Electoral Studies. The personal vote and voter turnout. Joseph W. Robbins * abstract. Contents lists available at ScienceDirect

Transcription

1 Electoral Studies 29 (2010) Contents lists available at ScienceDirect Electoral Studies journal homepage: The personal vote and voter turnout Joseph W. Robbins * Department of Political Science, Shepherd University, Shepherdstown, WV 25405, USA article info abstract Article history: Received 10 December 2009 Received in revised form 6 May 2010 Accepted 1 July 2010 Keywords: Electoral systems Personal vote index Voter turnout Political participation The level of electoral turnout is arguably the most widely monitored form of electoral participation. Consequently, electoral systems have often been cited as having a significant effect on turnout levels even though scholars do not agree on the effects of these complex institutions. Since most previous studies have relied on categorical or dichotomous electoral system indicators, this study utilizes Carey and Shugart s personal vote index to gain theoretical leverage on other electoral system components. In short, I find that where electoral competition is predicated on party, rather than candidates, reputations, turnout levels rise. The results of a time-series cross-sectional analysis reveal that the personal vote index significantly influences turnout levels even when controlling for a host of other factors. Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction Elections are generally perceived as one of the most vital components of a democratic state. Particularly important, is the rate at which people turn out to vote in these contests. Some argue that greater participation levels are a boon for democratic health. Lijphart s (1997) work underscores this idea as he notes that higher turnout rates reflect greater representation of the entire populace. Although it is not the only form of electoral participation, voter turnout is arguably the most widely recognized political activity examined in prior work. Accordingly, uncovering the factors behind electoral participation is an important endeavor for political scientists. To date, much has been written on the causes behind voter turnout. These works have emphasized the importance of cultural explanations (Almond and Verba, 1963; Inglehart, 1988; Putnam, 1993), socioeconomic conditions (Pacek and Radcliff, 1995a,b,c; Radcliff 1988) and various political factors (Fornos et al., 2004; Kostadinova, 2003; * Tel.: þ address: jrobbins@shepherd.edu. Pacek et al., 2009). Meanwhile, other researchers underscore the relevance of institutional variables and, in particular, electoral systems (Jackman, 1987; Jackman and Miller, 1995; Kostadinova and Power, 2007; Powell, 1980, 1986). Most of these studies demonstrate that proportional representation (PR) systems engender higher turnout figures; however, not everyone agrees with this conclusion. Despite the emphasis on electoral systems, a theoretical impasse has emerged. Because larger districts (i.e., PR systems) often produce more multiparty systems (Duverger, 1954), many scholars have surmised that turnout should be higher since there are more parties from which voters can choose and since races should be more competitive (Blais and Carty, 1990; Blais and Dobrzynska, 1998; Franklin, 2003; Gosnell, 1930; Jackman and Miller, 1995; Ladner and Milner, 1999; Powell, 1980, 1986; Tingsten, 1937). Conversely, others have shown that larger electoral districts may not, in fact, foment higher turnout levels (Brockington, 2004; Kostadinova, 2003; Kostadinova and Power, 2007). As a result, additional work is needed to better understand the relationship between electoral systems and voter turnout. What, then, is the next step in addressing this impasse? One solution has been to delve deeper into the links /$ see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi: /j.electstud

2 662 J.W. Robbins / Electoral Studies 29 (2010) between parties and voters by using survey data (e.g., Brockington, 2004; Karp and Banducci, 2008; Karp et al., 2008). Another possible solution is to take a closer look at electoral systems themselves. After all, much of the extant work uses categorical or binary indicators of electoral systems (Blais and Dobrzynska, 1998; Endersby and Krieckhaus, 2008; Fornos et al., 2004; Jackman, 1987; Kuenzi and Lambright, 2007; Powell, 1986). 1 While laudable, this neglects a number of other relevant electoral system variables which may also shape turnout levels. Much of the extant research has neglected to adequately account for other pertinent electoral system components that may also affect turnout rates. This is an important omission since previous work has argued that electoral systems shape party mobilization efforts, and, hence, turnout levels (Canon, 1999; Cox, 1999). But the previous work on voter turnout has yet to assess electoral considerations other than district size. This is surprising since Carey and Shugart (1995) have identified three additional noteworthy components found in most electoral systems: the degree to which votes are pooled across candidates; if parties control the ballot structure; and, whether people vote for candidates, parties or both, at the polls. Together, these rules represent the personal vote index, which reveals how personalistic or candidate-centered a given electoral system is. Therefore, in addition to controlling for district magnitude and PR systems, I build upon the existing work on voter turnout by including the personal vote index derived from Carey and Shugart s (1995) seminal study. I predict that the personal vote metric should be inversely related to turnout. In short, I anticipate that parties should work harder to increase turnout levels when their electoral fate hinges upon their own reputations. After including a series of standard controls, the results clearly support the notion that there is an inverse relationship between the incentives to cultivate a personal vote and voter turnout. Furthermore, the results also reveal that the personal vote index is better able to predict turnout levels than PR system dummies. The rest of this paper proceeds as such. The next section recasts the literature on electoral systems and voter turnout, followed by a detailed explanation of the argument that the incentives to cultivate a personal vote should be inversely related to voter turnout. This discussion will focus on how both individual voters and parties respond to electoral system incentives, and how this relates to turnout. Afterwards, I elaborate on the data, variables, and methods used in the analysis. And, following the results of a crosssectional time-series analysis, the key findings are presented along with an explication of the implications for future research. 1 One exception is Tavits (2009). Her recent study examines the assumption that directly elected presidents behave differently than indirectly-elected ones. In chapter six of her book, Tavits looks at list type in addition to district magnitude but this still overlooks a number of other relevant electoral attributes. 2. Electoral systems and voter turnout By now, numerous studies have affirmed the link between electoral systems and voter turnout. Much of this discussion hinges on party strategies as products of electoral rules. To build on the previous research, I argue that the incentives to cultivate a personal vote shape party strategies, and by extension, have an effect on voter turnout. Specifically, more party-centric systems should exert a positive effect on voter participation. That is, parties should be more inclined to mobilize voters everywhere if their electoral fate is predicated on the party s reputation, rather than individual candidates. This is consistent with the broader literature on voter turnout which has, for some time, found that voter turnout is contingent upon electoral laws. Before turning to the specifics of my argument, I present a brief overview of the research on electoral laws and their impact on voter turnout levels. As mentioned above, a consensus has yet to emerge regarding the benefits of one particular electoral system with respect to voter turnout. A number of studies have conjectured that PR systems increase incentives for individuals to vote. This conclusion is based on both theoretical (Cox, 1999) and empirical evidence (Blais and Carty, 1990; Blais and Dobrzynska, 1998; Franklin, 2003; Jackman, 1987; Jackman and Miller, 1995; Ladner and Milner, 1999; Powell, 1986). The ensuing discussion will review the relationship between electoral rules (i.e., the incentives to cultivate a personal vote) and voter turnout by focusing on how individuals and parties respond to these rules. As such, a few additional comments on the microfoundational relevance of electoral system mechanisms are warranted. At the aggregate-level, the literature suggests that PR systems foster greater participation since larger electoral districts tend to be more prominent contests that have more nationwide appeal. Thus, turnout should be higher when there is more at stake which is often the case in PR contests. In addition, single member districts (SMD) tend to be less competitive (Powell, 1980) which means that voters may refrain from voting when they believe that their vote will not make a difference in the outcome. Conversely, PR systems may result in higher turnout rates since districts are typically more competitive. Furthermore, since PR systems tend to produce multiparty systems (Duverger, 1954), parties have less of an incentive to appeal to the median voter (Cox, 1990). When the political landscape is populated with multiple parties, these organizations attempt to differentiate themselves from their competitors (Katz, 1980). As a result, voters especially those on the ideological fringe are more likely to participate when they have more options (Downs, 1957). Lastly, given the threshold for securing representation, every vote is potentially more valuable in a PR race. Since parties competing in PR contests do not need to secure a majority or plurality of votes to gain representation, there is a greater probability that each individual s vote will have an impact on the electoral outcome. Much of the existing literature finds that parties in PR races should foster turnout everywhere to better their chances of electoral success (Gosnell, 1930; Powell, 1986;

3 J.W. Robbins / Electoral Studies 29 (2010) Tingsten, 1937). In fact, several studies indicate that larger, more nationwide contests boost turnout because parties exert greater effort cajoling voters to the polls. The first reason underlying this finding has to do with the disproportionality between seats and votes (Ladner and Milner, 1999). Parties in plurality contests often do not devote equal resources and efforts in all races, and tend to focus on competitive or marginal contests; therefore, party mobilization is often lacking in less competitive districts (Cox, 1999). Meanwhile, a number of studies have already posited that parties are often rewarded at the polls for their mobilization efforts (Cox, 1999; Gerber and Green, 2000), even though they do not exert the same effort in all districts. Consequently, mobilization strategies are likely to be a function of electoral rules. Another reason for higher turnout stems from the linkages established in PR races. In general, political parties forge stronger ties under these contests than they do in plurality or SMD races. Accordingly, this may facilitate mobilization efforts, which should result in more votes cast. Because of this, Powell (1980) contends that higher turnout is more likely in PR systems due to the stronger ties between parties and various social groups. Multi-member PR electoral systems have regularly been linked with higher turnout numbers for other reasons. Some argue that this is because political parties will mobilize more fervently in all electoral districts. In order to secure representation, parties competing in these contests do not need to win as many votes as they would in a plurality system. Accordingly, campaigning everywhere is an effective strategy, especially when other electoral rules permit pooling a party s votes across candidates or districts. In addition, other scholars suggest that turnout numbers are higher in PR races because these contests often engender greater competition throughout the electoral system. The extant work has previously held that there is a higher probability of voting in close or more competitive races, and thus, it follows that PR races should produce higher turnout rates. Not all scholars agree that PR systems correspond with higher turnout. There are at least two reasons why these systems may hamper participation. First, because multimember PR races are often the precursor to coalition governments in parliamentary systems, individuals may abstain from voting since it is unlikely that their vote will decide the government s final composition. That is, since parties and political elites must wrangle and jockey for positions in the governing coalition, the vote cast for the initial parliamentary contest is but a first in a series of multiple votes. Hence, voters may be discouraged by this and withdraw from the electoral process. Second, these contests may also confound or limit denizens ability to differentiate among parties and their respective policy platforms. With more options, the costs of being informed on each competitor rise, which may discourage voters from participating (Fornos et al., 2004; Jackman, 1987; Jackman and Miller, 1995; Kostadinova, 2003; Kostadinova and Power, 2007). While Downs contends that these races will result in ideologically separate political parties, multiparty systems may still attenuate individuals propensity to evaluate their prospective choices in the ballot booth. Although researchers continue to debate the merits of electoral systems with respect to voter turnout, most political scientists concur that multi-member PR systems raise turnout levels. The merits of these studies are wellnoted and have certainly furthered our understanding of the relationship between electoral systems and voter turnout. However, the aforementioned studies have generally neglected to consider other electoral nuances that may also influence turnout levels. Fortunately, recent work has advanced our understanding, and measurement, of electoral rules (Carey and Shugart, 1995; Wallack et al., 2003). That said, focusing on additional electoral laws may add to our understanding of voter turnout. I now turn to a discussion of these additional electoral rules and explain how they relate to predicting aggregate participation levels. 3. The argument What s the relevance of the personal vote index to voter turnout? Aside from the evidence cited above, there are other reasons why electoral system nuances may influence potential voters. In particular, recent experimental evidence reveals that electoral laws affect how individuals process information. Chin and Taylor-Robinson (2005) discover that party effects were more influential to individuals in the closed-list PR systems than to those in SMD systems. Likewise, Tavits (2009) considers this possibility in her investigation of the implications for directly elected presidents. Because most previous studies have largely neglected considerations such as list type and other pertinent electoral system attributes, our understanding of the relationship between electoral systems and turnout remains incomplete Hypotheses The first electoral system variable assessed is district magnitude. Most existing studies have tested the effects of district size through dichotomous or categorical measures. Rather than relying on this approach alone, I include a continuous measure (District Magnitude) to control for district size. In line with most previous studies, I expect district size to positively correspond with turnout. Parties that compete in larger or nationwide districts should try and convince more people to vote, as they attempt to mobilize voters everywhere in hopes of increasing their vote share. Thus, I expect that: Hypothesis 1: District magnitude should be positively related to turnout levels. Although countries may have similar district magnitudes, the other rules governing the electoral process may vary tremendously. Consider the cases of Moldova and Slovakia. Both countries have similar district magnitudes (both countries mean magnitudes surpass 100 seats per district), but their personal vote scores unveil interesting differences. Slovakia s electoral system places greater emphasis on individual candidates, whereas Moldova s

4 664 J.W. Robbins / Electoral Studies 29 (2010) system is more party-centric (mean scores are 3 and 0, respectively). 2 Put differently, there are greater incentives to cultivate a personal vote in Slovakia than in Moldova. Another interesting juxtaposition lies with Japan and Ecuador. The Japanese electoral system is predicated on the candidate rather than the party, and due to this, their personal vote score is rather high (mean score ¼ 3). Yet, even though Ecuador has roughly the same district magnitude as Japan (5.58 for Ecuador, and 5 for Japan), it places less emphasis on individual candidates (mean personal vote ¼ 1.17). As such, these nuances should play a key role in constructing electoral strategies for politicians and parties in more party-centric states such as Moldova and/or Ecuador, thus possibly resulting in higher turnout levels. Clearly, district magnitude is not the only electoral system component that should have an effect on turnout. After all, previous research has shown that other electoral laws have been linked to policy-making as legislators use public policies to further re-election efforts (Carey and Shugart, 1995; Wallack et al., 2003). That said, these laws could significantly alter campaign strategies, and, hence, turnout levels. Fortunately, Carey and Shugart (1995) have bolstered our understanding of these integral rules. In their renowned work, they have identified three key characteristics that influence political parties and candidates: whether parties have control over nominations or ballot structure (Ballot), if votes are pooled across party nominees (Pool), and if votes are cast for parties or individual candidates (Vote). Borrowing from the extant personal vote literature, I anticipate that the incentives to cultivate a personal vote should significantly influence electoral strategies. Ballot control, or list type, is especially intriguing since it determines the amount of control parties possess over ballot formation, and therefore parties incentives may vary with respect to list type. Although many advanced industrial democracies utilize this arrangement, others, such as Brazil, opt for an open-list system. In open-list systems, politicians are exposed to both intra- and inter-party competition. These characteristics, moreover, are likely to create stronger incentives for politicians to create a personal vote or build their own reputations rather than relying solely on the party s reputation. As such, parties may not exert the same level of resources to solicit support or mobilize voters given the incentives created by an openlist system. In contrast, individual candidates running on an open-list will appeal to his/her supporters but will likely avoid mobilization strategies that involve the entire populace. After all, they want their supporters to turnout in higher numbers, not all voters. In closed-list systems, parties create the lists and may behave differently when they construct their own ballots. In this system, parties place greater emphasis on mobilizing voters everywhere in hopes of soliciting additional support. After all, if they construct the list then they are responsible for the success of their candidates and will devote more time, energy and 2 All personal vote data are obtained from Johnson and Wallack s (2008) dataset. resources culling individuals to the polls. If multiple parties respond similarly, then, national turnout should increase as more parties reach out to additional voters. Closed-list systems exist in countries where parties control ballot formation. If a party s success is contingent upon its own mobilization efforts, then there should be more emphasis placed on increasing turnout in order to secure more electoral support, such as in open-list systems, then turnout should be lower. Candidates will campaign much the same as parties but rather than appealing to voters everywhere, individual candidates should focus their efforts on narrower constituencies. In open-list systems, moreover, party leaders are not likely to benefit from higher turnout and should expend less effort and resources trying to amplify turnout rates. Furthermore, because these rules influence electoral outcomes, then individual voters should respond to these rules as well. If an individual recognizes that his/her candidate is unlikely to gain a seat, then why should s/he vote? Hence, fewer voters should cast a ballot as parties and candidates allocate fewer resources mobilizing would-be supporters. Formally stated, Hypothesis 2: Voter turnout should be higher where ballot formation is controlled by parties or under closed-list systems. The second measure (Pool) represents systems where votes are pooled across the party. Where votes are pooled across candidates they can utilize extra votes and are responsible for divvying up these votes. Consequently, parties stand to gain more with higher turnout levels. Parties stand to gain additional support if leftover votes can benefit the party as a whole. These remaining votes can increase the party s seat share, thereby giving them a larger share in the legislature. Since parties stand to gain more when votes are pooled across candidates, they are expected to work harder to produce higher turnout levels. Or, with more to gain, parties should alter their electoral strategies in such a way as to maximize their reach to voters. As more parties seek to maximize vote totals, there should be a positive effect on overall turnout. Predictably, turnout rates should increase where votes are pooled across candidates and decline where they are not. Or, if votes are not pooled and there is more incentive to emphasize individual candidates, then turnout rates should fall. In other words, Hypothesis 3: Voter turnout should be higher in states where votes are pooled. The final component of the personal vote index (Vote) measures whether denizens vote for parties or individual candidates. Parties should exert greater effort in mobilizing voters and increasing overall turnout where people vote for a party rather than individual candidates. Parties should respond to such a mechanism by ensuring that more voters participate since they the parties have more to gain. In contrast, the need for higher participation declines where votes are cast for individual candidates rather parties. In systems that place a premium on the candidate s

5 J.W. Robbins / Electoral Studies 29 (2010) reputation (rather than the parties) and where politicians prefer support from a more narrowly targeted constituency, fewer people are expected to cast a ballot. Thus, Hypothesis 4: Voter turnout should be higher where voters vote for parties rather than individual candidates. Finally, more party-centric systems should exert a positive effect on voter participation. That is, parties should be more inclined to mobilize voters everywhere if their electoral fate is predicated on the party s reputation. That said, I have combined the Ballot, Pool and Vote figures to form the personal vote index measure used in the analysis below. Higher personal vote scores reflect more candidatecentered systems, which should be inversely related to turnout levels. Or, Hypothesis 5: The personal vote index should be inversely related with turnout levels. 4. Research design In order to test the stated hypotheses, a large-n pooled cross-sectional time-series approach is employed. The cases evaluated here include democracies from Africa, Asia, East-Central Europe, Latin America and the OECD. The countries included reached at least a 6 on the Polity2 measure from the Polity IV database (Polity IV Project, 2000). Due to data limitations, not all democracies are included in the analysis. A list of the countries and years used in the analysis can be found in the appendix Dependent variable: voter turnout The dependent variable in this study is the percentage of eligible registered voters who cast a ballot in a legislative election. This measure has been commonly used in previous studies (Blais and Dobrzynska, 1998; Franklin, 1999; Kostadinova, 2003). Not all scholars agree on the best measure for voter turnout, though Geys (2006, p. 639) provides some insight into why this metric may be preferable to other voter turnout calibrations. I have also substituted this measure with another common indicator, voter turnout as a percentage of the age-eligible voting population but this substitution does not fundamentally alter the results presented below, and was thus omitted from the analysis. Legislative elections are selected for multiple reasons. With few exceptions, most studies have also emphasized turnout in these contests (Blais and Dobrzynska, 1998; Kostadinova, 2003; Kostadinova and Power, 2007; Powell and Whitten, 1993). Also, the number of presidential elections pales in comparison to the number of legislative contests. Numerous observations were lost when presidential and legislative races were tested separately; therefore, I focus only on legislative contests. Furthermore, there are also coding difficulties with some presidential regimes. For instance, should researchers include systems where the president is selected by the legislature? Should all presidential regimes be treated equally? Due to these questions, this study Table 1 Personal vote index. Variable Value Electoral rule Ballot 0 Party leaders present a fixed ballot. 1 Party leaders present party ballots, but voters may disturb list. 2 Party leaders do not control access to ballots or rank. Pool 0 Votes for candidates pooled across the whole party. 1 Votes for candidates are pooled at the sub party level. 2 Votes for candidates are not pooled. Vote 0 Voters cast votes for a single party. 1 Voters cast votes for multiple candidates. 2 Voters cast a single vote below the party level. Source: Carey and Shugart (1995). measures turnout as the percentage of registered voters who turned out in legislative contests. Turnout data were gathered from the International Institute for Democracy and Electoral Assistance s (IDEA) website (IDEA, 2008a,b) Electoral system variables The first two electoral system indicators used are a binary indicator for PR systems and district magnitude. Although previous work utilizes either Powell s (1986) categorical measure for nationally competitive districts or dichotomous indicators, this study tests both the binary PR variable as well as district magnitude. 3 This should provide greater theoretical leverage for assessing the salience of electoral systems than either variable in isolation. These data were obtained from Beck et al. s (2004) Dataset on Political Institutions. The other primary electoral system variable considered is the personal vote index. This is a composite index that accounts for whether votes are pooled across candidates, if party leaders control ballot access, and if votes are cast for parties rather than individual candidates (see Table 1). The updated version of this measure is used in the analysis below, and these data are obtained from the Johnson and Wallack (2008) dataset, which is based on the aforementioned Carey and Shugart s (1995) index. Higher personal vote scores represent systems where there is a greater incentive to cultivate a personal vote. Thus, higher personal vote values should be inversely related to participation levels (Hypothesis 5). In addition to simply running the models with the composite index, I have also tested each component separately. However, none of these measures are truly continuous. In order to facilitate interpretation, then, these data should be modified. I have recoded these variables such that 0 indicates those states where parties have control over the ballot, votes are pooled by the party, and voters cast a ballot for parties, not candidates (denoted Ballot ¼ 0, Pool ¼ 0, and Vote ¼ 0, respectively). Each variable, moreover, should be inversely related to turnout since the most 3 I have estimated the models with both the standard district magnitude and the more conservative logged version (see the Reviewer s Appendix), but the primary results are unchanged regardless of the specification chosen.

6 666 J.W. Robbins / Electoral Studies 29 (2010) party-centric systems are denoted by ballot ¼ 0, pool ¼ 0, and/or vote ¼ 0. This approach should also facilitate interpreting the substantive findings. Given the diversity of countries in the dataset, it is also necessary to discuss the operationalization of mixed member electoral systems. Coding these cases requires additional clarification with respect to both electoral system measures. District magnitudes in mixed-systems are based on the average magnitude between the two tiers (i.e., the PR and SMD tiers). This approach is commonly used, and these data are found in the Beck et al. (2004) dataset. For the personal vote, there is additional information available from the Johnson and Wallack (2008) dataset that presents the personal vote scores for mixed systems. Therefore, these values require no additional calculations to account for these hybrid regimes. Table 2 Summary statistics for dependent and independent variables. Independent Variable Mean Standard Minimum Maximum Deviation Turnout (%) Ballot Vote Pool Personal vote index PR systems District magnitude Concurrent Election Compulsory Level Bicameral OECD member ENP GDP per capita , , DGDP per capita Inflation , Unemployment (%) Urban population (%) Additional controls Aside from the variables already mentioned, I also employ a set of standard economic, political and social controls: concurrent elections, compulsory systems, GDP per capita, change in GDP per capita, inflation, unemployment, effective number of parties (ENP) 4, federalism and percent urban population (Aguilar and Pacek, 2000; Almond and Verba, 1963; Blais and Dobrzynska, 1998; Fornos et al., 2004; Franklin, 1999; Gray and Caul, 2000; Inglehart, 1988; Jackman, 1987; Jackman and Miller, 1995; Kostadinova, 2003; Kostadinova and Power, 2007; Kuenzi and Lambright, 2007; Norris, 2004; Pacek and Radcliff, 1995b; Powell, 1986; Powell and Whitten, 1993; Power and Roberts, 1995; Remmer, 1991). A trichotomous measure for compulsory laws is used and, following the IDEA s coding scheme states that never enforce their obligatory rules are denoted by 0, while those that sometimes enforce their laws are coded 1, and those that diligently monitor turnout are coded 2. The relationship between the personal vote index and voter turnout is not impacted by the use of this trichotomous variable (versus a dichotomous measure) and given the variation in compulsory laws and enforcement, the IDEA coding scheme is retained. Unless stated otherwise, the economic controls have come from the World Bank s WDI (2006) database while the institutional measures were taken from Beck et al. s (2004) Database of Political Institutions. Each economic variable was lagged a year as is the standard practice. The percent urban population metric was obtained from the World Bank s WDI (2006) dataset. 5 Finally, a trichotomous indicator for federalist states, which is borrowed from 4 To be clear, the ENP variable is based on the inverse of party system fractionalization. Thus, this has been calculated with the fractionalization data from Beck et al. (2001). 5 I first attempted to control for cultural effects via literacy rates (as seen previously). Unfortunately, the measure available from the World Bank s World Development Indicators lacked sufficient data to include in this sample and I was forced to omit this variable. I have also considered using the circulation of daily newspapers (WDI, 2006), but missingness problems also precluded the use of this alternative control. Wibbels s (2000), is also included. Summary statistics of each variable can be found in Table Methodology There are several points that bear mentioning in this discussion as many different techniques have been used to estimate turnout models. Due to the temporal element inherent in such studies, a pooled cross-sectional timeseries estimation is used. This approach is chosen in order to maintain consistency with earlier studies. 6 Furthermore, a Prais Winsten regression analysis with panel-corrected standard errors (Beck and Katz, 1995) and random effects is employed for several reasons. Since I am interested in uncovering the country-to-country variation, a random effects model is better able to capture these differences (Beck and Katz, 2004, 2006). Furthermore, fixed effects models may eliminate stable, or time invarying covariates (e.g., compulsory voting systems, district magnitude, bicameral systems, etc.) from the analysis (2004, 2006); hence, random effects models are better suited for the current study. 7 In addition, after conducting multiple Wooldridge (2002) tests, the results revealed that serial correlation was present in the models. Thus, an AR1 process was utilized to address the concerns of serial correlation. 6 Although some may question the use of time-series I follow the approach seen in previous studies (e.g., Fornos et al., 2004; Pacek et al., 2009; Radcliff, 1992). These hypotheses were also tested without a time-series specification and the substantive results were virtually the same. For consistency, then, I have retained the time-series approach. These findings are available upon request. 7 In addition, given that the dependent variable (parliamentary turnout) is a percentage it is inherently truncated and not a continuous measure. This implies that a linear regression equation may be inappropriate, but after using the preferred tobit estimation these findings are substantively the same and thus I have kept the standard regression instead. A truncated regression estimation was also conducted and the findings hold regardless of the technique used. In sum, regardless of the estimation employed, I am confident that the results are not spurious nor are they an artifact of one particular model.

7 J.W. Robbins / Electoral Studies 29 (2010) Table 3 Time-series analysis results of voter turnout in parliamentary elections. Independent variables Model 1 Model 2 Model 3 Model 4 Ballot ¼ 0 (0,1) 3.27 (1.51)** Vote ¼ 0 (0,1) 3.04 (1.19)** Pool ¼ 0 (0,1) 3.91 (1.71)** Personal vote index 1.23 (0.52)** District magnitude 0.06 (0.02)*** 0.06 (0.02)*** 0.05 (0.02)*** 0.05 (0.02)** Concurrent (0,1) 5.34 (2.28)** 5.15 (2.25)** 5.76 (2.35)** 5.72 (2.46)** Compulsory 5.17 (1.79)*** 5.23 (1.82)*** 3.88 (1.88)** 4.59 (1.74)*** Bicameral (0,1) 6.22 (1.70)*** 6.33 (1.73)*** 5.29 (1.84)*** 5.70 (1.71)*** GDP per capita t (0.00)** 0.00 (0.00)** 0.00 (0.00)** 0.00 (0.00) ENP 0.26 (0.43) 0.16 (0.44) 0.31 (0.44) 0.31 (0.45) Inflation t (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Unemployment Pct. t (0.19) 0.17 (0.19) 0.07 (0.20) 0.09 (0.20) Urban population (%) 0.43 (0.08)*** 0.44 (0.08)*** 0.44 (0.08)*** 0.43 (0.08)*** OECD (0,1) (2.62)*** (5.71)*** (2.74)*** (2.79)*** Constant (5.76)*** (5.71)*** (5.85)*** (5.83)*** Observations Groups r Wald X *** 0.00*** 0.00*** 0.00*** *p < 0.10; **p < 0.05; ***p < 0.01; panel-corrected standard errors in parentheses. Additional estimations have been conducted to mitigate concerns that these findings are driven by particular cases. Specifically, a series of country-specific measures have been included, though not reported, to simulate a fixed effects model as seen in other studies (Radcliff, 1992; Pacek and Radcliff, 1995a,b,c). 8 The stated hypotheses are still in the expected direction and statistically significant. Additional robustness checks are enumerated at the end of the next section. 5. Discussion and findings The results of the analyses are provided in Tables 3 6.In Models 1 3 the individual components of the personal vote index are separately considered and in Model 4 the composite measure is reported. 9 Additionally, separate models are listed in Tables 3 6 since both GDP per capita and change in GDP per capita are tested separately. The overall findings are elaborated in the remainder of this section. Each model is statistically significant, therefore revealing that these variables jointly influence turnout levels in the countries surveyed. Do electoral systems shape turnout levels? From the results above, the answer is clear. Both district magnitude and the personal vote index measures are statistically significant and in the hypothesized directions. This shows that these variables correspond with higher turnout levels. Indeed, district magnitude corresponds with higher turnout rates which is consistent with the previous literature that finds that parties exert greater efforts to mobilize voters in these systems (Blais and Carty, 1990; Blais and Dobrzynska, 1998; Gosnell, 1930; Jackman, 1987; Jackman and Miller, 1995; Powell, 1980, 1986). For each increase in district magnitude we see that turnout 8 These results are available upon request. 9 This same pattern is repeated in each of the tables. increases by roughly %. Obviously, this is not a staggering figure. However, if we consider the effect of district magnitude when set at its minimum, mean and maximum values then the findings are more compelling. 10 Using this approach, we can more clearly see the effect of district magnitude with respect to overall participation levels. When set at its minimum value (district magnitude ¼ 1, or countries such as India, the United Kingdom and the United States, etc.), turnout levels are just over 60%. This is a small change from its mean value, where turnout levels are slightly higher at 61%. However, if district magnitude is set at its maximum (district magnitude ¼ 150,asseenincountriessuchasSlovakia,the Netherlands), the change is more dramatic. Based on this change, turnout levels increase to 68%. Thus, there is a drop of approximately 8% when we compare states with single member district electoral systems versus those with much larger district magnitudes commonly found in PR systems. District structure is an accurate predictor of voter turnout. But, is the personal vote index equally adept at forecasting turnout? In short, yes. More candidate-centered electoral systems are inversely related to overall turnout levels. Even when the constituent parts are disaggregated this result still holds. Each dichotomous component Ballot, Vote, and Pool inversely affect turnout rates. These findings confirm hypotheses 2 4. Meanwhile, for every increase in the personal vote measure (moving from more party-centered to candidate-centered regimes), we see a decrease in the dependent variable of approximately 1%. In other words, party-centric systems are correlated with higher turnout levels, ceteris paribus. Again, we can look at the expected turnout figures after manipulating the personal vote index. At its mean value, turnout levels are 10 These interpretations were calculated using King et al. s (2000) Clarify program. Each continuous independent variable was set at their respective mean, while the remaining variables were set at their respective measure of central tendency.

8 668 J.W. Robbins / Electoral Studies 29 (2010) Table 4 Time-series analysis of voter turnout in parliamentary elections. Independent variables Model 5 Model 6 Model 7 Model 8 Ballot ¼ 0 (0,1) 3.36 (1.52)** Vote ¼ 0 (0,1) 3.52 (1.52)* Pool ¼ 0 (0,1) 3.82 (1.66)** Personal vote index 1.50 (0.52)*** District magnitude 0.06 (0.02)*** 0.06 (0.02)*** 0.05 (0.02)*** 0.05 (0.02)** Concurrent (0,1) 6.20 (2.52)** 5.45 (2.33)** 6.51 (2.51)** 5.85 (2.51)** Compulsory 5.66 (1.73)*** 5.92 (1.75)*** 4.50 (1.76)** 5.11 (1.69)*** Bicameral (0,1) 7.20 (1.70)*** 7.21 (1.71)*** 6.27 (1.82)*** 6.32 (1.68)*** DGDP per capita t (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) ENP 0.20 (0.44) 0.17 (0.44) 0.28 (0.44) 0.37 (0.45) Inflation t (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Unemployment Pct. t (0.19) 0.03 (0.18) 0.16 (0.20) 0.08 (0.19) Urban Population (%) 0.35 (0.07)*** 0.37 (0.07)*** 0.36 (0.07)*** 0.38 (0.07)*** OECD (0,1) 7.06 (2.35)*** 6.64 (2.27)*** 7.13 (2.38)*** 7.24 (2.39)*** Constant (5.79)*** (5.68)*** (5.81)*** (5.72)*** Observations Groups r Wald X *** 0.00*** 0.00*** 0.00*** *p < 0.10; **p < 0.05; ***p < 0.01; panel-corrected standard errors in parentheses. roughly 62%. If the index is set at its minimum value there is an increase of roughly 3% (64.6%). Conversely, where there is greater emphasis on personal votes, the corresponding turnout level is around 57%. Therefore, if we move from a country such as Moldova, Portugal, or South Africa to countries like Colombia, France or the United States, there is a decrease in turnout of more than 7%. Some readers, however, may question whether or not this relationship holds when controlling for electoral district type. To answer this question, it is necessary to evaluate the salience of the PR binary measure. In doing so, I find that the primary results (i.e., the personal vote index and its constituent parts) remain statistically significant and negatively signed. Furthermore, the results in Tables 5 and 6 also report that the PR variable is insignificant when considered alongside the personal vote measures. This corroborates the primary theme of this paper that additional electoral indicators are needed to fully understand the effects of these complex institutions. In sum, we find that electoral system components have an effect on turnout levels even when controlling for other factors. Finally, I have also considered the effect of the most party- and candidate-centered systems. The former refers to systems where district magnitude is set at its maximum value and the personal vote index is set at its minimum. In this case, expected turnout is over 71%, which is the largest entry in Table 7. In contrast, I have similarly assessed the opposite scenario, where there are more incentives to cultivate a personal vote (i.e., where district magnitude is set at 1 and where the incentive to cultivate a personal vote is set at its maximum value). In doing this, the anticipated Table 5 Time-series analysis of voter turnout in parliamentary elections with PR dummy. Independent variables Model 9 Model 10 Model 11 Model 12 Ballot ¼ 0 (0,1) 4.63 (1.99)** Vote ¼ 0 (0,1) 4.45 (2.31)* Pool ¼ 0 (0,1) 5.05 (2.71)* Personal vote index 2.31 (0.96)** PR systems (0,1) 0.45 (2.92) 1.65 (2.79) 1.67 (3.51) 0.52 (3.20) District magnitude 0.08 (0.02)*** 0.07 (0.02)*** 0.06 (0.02)*** 0.08 (0.03)*** Concurrent (0,1) 7.71 (2.65)*** 6.78 (2.51)*** 7.27 (2.64)*** 7.99 (2.68)*** Compulsory 5.78 (2.11)*** 5.45 (2.23)*** 2.62 (2.93) 4.96 (2.14)*** Bicameral (0,1) 7.41 (2.26)*** 7.18 (2.25)*** 5.45 (2.27)** 7.05 (2.22)*** GDP per capita t (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) ENP 0.09 (0.61) 0.19 (0.63) 0.45 (0.64) 0.04 (0.62) Inflation t (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Unemployment Pct. t (0.21) 0.00 (0.20) 0.11 (0.22) 0.14 (0.21) Urban population (%) 0.39 (0.10)*** 0.42 (0.10)*** 0.44 (0.10)*** 0.41 (0.10)*** OECD (0,1) (3.12)*** (3.02)*** (3.28)*** (3.26)*** Constant (7.93)*** (7.93)*** (7.99)*** (7.88)*** Observations Groups r Wald X *** 0.00*** 0.00*** 0.00*** *p < 0.10; **p < 0.05; ***p < 0.01; panel-corrected standard errors in parentheses.

9 J.W. Robbins / Electoral Studies 29 (2010) Table 6 Time-series analysis of voter turnout in parliamentary elections with PR dummy and change in GDP per capita. Independent variables Model 13 Model 14 Model 15 Model 16 Ballot ¼ 0 (0,1) 4.72 (1.94)** Vote ¼ 0 (0,1) 4.98 (2.30)** Pool ¼ 0 (0,1) 4.54 (2.55)* Personal Vote Index 2.25 (0.95)** PR systems (0,1) 1.76 (2.83) 2.86 (2.74) 3.65 (3.30) 0.81 (3.05) District magnitude 0.08 (0.02)*** 0.07 (0.02)*** 0.06 (0.02)*** 0.08 (0.02)*** Concurrent (0,1) 7.47 (2.67)*** 6.59 (2.50)*** 6.97 (2.64)*** 7.93 (2.74)*** Compulsory 6.08 (2.10)*** 5.88 (2.18)*** 2.84 (3.09) 5.21 (2.15)** Bicameral (0,1) 7.27 (2.20)*** 7.03 (2.16)*** 5.20 (2.20)** 6.96 (2.16)*** DGDP per capita 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) ENP 0.27 (0.65) 0.33 (0.68) 0.69 (0.68) 0.20 (0.66) Inflation t (0.00) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Unemployment Pct. t (0.22) 0.07 (0.21) 0.16 (0.22) 0.23 (0.22) Urban Population (%) 0.33 (0.08)*** 0.36 (0.08)*** 0.37 (0.08)*** 0.35 (0.07)*** OECD (0,1) 9.23 (2.72)*** 8.70 (2.73)*** 7.10 (2.67)*** (2.88)*** Constant (7.69)*** (7.66)*** (7.84)*** (7.68)*** Observations Groups r Wald X *** 0.00*** 0.00*** 0.00*** *p < 0.10; **p < 0.05; ***p < 0.01; panel-corrected standard errors in parentheses. turnout rate is just over 50% of registered voters (56.2%). When the most party- and candidate-centered systems are compared, the difference in aggregate turnout is roughly 15%. All said, in systems where personal votes are at a premium, the corresponding effect on turnout levels is substantial. These results are summarized in Table 7. In addition to the primary findings discussed, there are other notable results from Tables 3 6. The findings for bicameral, compulsory and concurrent systems are significant and in the hypothesized directions the former depresses turnout while the latter two increase it. Judging from previous work on voter turnout, these results are exactly what one would expect to find. The effective number of parties, moreover, has also been emphasized in voter turnout studies; however, the results do not support this finding. Although the coefficients are in the expected direction, it falls short of traditional thresholds of statistical significance. Thus, once other aspects of electoral systems are controlled for, the importance of the ENP disappears. Also, both the OECD and percentage of urban population variables were accurate predictors of turnout levels. We see that OECD states Table 7 Estimated values for turnout levels. Expected turnout levels Personal vote (%) District magnitude (%) Minimum Mean Maximum Party-Centric a Candidate-Centric b a This represents the expected turnout level when district magnitude is set at its maximum value and the personal vote index is set at 0. b This reflects the expected turnout when district magnitude is set at its minimum value and the personal vote index is set at its maximum value (5). are associated with greater participation levels, as are more urbanized states. However, the various economic indicators did not reveal consistent, statistically significant relationships. The only exception is the relationship between GDP per capita (Models 1 3), but this is tenuous as it is insignificant in the remaining models. The null finding for most of these controls is particularly puzzling since there have been a litany of works espousing the importance of economic conditions vis-à-vis participation rates. However, this is certainly not the only study to uncover such a finding (e.g., Fornos et al., 2004: 923). 6. Robustness checks A number of robustness checks have been undertaken to test the reliability of the results in Models One of the first attempts involved testing for collinearity among the independent variables (including the two electoral system indicators). For the most part, concerns of collinearity are unjustified. Even the correlation between the primary electoral variables is small (Pearson s correlation ¼ 0.20); hence, including both measures in Models 1 16 does not distort the substantive findings. Furthermore, additional tests were conducted to evaluate the potential distorting effects of any outliers. After all, it is possible that the findings in Tables 3 6 are based on the influence of outliers or the estimation technique used. To detect the presence of outliers, such as the United States and/or Switzerland, Cook s D test was used to assess the leverage exerted by potential outliers. After identifying these cases, I ran additional tests without these cases and the primary results were the same as those seen above. Thus, the inclusion of these cases does not alter the substantive findings. As seen in many previous studies on turnout, I have also evaluated the significance of regional dummy variables (Models in the Reviewer s

10 670 J.W. Robbins / Electoral Studies 29 (2010) Appendix). Again, the findings are robust even in the presence of regional controls. Additional estimation techniques were also considered to determine if the results were an artifact of the test chosen. The first alternative method employed was to simulate a fixed effects model where a series of countryspecific indicators was added. After omitting those country variables that were insignificant, the results still hold in the presence of these controls. 11 Additionally, some have opted for another estimation procedure when the number of countries exceeds the time periods in a sample (see Fornos et al., 2004). Since this aptly describes the dataset in use, it was necessary to run a time-series analysis with a generalized least squares estimation (Liang and Zeger, 1986; Zeger and Liang, 1986). 12 Again, the results hold: the personal vote index is inversely related to turnout. Truncated regression and tobit estimations were also conducted but the results are essentially the same as those found in Tables 3 6. Therefore, based on these alternative specifications, the relationship between electoral systems and turnout levels is not sensitive to the method selected. Instead, these additional checks bolster the confidence that the incentives to cultivate a personal vote play an important role in predicting voter turnout levels. Finally, I have also tested the effects of presidentialism, total population, Polity2, and the natural log of district magnitude. These results are available in the Reviewer s Appendix (Models 17 24). Including these additional measures does not impact the main substantive results discussed above. Conclusion After reviewing several elections in a large sample of democratic states, the evidence affirms the importance of personal vote incentives. This study has included the personal vote database which is based on the index initially posited by Carey and Shugart (1995). The findings are intriguing for several reasons. The results suggest that while district magnitude is an important indicator of aggregate participation levels, so too is the incentive to cultivate a personal vote. Indeed, even when standard controls are included, the results continue to underscore the importance of electoral systems in predicting voter turnout. Furthermore, the evidence suggests that partycentered systems increase the need for parties to encourage turnout especially in comparison to candidatecentered systems. This finding is novel in that previous studies on aggregate turnout have yet to assess the relevance of other electoral system mechanisms. Therefore, this work has sought to move beyond dichotomous indicators of electoral systems. In doing so, we see that this significantly affects turnout levels. This study has attempted to utilize Carey and Shugart s (1995) personal vote index to predict voter turnout in legislative contests. Much of the work to date has applied this index to fiscal policy or other economic considerations. 11 These results are available upon request. 12 These results are available upon request. By linking the personal vote index to voter turnout, this study hopes to engender more work on the importance of electoral systems and other political institutions that are relevant in such discussions. Hence, without considering the full effects of electoral systems future works may be limited as they are unable to fully capture relevant electoral system aspects. Acknowledgements The author would like to thank Martin Edwards, Frank Thames, Harold Clarke and the anonymous reviewers for their helpful comments. Any remaining errors are the sole responsibility of the author. Appendix Country Parliamentary elections used Albania Argentina Australia Austria Belgium Bolivia Botswana Brazil Bulgaria Canada Chile Colombia Costa Rica Czech Republic Denmark Dominican Republic Ecuador El Salvador Estonia Finland France Germany Greece Honduras Hungary India Ireland Israel Italy Japan Latvia Lithuania Luxembourg Mexico Moldova Mozambique Namibia Netherlands New Zealand Norway Panama Paraguay Philippines Poland Portugal Romania Russia Slovakia Spain Sweden United Kingdom United States