections to be more important or whether they are encouraged to participate more by the activity of political parties.

This issue is closely related to an ongoing debate in the general literature on turnout. The standard wisdom is that turnout is higher when election contests

are closer. Our study has confirmed this result. Why, though, is turnout higher when elections are closer? Is it because closeness increases the value of turnout to rational voters (Matsusaka & Palda, 1993), or is it because parties mobilize more resources (Cox & Munger, 1989)? One of the problems with providing an answer to this question is that closeness may be endogenous to party mobilization. Closeness may make parties mobilize more, leading to closer elections and more party mobilization, and so on. By contrast, the importance of elections does not depend on the degree of party mobilization. Their importance is determined by the power of the institution that is being elected. Therefore, to test whether turnout is higher at first-order elections because instrumental voters consider them to be more important or because parties mobilize more resources, we simply need to measure the extent of party mobilization. We can do so directly through the relative levels of campaign expenditure, the amount of advertising, and so on. Therefore, this article not only establishes a research agenda but does so in a way that is susceptible to robust empirical testing.

Conclusion

This article has combined two previously separate literatures, the literature on the relative importance of elections and the literature on the variation in presidential power under semipresidentialism. To date, the literature on the relative importance of elections has focused overwhelmingly on the differential importance of elections to national-level institutions compared with elections to subnational or supranational institutions (Reif & Schmitt, 1980). This article has extended the logic of first- and second-order elections to the national level. We have demonstrated that there is variation in the importance of elections to national-level institutions as well. Equally, until now the literature on the variation in presidential power under semipresidentialism has tended to focus on the effect of such variation on democratic performance (Elgie, 2011; Elgie & Schleiter, 2011) or levels of conflict within the executive (Protsyk, 2006). Tavits (2009) has argued that voter fatigue accounts for the lower level of turnout at legislative elections under semipresidentialism relative to parliamentarism even controlling for presidential power. In this article we have demonstrated that variation in presidential power under semi-presidentialism means that even though these regimes have two national elections they are not necessarily of the same importance. We have clearly demonstrated that the importance of legislative elections and, therefore, the rate of turnout depend on the power of the president.

This article has also contributed to our general understanding of first- and second-order elections. In the literature, this distinction is understood to be

binary. However, this article has shown that the distinction should be understood as a continuum ranging from less important to more important elections. This result could be applied to the comparative study of subnational elections. Are certain subnational elections more important than others? In addition, although we have not aimed to demonstrate whether voters intrinsically perceive certain elections to be more important than others and, therefore, turn out to vote in greater or lesser numbers, or whether parties perceive them to be more important and devote differential resources to them, thus generating more or less interest among voters accordingly, we have demonstrated that the validity of either proposition could be tested without any inherent endogeneity problem.

Overall, this article has shed new light on the relative importance of elections under semipresidentialism. Given that so many countries have now chosen this form of government and that others are considering doing so, the findings have important implications for constitutional choice. Citizens need to be aware that depending on presidential power the choice of semipresiden-tialism is likely to lead to differential turnout at national-level elections. Knowledge of such an effect may have a bearing on whether certain countries wish to remain semipresidential and on whether other countries wish to choose this type of constitutional arrangement.

Acknowledgments

We would like to thank the anonymous reviewers for their very helpful comments as well as Shane Martin for his feedback. As ever, the authors are solely responsible for the final article and its contents.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

1. Retrieved from http://www.freedomhouse.org/template.cfm?page=439 (accessed June 17, 2011).

2. We were unable to include Kyrgyzstan (1995-2000) in the sample. Some elections held in the other semipresidential countries are not included in the sample because of missing data for the dependent and/or the independent variables.

3. We do not include the election in the year in which the country loses its status as an electoral democracy.

4. Retrieved from http://www.idea.int/vt/index.cfm (accessed June 21, 2011).

5. Retrieved from http://www.idea.int/vt/country_view.cfm?id=98 (accessed June 21, 2010). The official first- and second-round turnout figures from the Croatian State Electoral Commission are available at http://www.izbori.hr/izbori/dipFiles. nsf/0/9EAC74D78305D7CBC12578170048CC68/$File/konacno_izvjesce_ OESS-a.pdf (accessed June 21, 2010).

6. The official first-round turnout figures for the 2000 presidential election are available at http://www.izbori.hr/2000Pred/Pred1Krug.htm (accessed June 21, 2011).

7. These sources included the African Elections Database (http://africanelections. tripod.com/), the Electoral Institute for the Sustainability of Democracy in Africa website (http://www.eisa.org.za/WEP/ea.htm), and Wikipedia country election pages (http://en.wikipedia.org/wiki/Elections_by_country).

8. They are Armenia (1999), Austria (1986-), Bulgaria (1992-), Cape Verde (1991-), Central African Republic (1993-1999, 2005), Congo-Brazzaville (1992-1993), Croatia (1992-), East Timor (2002-), Finland (1988-), France (1988-), Georgia (2004), Guinea-Bissau (19994-1999, 2005-2009), Haiti (1995, 2006), Iceland (1988-), Ireland (1989-), Lithuania (1992-), Macedonia (1994-), Madagascar (1992-2007), Mali (1992-), Mauritania (2007), Moldova (1994-1998), Mongolia (1993-), Montenegro (2006-), Mozambique (1994-2004), Namibia (1989-), Niger (1993-1995, 1999-2004), Peru (1990, 2001-), Poland (1991-), Portugal (1987-), Romania (1992-), Russia (1993-2003), Sao Tomé and Principe (1991-), Senegal (2001-), Serbia (2007-), Slovakia (1999-), Slovenia (1992-), Sri Lanka (19892010), Taiwan (1998-), and Ukraine (1994-).

9. The number of observations in Table 1 (45) is greater than the number of countries in our sample (39) because in some countries the value of the Siaroff index changed over the period.

10. For a detailed presentation of empirical analysis using interactive variable, see Brambor, Clark, and Golder (2005) and Kam and Franzese (2007). The standard error of this sum of estimated coefficients is calculated using the following formula: ô = Jvar (y) + Z2 var(S) + 2Zcoviy S) where Z = Legis.

11. The standard error is calculated as explained in the previous note, where this time Z = Siaroff.

12. The source of the population figures is the World Bank's World Development Indicators. Retrieved from http://data.worldbank.org/data-catalog/world-development-indicators (accessed June 21, 2011).

13. The source for the GDP data is the ERS International Macroeconomic Data Set, which provides annual figures for real 2005 GDP per capita in U.S. dollars. Retrieved from http://www.ers.usda.gov/Data/Macroeconomics/ (accessed June 21, 2011). The GDP data are missing for East Timor. Thus, this country is not included in the regressions presented in Table 2. However, it is included in the estimations in Table 3.

14. The classifications for our period are available at http://www.freedomhouse.org/ template.cfm?page=439 (accessed June 21, 2011).

15. Retrieved from http://www.tcd.ie/Political_Science/staff/michael_gallagher/ ElSystems/index.php (accessed June 28, 2011).

16. Most studies of cohabitation focus on the degree of conflict between the president and the prime minister that it generates. See, for example, Pierce (1991).

17. Four of these articles use a regression based on average turnout over the period, and nine use ordinary least squares (OLS) to estimate their panel data set.

18. More precisely, the null hypothesis that the two estimates should not differ systematically is not rejected at the 10% level.

19. Other panel data methods of estimation have been used. For various reasons, they are less adapted, as we explain later in the section on robustness.

20. Another solution would have been to include in the estimation the lagged dependent variable. However, as our data set covers two different types of election, it raises the problem to define this lagged variable. Is it the previous election or the previous election of the same type?

21. The trend variable is equal to 1 for the elections that occurred in the first year of our sample, that is, in 1986 and to 25 in 2010.

22. The covariance is equal to -0.143, which amounts to a standard error for the sum of the two coefficients equal to 0.85 and thus to a value of Student t of 2.32.

23. This result is not shown to save space. We did not find any significant relationship between turnout and the level of democracy using the random effects model. In the OLS model, the estimated coefficient associated with the democracy variable was also not significant.

24. Which is equal to (2k + 2)/n, with k the number of regressors.

25. The expenditure figures are available from the Permanent Electoral Authority. Retrieved from http://www.roaep.ro/ro/section.php?id=28&l2=39&l3=45&lm= archived and http://www.roaep.ro/ro/section.php?id=28&l2=39&l3=45 for the 2008 legislative election and the 2009 presidential election, respectively (both accessed September 27, 2011).

References

Baltagi, Badi H., Matyas, Laszlo, & Sevestre, Patrick. (2008). Error components models. In Laszlo Matyas & Patrick Sevestre (Eds.), The econometrics of panel

data: Fundamentals and recent developments in theory and practice (pp. 49-87). Berlin, Germany: Springer-Verlag.

Blais, André. (2000). To vote or not to vote: The merits and limits of rational choice theory. Pittsburg, PA: University of Pittsburg Press.

Blais, André. (2007). Turnout in elections. In Russell J. Dalton & Hans-Dieter Klingemann (Eds.), The Oxford handbook of political behaviour (pp. 621-635). Oxford, UK: Oxford University Press.

Blais, André, Massicotte, Louis, & Dobrzynska, Agnieszka. (2003). Why is turnout higher in some countries than in others? Retrieved from http://www.elections.ca/ res/rec/part/tuh/TurnoutHigher.pdf

Brambor, Thomas, Clark, William Robert, & Golder, Matt. (2005). Understanding interaction models: Improving empirical analyses. Political Analysis, 13(1), 1-20.

Clark, Nick, & Rohrschneider, Robert. (2009). Second-order elections versus firstorder thinking: How voters perceive the representation process in a multi-layered system of governance. Journal of European Integration, 35(1), 645-664.

Costa Lobo, Marina, & Amorim Neto, Octavio (Eds.). (2010). O Semipresidencial-ismo nos Países de Lingua Portuguesa [Semi-presidentialism in Portuguese-speaking Countries]. Lisbon, Portugal: Imprensa de Ciências Sociais.

Cox, Gary W., & Munger, Michael C. (1989). Closeness, expenditures and turnout in the 1982 US House elections. American Political Science Review, §3(1), 217-230.

Dettrey, Bryan J., & Schwindt-Bayer, Leslie A. (2009). Voter turnout in presidential democracies. Comparative Political Studies, 42(10), 1317-1338.

Downs, Anthony. (1957). An economic theory of democracy. New York, NY: Harper & Row.

Elgie, Robert. (1999). The politics of semipresidentialism. In Robert Elgie (Ed.), Semi-presidentialism in Europe (pp. 1-21). Oxford, UK: Oxford University Press.

Elgie, Robert. (2011). Semi-presidentialism: Sub-types and democratic performance. Oxford, UK: Oxford University Press.

Elgie, Robert, & Schleiter, Petra. (2011). Variation in the durability of semi-presidential democracies. In Robert Elgie, Sophia Moestrup, & Yu-Shan Wu (Eds.), Semi-presidentialism and democracy (pp. 42-60). London, UK: Palgrave.

Endersby, James W., & Krieckhaus, Jonathan T. (2008). Turnout around the globe: The influence of electoral institutions on national voter participation, 1972-2000. Electoral Studies, 24(4), 601-610.

Fornos, Carolina A., Power, Timothy J., & Garand, James C. (2004). Explaining voter turnout in Latin America, 1980 to 2000. Comparative Political Studies, 37(8), 909-940.

Geys, Benny. (2006). Explaining voter turnout: A review of aggregate-level research. Electoral Studies, 25, 637-663.

Gray, Mark, & Caul, Miki. (2000). Declining voter turnout in advanced industrial democracies, 1950 to 1997: The effects of declining group mobilization. Comparative Political Studies, 33(9), 1091-1122.

Hix, Simon. (1998). Elections, parties and institutional design: A comparative perspective on European Union democracy. West European Politics, 21(3), 19-52.

Hix, Simon, & Marsh, Michael. (2011). Second-order effects plus pan-European political swings: An analysis of European Parliament elections across time. Electoral Studies, 30, 4-15.

Kam, Cindy D., & Franzese, Robert J., Jr. (2007). Modeling and interpreting interactive hypotheses in regression analysis. Ann Arbor: University of Michigan Press.

Koepke, Jason R., & Ringe, Nils. (2006). The second-order election model in an enlarged Europe. European Union Politics, 7(3), 321-346.

Kostadinova, Tatiana, & Power, Timothy J. (2007). Does Democratization Depress Participation? Voter Turnout in the Latin American and Eastern European Transitional Democracies. Political Research Quarterly, 60(3), 363-377.

Krouwel, André. (2003). Measuring presidentialism and parliamentarism: An application to Central and East European Countries. Acta Politica, 38, 333-364.

Kuenzi, Michelle, & Lambright, Gina M. S. (2007). Voter Turnout in Africa's Multiparty Regimes. Comparative Political Studies, 40(6), 665-690.

Liang, Kung-Yee, & Zeger, Scott L. (1986). Longitudinal data analysis using generalized linear models. Biometrika, 73(1), 13-22.

Lijphart, Arend. (1997). Unequal participation: Democracy's unresolved dilemma. American Political Science Review, 91(1), 1-14.

Lutz, Georg, & Marsh, Michael. (2007). Introduction: Consequences of low turnout. Electoral Studies, 26, 539-547.

Marsh, Michael. (1998). Testing the second-order election model after four European elections. British Journal of Political Science, 28(4), 591-607.

Matsusaka, John G., & Palda, Filip. (1993). The Downsian voter meets the ecological fallacy. Public Choice, 77(4), 855-878.

Metcalf, Lee Kendall. (2000), Measuring presidential power. Comparative Political Studies, 33(5), 660-685.

Moestrup, Sophia. (1999). The role of actors and institutions: The difficulties of democratic survival in Mali and Niger. Democratization, 6(2), 171-186.

Pacek, Alexander C., Pop-Elches, Grigore, & Tucker, Joshua A. (2009). Disenchanted or discerning: Voter turnout in post-communist countries. Journal of Politics, 71(2), 473-491.

Pierce, Roy. (1991). The executive divided against itself: cohabitation in France, 1986-1988. Governance, 4(3), 270-294.

Protsyk, Oleh. (2006). Intra-executive competition between president and prime minister: Patterns of institutional conflict and cooperation under semipresidentialism. Political Studies, 54, 219-244.

Rallings, Colin, & Thrasher, Michael. (2007). The turnout "gap" and the costs of voting—A comparison of participation at the 2001 general and 2002 local elections in England. Public Choice, 131, 333-344.

Reif, Karlheinz, & Schmitt, Hermann. (1980). Nine second-order national elections: a conceptual framework for the analysis of European election results. European Journal of Political Research, 8, 3-44.

Samuels, David, & Shugart, Matthew. (2010). Presidents, parties, prime ministers: A framework for analysis. Cambridge, UK: Cambridge University Press.

Schleiter, Petra, & Morgan-Jones, Edward. (2010). Who's in charge? Presidents, assemblies, and the political control of semi-presidential cabinets. Comparative Political Studies, 43(11), 1414-1441.

Schmitt, Hermann. (2005). The European Parliament elections of June 2004: Still second-order? West European Politics, 28(3), 650-679.

Sedelius, Thomas, & Ekman, Joakim. (2010). Intra-executive conflict and cabinet instability: Effects of semi-presidentialism in Central and Eastern Europe. Government and Opposition, 45(4), 505-530.

Shugart, Matthew Soberg, & Carey, John M. (1992). Presidents and assemblies: Constitutional design and electoral dynamics. Cambridge, UK: Cambridge University Press.

Siaroff, Alan. (2003). Comparative presidencies: The inadequacy of the presidential, semi-presidential and parliamentary distinction. European Journal of Political Research, 42(3), 287-312.

Skach, Cindy. (2005). Borrowing constitutional designs: Constitutional law in Weimar Germany and the French Fifth Republic. Princeton, NJ: Princeton University Press.

Tavits, Margit. (2009). Direct presidential elections and turnout in parliamentary contests. Political Research Quarterly, 62(1), 42-54.

Van Cranenburgh, Oda. (2008). "Big men" rule: Presidential power, regime type and democracy in 30 African countries. Democratization, 15(5), 952-973.

Van der Eijk, Cees, Franklin, Mark, & Marsh, Michael. (1996). What voters teach us about Europe-wide elections: What Europe-wide elections teach us about voters. Electoral Studies, 15(2), 149-166.

Wooldridge, Jeffrey M. (2003). Cluster-sample methods in applied econometrics. American Economic Review, 93(2), 133-138.

Wooldridge, Jeffrey M. (2010). Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press.

Robert Elgie is Paddy Moriarty Professor of Government and International Studies at Dublin City University, Ireland. He is the coeditor with Andrew Appleton of the journal French Politics, published by Palgrave Macmillan. He has published widely on semipresidentialism, including Semi-Presidentialism: Sub-types and Democratic Performance (Oxford University Press, 2011).

Christine Fauvelle-Aymar is associate professor of economics at the University of Tours, France. She has done extensive research on turnout in different electoral contexts and on voting behavior in France. Her research has been published in journals such as French Politics, Electoral Studies, and Public Choice.