- Email: [email protected]
Partisan turnout bias and district magnitude
Electoral Studies 35 (2014) 150e158
Contents lists available at ScienceDirect
Electoral Studies journal homepage: www.elsevier.com/locate/electstud
Partisan turnout bias and district magnitude* Ignacio Lago a, *, Marina Costa Lobo b a b
Universitat Pompeu Fabra, Department of Political and Social Sciences, Ramon Trias Fargas 25-27, 08005 Barcelona, Spain Instituto de Ci^ encias Sociais, Universidade de Lisboa, Av. Professor Aníbal de Bettencourt, 9, 1600-189 Lisboa, Portugal
a r t i c l e i n f o
a b s t r a c t
Article history: Received 19 September 2012 Received in revised form 10 March 2014 Accepted 3 June 2014 Available online 12 June 2014
This article examines how the partisan turnout bias (i.e. turnout rate differences across districts that are linked to the partisan vote shares in those districts) changes over time in PR districted electoral systems. We argue that the bias after the founding election is the unintended consequence of parties and voters' strategic behaviors when they respond to the incentives provided by the electoral system. By looking at the case of Portugal, one of the countries with the largest variation in district magnitude, we find that the increasing asymmetry in turnout rates across districts makes the bias more severe as time goes by. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Partisan turnout bias Electoral system District magnitude Strategic behavior
1. Introduction Electoral systems are not neutral institutions regulating the electoral competition among parties, but rules shaped by ruling parties to their own advantage. Even when parties are of the same size, electoral systems can create relative advantages for some parties depending on their ideology. Partisan biases refer to situations where some parties are able to win a larger share of the seats with a lesser (share of the) vote than other parties (Grofman et al., 1997: 457). Not surprisingly, electoral systems scholarship has focused much attention on the determinants and consequences of partisan biases, mainly in single-member, simple-plurality systems. In this paper we focus on the partisan turnout bias (i.e., turnout rate differences across districts that are linked to the partisan vote shares in those districts, in that certain -vis parties are more likely to have 'cheap seats' vis-a turnout, Grofman et al., 1997: 457) in countries using * The authors are grateful to the anonymous reviewers for comments on earlier versions. A preliminary version of this article was presented at the International Conference “The Effects of District Magnitude”, Instituto ^ncias Sociais, Universidade de Lisboa, Lisbon, May 2012. de Cie * Corresponding author. E-mail addresses: [email protected], [email protected] (I. Lago), [email protected] (M.C. Lobo).
http://dx.doi.org/10.1016/j.electstud.2014.06.001 0261-3794/© 2014 Elsevier Ltd. All rights reserved.
electoral systems with districts of varying magnitudes to explore how the bias changes over time. Our argument is that in proportional representation (PR) districted electoral systems in which districts are of different sizes, there are systematic differences across elections in the partisan turnout bias. Based on data from Portugal, a Third-Wave democracy with a large variation in district magnitude, we show that the bias is the unintended consequence of strategic behaviors by the parties which lost out by this bias. In addition, it is the case that those parties are mainly leftist (and small) parties. As leftist parties (and small parties) tend to do better in large (and urban) districts than in small (and rural) districts, they will progressively invest more heavily in the former once the distribution of partisan support is known after the founding election. Given that turnout tends to be higher in large districts than in small districts, the strategic behavior of leftist parties will make the gap greater. All else equal, the consequence is that the partisan turnout bias will be more severe as time goes by: seats will be cheaper in number of votes for rightist parties and more expensive for leftist parties. In sum, our argument hinges on the classical proposition that low turnout biases election outcomes such that right-wing parties gain at the expense of left-of-centre alternatives (DeNardo, 1980; see also the symposium about
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
the consequences of low turnout in Electoral Studies, 2007). Our findings have substantial implications for the quality of democracy in countries using PR districted electoral systems; we show that there is no “level playing field” for representatives of rightist and leftist parties when disputing an election. Moreover, when electoral system reform is discussed, this issue is seldom mentioned, thus making it likely that further reforms will not address it, or may even aggravate the problem. The article is organized as follows. In the next section, the argument addressing how the partisan turnout bias should evolve after the founding election in a PR districted electoral system is presented. The third section describes our case study, data, and methods. The fourth section discusses the results of the empirical analysis. We conclude with the presentation of our findings and a discussion of their implications. 2. Arguments Our basic story is that the partisan turnout bias in PR districted electoral systems is not constant, but changes over time as the unintended consequence of parties and voters' strategic behaviors when they respond to the incentives provided by the electoral system. The prequel of the argument is the conventional wisdom within electoral studies that turnout is higher in proportional representation systems and/or larger districts (see Blais and Aarts, 2006 for a review of the literature). Although the empirical evidence is robust, the two main mechanisms, proportionality and competitiveness, have weak or no effects both in cross-national and cross-district national analyses. For instance, using data from the Comparative Study of Electoral Systems (CSES), Fisher et al. (2008) find that turnout is lower under plurality rule than under PR because those with weaker motivations to vote are discouraged from voting. Similarly, relying on districtspecific national data from Switzerland and Spain, Grofman and Selb (2011) show that proportionality and competition do not increase with district magnitude. In sum, “most of the literature supports the view that PR fosters turnout, but there is no compelling explanation of how and why” (Blais, 2006: 116). In our view, however, the relationship between turnout and district magnitude becomes clearer if one employs a longitudinal analysis. Based on what Tavits and Annus (2006) define as the developmental argument of strategic behaviors, Gallego et al. (2012) show in a recent piece that the proportionality-turnout nexus in PR districted electoral systems is a long-term relationship connected with the psychological effect of electoral systems.1 First, parties will invest more heavily in those districts where their mobilization can make a difference. While the targets of mobilization for large parties can be in small and large districts depending on the existence of marginal seats, small parties
1 Their analysis of the learning process over time is focused specifically on Spain, one of the two countries included in the Grofman and Selb's paper.
151
have no incentive to campaign or even enter the race in small districts in which they are not competitive. Therefore, once actors have sufficient information about the relative chances of potential competitors, aggregate campaigning efforts will be more targeted at large districts as time goes by. Consequently, turnout differences across small and large districts will tend to increase after the founding election. Second, voters will not show up for less-competitive elections or waste their vote on parties that do not have a chance of winning. When voters are able to separate winners from losers in district-level races and they are sure who is ‘out of the running’, they face incentives to desert minor parties or, if they do not have a clear second preference, to abstain. All else equal, as this happens in small but not in large districts, differences in turnout between large and small districts will increase after the founding election. In sum, in electoral systems with districts of varying magnitude, such as Portugal, the asymmetry in turnout rates across districts should be higher when actors start to respond to the incentives provided by the electoral system.2 This is exactly what Fig. 1 shows. As can be seen, the correlation between turnout and district magnitude is positive both in the founding and the last election. That is, turnout is higher in large districts. However, the correlation is much stronger in the latter (0.57, statistically significant at the 0.01 level) than in the former (0.18, not statistically significant).3 The second element of our basic story is the asymmetry in the distribution of partisan voting strength across districts. This should not be a surprising correlation in PR districted electoral systems. As explained by Monroe and Rose (2002: 69), to the extent that high population areas are also high population density areas within the country, urban areas are represented by high-magnitude districts and rural areas by low-magnitude districts. If partisan sentiment tends to be correlated with urbanerural demographics, then partisan sentiment is also correlated with district magnitude. Roughly speaking, this means that the ideological left should tend to do better in large districts and the right in small districts.4 The empirical evidence from the fourteen Lower-House elections held in Portugal from 1975 to 2011 supports this argument. As can be seen in Fig. 2, in the founding election leftist parties (the PS and the PCP) did better in large districts and rightist parties (the PSD and the CDS) in small districts. However, as we will see later in detail, this pattern of distribution of partisan support across districts does not remain constant over time.
2 Grofman and Selb (2011) also rely on data from PR districted electoral systems in which district magnitude varies (Switzerland and Spain). However, as subnational parties are so strong in both countries, the relationship between turnout and proportionally and the number of parties is undermined considerably. 3 District magnitude goes from 4 to 55 in the 1975 election and from 2 to 47 in the 2011 election. 4 Urbanization is not the only feature that can be correlated with district magnitude. Demographics as employment or average wage, for instance, are not randomly distributed across districts (Monroe and Rose, 2002: 69).
152
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
Fig. 1. Relationship between turnout and district magnitude in the 1975 and 2011 elections in Portugal.
When there is a turnout-related effect on partisan bias in a founding election, the bias should favor rightist parties (or hinder leftist parties) as rightist parties tend to do better in small districts where turnout is lower. However, the slope of the bias will not be flat after the founding election. The intensity of the partisan turnout bias after the founding election will depend on the change in the asymmetry across districts in (i) the distribution of partisan voting strength and (ii) turnout rates. In principle, there are no reasons to expect great changes in the first dimension. However, as we have seen, the asymmetry in turnout rates should increase once the psychological effect of electoral systems comes into play. Minor (mainly leftist) parties will invest their resources in those districts where they have real electoral chances, the large ones, and (some) supporters of minor parties will decide to abstain in those districts where their favorite party does not have real electoral chances, the small ones. Consequently, both the turnout gap between large and small districts and the initial bias in favor of rightist parties will be exacerbated as time goes by. In sum, the partisan turnout-bias after the founding election in PR districted electoral systems emerges as an unintended consequence of political actors when they respond to the incentives provided by the electoral system. Paradoxically, when acting instrumentally to win more seats, leftist voters and party elites are acting against their own interests and in favor of rightist parties. The foregoing discussion generates two hypotheses about the slope of the turnout-related effect on partisan bias: Hypothesis 1. All else equal, the partisan turnout bias should increase as the asymmetry in the distribution of partisan voting strength across districts increases.
Hypothesis 2. All else equal, the pro-rightist bias in the founding election should rise over successive elections; increasingly negative for leftist parties and increasingly positive for rightist parties.
3. Data and methods The causal mechanisms explaining how (the variation in) district magnitude shapes the partisan turnout bias over time are examined through a quantitative analysis of the fourteen Lower-House elections held in Portugal from 1975 to 2011. Portugal fits the ideal institutional, social and political conditions for testing our argument.5 First, following the definitions of Monroe and Rose (2002), Portugal uses a pure ‘districted electoral system’ in which MPs6 are elected in twenty districts ranging from 2 to 47 seats (resident, 2011, see Lobo, 2009). In the sample of seventeen contemporary districted PR systems compiled by Monroe and Rose, Portugal has the highest district magnitude variation. Second, as can be seen in Fig. 3, district magnitude is strongly correlated with urbanization. The indicator for urbanization is the population density at the district level, with density measured as the number of persons per square kilometer.7 Correlations go from 0.9 in the 1975 and 1976 elections to 0.95 in the 2009 and 2011 election. Third,
5 Not surprisingly, Portugal was also selected by Monroe and Rose as their case study when analyzing the partisan political consequences of magnitude variation. 6 An electoral reform in 1991 in Portugal reduced the number of MPs from 250 to 230 thus marginally altering district magnitudes. This change in district magnitudes does not change our results appreciably. 7 The two non-resident districts have not been considered.
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
153
Fig. 2. Relationship between parties' vote shares and district magnitude in the 1975 (left side) and 2011 (right side) elections.
the Portuguese party system is highly nationalized8 (Lago and Montero, 2013) and quite stable and both factors make the empirical analysis relatively easy.9 Since the founding election in 1975, the Portuguese party system has consisted of two major political parties, the centre-left Socialist Party (PS) and the centre-right Social Democratic Party (PSD), and two minor parties, the Communist Portuguese Party (PCP-PEV), and the conservative alliance between the Social Democratic Center (CDS) and the Popular Party (PP) (Lobo and Magalhaes, 2011).10 The empirical analysis proceeds in two steps. The first step consists of showing that the correlation between district magnitude and turnout increases over time as a consequence of the psychological effect of the electoral system. The dependent variable is the level of turnout at the district level and the key independent variable is the (log) of district magnitude. The data source is the National Committee for Elections website (www.cne.pt). To show the increasing asymmetry in turnout rates across districts, several interactions are run between district magnitude and First election ea dummy variable that equals 0 for districts in the founding election and 1 for all remaining districtse and (the log of) Time Trend -a continuous variable than equals 1 for the 1975 election, 2 for the 1976 election … and 14 for the 2011 election) (model 5). The expected
8 Grofman and Selb (2011) rely on data from PR districted electoral systems in which district magnitude varies (Switzerland and Spain). However, as subnational parties are so strong in both countries, the relationship between turnout and proportionally and the number of parties is undermined considerably. 9 Although new viable parties have emerged since the founding election, mainly the Partido Renovador Democr atico (PRD) and the Bloco de Esquerda (BE), they have been presenting candidates in all districts. 10 The PS has not been considered in the 1980 election as its coalition with two minor leftist parties, the UEDS and the ASDI, makes the comparison between this election and the remaining not possible. Similarly, the PSD and the CDS have been excluded in the 1979 and 1980 elections given that they ran the elections together.
sign of both interactions is positive (i.e, the impact of district magnitude should increase as time goes by). The second stage of the analysis consists of showing that the partisan turnout bias depends on the change in the asymmetry across districts in (i) the distribution of partisan voting strength and (ii) turnout rates. To measure the turnout-related effects on partisan bias, we follow the formulation provided by Grofman et al. (1997). As some of the differences in turnout rates across districts may be due to malapportionment or the geographic distribution of partisan support, Grofman et al. use the difference between Ri and Mi as the measure of the aspect of partisan bias than can be taken to be purely turnout-related in nature: Ri is the party i's share of the total national raw vote, i.e., party i's share of the total votes won by party's candidates across all the districts and Mi, a malapportionment-corrected figure, P is Mi ¼ j ðpij dðjÞ Þ, where pij is the proportion of votes P won by party i in the jth district, i.e., pij ¼ vij = i vij , and d(j) is the ratio of raw population in the jth district to total raw national population. The difference between Ri and Mi captures the existence of the bias. If the difference is negative (i.e., Ri Mi < 0), the bias is positive as the party i wins its seats, on average, in districts with lower levels of turnout than other parties and then its seats are cheaper in votes, i.e., each seat requires less votes to be won than the average. If the difference is positive (i.e., Ri Mi > 0), the bias is negative as the seats of the party i in that district are more expensive in votes, i.e., each seat requires more votes to be won than the average. The two independent variables when explaining the determinants of the partisan turnout bias have been operationalized as follows: The correlation between district magnitude and partisan support across districts, Distribution of Support, has been defined as the difference between the share of the vote for a given party in Lisbon and its mean district
154
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
Fig. 3. Correlation between district magnitude and population density in Portugal*. *All the correlations are statistically significant at the 0.01 level.
share of the vote. If the value is positive, this means that the party does better in Lisbon than in the mean of the remaining districts and the other way around when the value is negative. Lisbon is the largest district in Portugal: around 21e22 percent of the total seats are allocated there and around one in five Portuguese people are entitled to vote in the district. The variable has positive values when a party draws its support mainly from urban areas, and negative values when a party draws its support mainly from rural areas. As can be seen in Table 1, the CDS is the party with the most homogeneous electoral results across districts: there are virtually no differences between its support in Lisbon and the rest of the country. On the other hand, the PSD is the party getting the best results in the small districts and the PCP in the largest district. Interestingly, the variable has both positive and negative values for three of the four parties. That is, the PS, the CDS and the PCP do better in urban than in rural districts in some elections, but not in others. The PSD is the only party that always does better in rural districts. Therefore, we expect that only for the PSD should the partisan turnout bias have the same sign across all elections. The correlation between district magnitude and turnout has been defined in three different ways to test the robustness of our findings: (1) the turnout coefficient in every election shown in Fig. 3; (2) the linear and unweighted correlation between district magnitude and turnout in every election; (3) in congruence with the first independent variable, the difference between the turnout in Lisbon and the mean district turnout. If the value is positive, this means that turnout is higher in Lisbon than in the mean of the remaining districts and the other way around when the value is negative.
4. Results A necessary condition for observing a change in the partisan turnout bias is that the correlation between district magnitude and turnout increases after the founding election. This condition is strongly supported by the results in Table 2. According to model 1, the (the log) of district magnitude is positive but not statistically significant when explaining turnout at the district level in the founding election. However, in the second election the variable becomes statistically significant at the 0.01 level (model 2). Table 1 Descriptive statistics. Variables
Mean
Std. Dev.
Max.
N
Turnout (log) District magnitude First election (log) Time trend
71.16 2.14 0.93 1.80
11.90 0.75 0.26 0.75
40.7 0.69 0 0
95.27 4.06 1 2.63
280 280 280 280
PS Bias Distribution of support
0.06 1.53
0.05 1.53
0.11 3.36
0.25 1.58
13 13
PSD Bias Distribution of support
0.17 7.34
0.09 2.74
0.33 14.98
0.04 5.49
12 12
CDS Bias Distribution of support
0.01 0.49
0.02 1.84
0.04 3.69
0.02 2.43
12 12
0.12 4.94 2.27 0.40 2.55
0.12 2.74 0.86 0.09 1.57
0.01 1.47 0.30 0.18 0.17
0.34 8.67 3.55 0.57 6.13
14 14 14 14 14
PCP Bias Distribution of support Turnout coefficient Cor (magnitude, turnout) Turnout in Lisbon
Min.
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
155
Table 2 The impact of district magnitude on turnout over time. Variables
(log)District magnitude First election (log)District magnitude* first election (log)Time trend (log)District magnitude* (log)time trend Constant R-squared N # Clusters
Models Model 1 (first elec.)a
Model 2 (after first elec.)
Model 3 (first-second elec.)
Model 4 (all elections)
Model 5 (all elections)
0.34 (0.61)
3.64*** (0.98)
0.34 (0.62) 13.08*** (1.84) 1.65** (0.74)
0.34 (0.60) 28.94*** (1.81) 3.29*** (0.76)
0.20 (0.90)
16.46*** (0.76) 1.18*** (0.29) 90.79*** (1.59) 0.02 20 e
61.84*** (2.60) 0.06 260 20
90.79*** (1.61) 0.74 40 20
90.79*** (1.55) 0.28 280 20
95.89*** (2.20) 0.81 280 20
Estimation is by ordinary least squares. Standard Errors in parentheses. *p < 0.1; **p < 0.05; ***p < 0.01. a Robust standard errors.
Models 3, 4 and 5 directly test whether the impact of district magnitude on turnout changes over time. The three interactions between (the log) of district magnitude and the dummy variable identifying districts in the founding election versus the second one (model 3) and versus all the remaining elections (model 4), and the time trend (model 5) are positive and statistically significant at the 0.05 level or better. The fit of the model significantly improves when the dummy variable is replaced with the time trend. In sum, the learning argument clearly works: the asymmetry in turnout rates across districts increase as time goes by.11 The results are disaggregated in Fig. 4 where the coefficient of (the log of) magnitude in every election is displayed.12 The increasing impact of district magnitude is already apparent in the second election. In general terms, the correlation between district magnitude and turnout tends to increase as time (elections) goes by: the coefficient is statistically significant in each of the last ten elections with the largest coefficient being found in the 2011 election. According to our argument, all else equal, if the correlation between district magnitude and turnout changes across elections, the partisan turnout bias should also change. The partisan biases in the Lower-House elections in Portugal due to turnout rate differences across districts are displayed in Fig. 5. The magnitude of the bias is not especially large for none of the four parties and is not statistically significant in any of the elections. This is not surprising as the number of observations is low e20 districtsd and the differences in turnout across districts are small (see Table 1).13 For our purpose the most relevant aspect is the slope of the bias, not the level. There are three interesting results in Fig. 5. First, given that in the founding election leftist parties did better in large districts and turnout was
11 The results do not change appreciably if the two largest districts, Lisbon and Porto, are excluded from the analysis. 12 Model 1 in Table 2 has been run for every election. 13 In the founding election, for instance, the average turnout in the 20 districts was 91.56 and the standard deviation only 1.83.
positively correlated with district magnitude, the rightist parties, the CDS and above all the PSD, are the beneficiaries of the bias in the founding election. Second, the bias is far from being constant across elections. The PSD is the only party whose bias always has the same sign, negative values and a positive bias, but show a huge variation; from a value of 0.33 in the 1983 election to 0.04 in the 2002 election. This variability is even higher for the two leftist parties. The PS goes from 0.24 in the 1983 election to 0.23 in the 1991 election and the PCP goes from 0.34 in the 1979 election to 0.01 in the 2002 election. According to our second hypotheses, the initial pro-rightist parties bias should be stronger over time (elections) if the distribution of partisan support across districts does not change. However, as said before, the correlation between parties' vote shares and district magnitude is not constant and even has different signs as time goes by for three of the four parties dthe PSD is the exception. Finally, not surprisingly, for the party with the most homogeneous electoral results across districts, the CDS, the bias is virtually nonexistent; while the parties with the most urban and rural vote, the PCP and the PSD, have the greatest positive and negative values, respectively. In order to test our hypotheses, we have run three specifications in which the partisan turnout bias for every party and election is regressed on different operationalizations of the correlation between turnout and district magnitude, while controlling for the distribution of partisan support. As the slope of the partisan turnout bias should be different depending on the ideology of parties, the models are run separately for rightist and leftist parties. The results of the OLS estimates are displayed in Table 3. Although the observations should be clustered into parties, a regression with robust standard errors instead of a regression with clusters has been run. In that case the degrees of freedom would be reduced to reflect the number of independent groups included in the estimation and then the number of independent observations (degrees of freedom) would only be two in the estimations. Remember that, according to the formulation provided by Grofman et al. (1997) to measure the bias, positive coefficients mean that the bias is negative (i.e., seats are more expensive in votes).
156
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
Fig. 4. How (log) district magnitude affects turnout in Portugal.
Our two hypotheses are strongly supported. Both for rightist and leftist parties, the variable Distribution of Support is positive and statistically significant at the 0.01 level in the six models: the partisan turnout bias is increasingly negative as the correlation between their electoral results
and district magnitude increases. That is, when parties do better in large districts emainly Lisbone than in small districts, differences in turnout create disadvantages for them. Similarly, the impact of the correlation between turnout and district magnitude is statistically significant at the 0.05 level or better in all the models and has the expected signs: negative for rightist parties and positive for leftist parties. In other words, given that the bias was for rightist parties in the founding election, it is increasingly favorable for rightist parties and unfavorable for leftist parties as elections go by eremember that the impact of district magnitude on turnout is bigger over time. The model's fit is very good for such a parsimonious model. To explore the robustness of our findings, in Table 4 the models are clustered by party in order to correct for the non-independence of the data structure. The difference with the previous analysis is that now we are not separating the four parties according to their ideology. As the rightist and leftist parties are taken into account at the same time, the crucial parameter is the interaction between the distribution of partisan support and the different measures of the correlation between district magnitude and turnout: only when parties do better in large districts and turnout is positively correlated with district magnitude should we find a positive and statistically significant coefficient (i.e., a negative bias). This is exactly what can be observed in Table 4. In the three models, the interaction is statistically significant at the 0.05 level. 5. Conclusions
Fig. 5. Partisan turnout bias in Portugal. *Negative values mean that parties are the beneficiaries of the bias.
Partisan turnout bias depends on the asymmetry in the distribution of partisan support and turnout rates across districts. When a party does systematically better in
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
157
Table 3 The sources of partisan turnout bias in Portugal (I). Independent variables
Rightist parties
Distribution of support Turnout coefficient Cor (magnitude, turnout) Turnout in Lisbon Constant R squared N
Leftist parties
1
2
3
4
5
6
0.023*** (0.003) 0.056*** (0.016)
0.023*** (0.003)
0.022*** (0.003)
0.027*** (0.007) 0.082*** (0.027)
0.028*** (0.008)
0.027*** (0.008)
0.433*** (0.113) 0.137*** (0.044) 0.75 24
0.177** (0.049) 0.70 24
0.771*** (0.269) 0.031*** (0.007) 0.081*** (0.026) 0.78 24
0.188** (0.084) 0.46 27
0.312** (0.133) 0.43 27
0.035** (0.015) 0.075 (0.063) 0.40 27
***p < 0.01, **p < 0.05. Estimation is by OLS. Robust standard errors are given in parentheses.
districts with lower turnout, it will be the beneficiary of the bias: it will need fewer votes to pick up the seats than another party doing better in districts with higher turnout. In PR districted electoral systems in which districts are of different sizes, the partisan turnout bias in the founding election is not infrequent. As turnout tends to be higher in large districts and district magnitude is almost always correlated with urbanization, turnout rate differences across districts are linked to the partisan vote shares in those districts: the bias should favor rightist parties (or hinder leftist parties) as rightist parties tend to do better in small districts. In this article, we have used data from Portugal to show that the initial partisan turnout bias is exacerbated across elections once the psychological effect of the electoral system comes into play. Once actors have sufficient information about the relative chances of potential competitors, aggregate campaigning efforts will be more targeted at large districts, while small leftist parties will do worse in small districts and better in large district in comparison with the founding election. The empirical analysis proceeded in two steps. First, we demonstrated the increasing impact of district magnitude variation on turnout. The strongest effect was found in the last election considered, in 2011, and the weakest in the
Table 4 The sources of partisan turnout bias in Portugal (II).
References
Independent variables
1
2
3
Distribution of support
0.008 (0.03) 0.0133 (0.025)
0.003 (0.005)
0.010** (0.003)
Turnout coefficient Cor (magnitude, turnout) Turnout in Lisbon Distribution*turnout coefficient Distribution*Cor (Mag., Turn.) Distribution*turnout in Lisbon Constant R squared N # Clusters
0.132 (0.236) 0.005 (0.011) 0.008** (0.002) 0.073** (0.015) 0.006** (0.001) 0.020 (0.052) 0.71 51 4
**p < 0.05. Estimation is by OLS.
founding election. As expected, actors' strategic behavior leads to an increasing asymmetry as time goes by in turnout rates across districts. Second, the partisan turnout bias was measured. Our hypotheses were strongly supported. The partisan turnout bias after the founding election depends on the change in the asymmetry across districts in (i) the distribution of partisan voting strength and (ii) turnout rates. Finally, our analysis of how PR districted electoral systems generates a partisan turnout bias opens up two important research questions. On the one hand, given that asymmetries in turnout rates across districts are crucial, we expect that the slope of the bias will be greater when variance in district magnitude is high. The higher the variance in district magnitude, the greater the difference the psychological effect will be across districts. A comparison between Portugal and Luxembourg, for instance, would be enlightening. On the other hand, in PR districted electoral systems with a strong regional cleavage (Spain, for instance) the relationship between district magnitude and turnout is undermined. Therefore, when there are regional or local parties the partisan turnout bias should not be particularly important. In sum, there are four possible scenarios for testing our argument when crossing these two variables (variance and regionalization/nationalization). Portugal represents the ideal country for observing the greatest partisan turnout bias.
0.043 (0.087)
0.002 (0.022)
0.70 51 4
0.73 51 4
Blais, A., 2006. What affects turnout? Annu. Rev. Polit. Sci. 9, 111e125. Blais, A., Aarts, K., 2006. Electoral systems and turnout. Acta Polit. 41, 180e196. DeNardo, J., 1980. Turnout and the vote: the Joke's on the Democrats. Am. Polit. Sci. Rev. 74, 406e420. Electoral Studies, 2007. Special symposium: consequences of low turnout. Elect. Stud. 26, 539e632. Fisher, S.D., Lessard-Phillips, L., Hobolt, S.B., Curtice, J., 2008. Disengaging voters: do plurality systems discourage the less knowledgeable from voting? Elect. Stud. 27, 89e104. Gallego, A., Rico, G., Anduiza, E., 2012. Disproportionality and voter turnout in old and new democracies. Elect. Stud. 31, 159e169. Grofman, B., Koetzle, W., Burnell, T., 1997. An integrated perspective on the three potential sources of partisan bias: malapportionment, turnout differences, and the geographic distribution of party vote shares. Elect. Stud. 16, 457e470. Grofman, B., Selb, P., 2011. Turnout and the (effective) number of parties at the national and district levels: a puzzle-solving approach. Party Polit. 17, 93e117. Lago, I., Montero, J.R., 2013. Defining and measuring party system nationalization. Eur. Polit. Sci. Rev. http://dx.doi.org/10.1017/ S1755773913000027.
158
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
~o. In: Marina, C. Lobo, Lobo, M.C., 2009. Um sistema partid ario sob pressa ~es, Pedro (Eds.), As eleiço ~ es legislativas e presidenciais, Magalha 2005e2006. ICS, Lisbon, pp. 9e20. Lobo, M.C., Magalhaes, P.C., 2011. Room for Manoeuvre: euroscepticism in the Portuguese parties and electorate 1976e2005. South Eur. Soc. Polit. 16, 81e104.
Monroe, B.L., Rose, A.G., 2002. Electoral systems and unimagined consequences. Partisan effects of districted proportional representation. Am. J. Polit. Sci. 46, 67e89. Tavits, M., Annus, T., 2006. Learning to make votes count: the role of democratic experience. Elect. Stud. 25, 72e90.
Contents lists available at ScienceDirect
Electoral Studies journal homepage: www.elsevier.com/locate/electstud
Partisan turnout bias and district magnitude* Ignacio Lago a, *, Marina Costa Lobo b a b
Universitat Pompeu Fabra, Department of Political and Social Sciences, Ramon Trias Fargas 25-27, 08005 Barcelona, Spain Instituto de Ci^ encias Sociais, Universidade de Lisboa, Av. Professor Aníbal de Bettencourt, 9, 1600-189 Lisboa, Portugal
a r t i c l e i n f o
a b s t r a c t
Article history: Received 19 September 2012 Received in revised form 10 March 2014 Accepted 3 June 2014 Available online 12 June 2014
This article examines how the partisan turnout bias (i.e. turnout rate differences across districts that are linked to the partisan vote shares in those districts) changes over time in PR districted electoral systems. We argue that the bias after the founding election is the unintended consequence of parties and voters' strategic behaviors when they respond to the incentives provided by the electoral system. By looking at the case of Portugal, one of the countries with the largest variation in district magnitude, we find that the increasing asymmetry in turnout rates across districts makes the bias more severe as time goes by. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Partisan turnout bias Electoral system District magnitude Strategic behavior
1. Introduction Electoral systems are not neutral institutions regulating the electoral competition among parties, but rules shaped by ruling parties to their own advantage. Even when parties are of the same size, electoral systems can create relative advantages for some parties depending on their ideology. Partisan biases refer to situations where some parties are able to win a larger share of the seats with a lesser (share of the) vote than other parties (Grofman et al., 1997: 457). Not surprisingly, electoral systems scholarship has focused much attention on the determinants and consequences of partisan biases, mainly in single-member, simple-plurality systems. In this paper we focus on the partisan turnout bias (i.e., turnout rate differences across districts that are linked to the partisan vote shares in those districts, in that certain -vis parties are more likely to have 'cheap seats' vis-a turnout, Grofman et al., 1997: 457) in countries using * The authors are grateful to the anonymous reviewers for comments on earlier versions. A preliminary version of this article was presented at the International Conference “The Effects of District Magnitude”, Instituto ^ncias Sociais, Universidade de Lisboa, Lisbon, May 2012. de Cie * Corresponding author. E-mail addresses: [email protected], [email protected] (I. Lago), [email protected] (M.C. Lobo).
http://dx.doi.org/10.1016/j.electstud.2014.06.001 0261-3794/© 2014 Elsevier Ltd. All rights reserved.
electoral systems with districts of varying magnitudes to explore how the bias changes over time. Our argument is that in proportional representation (PR) districted electoral systems in which districts are of different sizes, there are systematic differences across elections in the partisan turnout bias. Based on data from Portugal, a Third-Wave democracy with a large variation in district magnitude, we show that the bias is the unintended consequence of strategic behaviors by the parties which lost out by this bias. In addition, it is the case that those parties are mainly leftist (and small) parties. As leftist parties (and small parties) tend to do better in large (and urban) districts than in small (and rural) districts, they will progressively invest more heavily in the former once the distribution of partisan support is known after the founding election. Given that turnout tends to be higher in large districts than in small districts, the strategic behavior of leftist parties will make the gap greater. All else equal, the consequence is that the partisan turnout bias will be more severe as time goes by: seats will be cheaper in number of votes for rightist parties and more expensive for leftist parties. In sum, our argument hinges on the classical proposition that low turnout biases election outcomes such that right-wing parties gain at the expense of left-of-centre alternatives (DeNardo, 1980; see also the symposium about
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
the consequences of low turnout in Electoral Studies, 2007). Our findings have substantial implications for the quality of democracy in countries using PR districted electoral systems; we show that there is no “level playing field” for representatives of rightist and leftist parties when disputing an election. Moreover, when electoral system reform is discussed, this issue is seldom mentioned, thus making it likely that further reforms will not address it, or may even aggravate the problem. The article is organized as follows. In the next section, the argument addressing how the partisan turnout bias should evolve after the founding election in a PR districted electoral system is presented. The third section describes our case study, data, and methods. The fourth section discusses the results of the empirical analysis. We conclude with the presentation of our findings and a discussion of their implications. 2. Arguments Our basic story is that the partisan turnout bias in PR districted electoral systems is not constant, but changes over time as the unintended consequence of parties and voters' strategic behaviors when they respond to the incentives provided by the electoral system. The prequel of the argument is the conventional wisdom within electoral studies that turnout is higher in proportional representation systems and/or larger districts (see Blais and Aarts, 2006 for a review of the literature). Although the empirical evidence is robust, the two main mechanisms, proportionality and competitiveness, have weak or no effects both in cross-national and cross-district national analyses. For instance, using data from the Comparative Study of Electoral Systems (CSES), Fisher et al. (2008) find that turnout is lower under plurality rule than under PR because those with weaker motivations to vote are discouraged from voting. Similarly, relying on districtspecific national data from Switzerland and Spain, Grofman and Selb (2011) show that proportionality and competition do not increase with district magnitude. In sum, “most of the literature supports the view that PR fosters turnout, but there is no compelling explanation of how and why” (Blais, 2006: 116). In our view, however, the relationship between turnout and district magnitude becomes clearer if one employs a longitudinal analysis. Based on what Tavits and Annus (2006) define as the developmental argument of strategic behaviors, Gallego et al. (2012) show in a recent piece that the proportionality-turnout nexus in PR districted electoral systems is a long-term relationship connected with the psychological effect of electoral systems.1 First, parties will invest more heavily in those districts where their mobilization can make a difference. While the targets of mobilization for large parties can be in small and large districts depending on the existence of marginal seats, small parties
1 Their analysis of the learning process over time is focused specifically on Spain, one of the two countries included in the Grofman and Selb's paper.
151
have no incentive to campaign or even enter the race in small districts in which they are not competitive. Therefore, once actors have sufficient information about the relative chances of potential competitors, aggregate campaigning efforts will be more targeted at large districts as time goes by. Consequently, turnout differences across small and large districts will tend to increase after the founding election. Second, voters will not show up for less-competitive elections or waste their vote on parties that do not have a chance of winning. When voters are able to separate winners from losers in district-level races and they are sure who is ‘out of the running’, they face incentives to desert minor parties or, if they do not have a clear second preference, to abstain. All else equal, as this happens in small but not in large districts, differences in turnout between large and small districts will increase after the founding election. In sum, in electoral systems with districts of varying magnitude, such as Portugal, the asymmetry in turnout rates across districts should be higher when actors start to respond to the incentives provided by the electoral system.2 This is exactly what Fig. 1 shows. As can be seen, the correlation between turnout and district magnitude is positive both in the founding and the last election. That is, turnout is higher in large districts. However, the correlation is much stronger in the latter (0.57, statistically significant at the 0.01 level) than in the former (0.18, not statistically significant).3 The second element of our basic story is the asymmetry in the distribution of partisan voting strength across districts. This should not be a surprising correlation in PR districted electoral systems. As explained by Monroe and Rose (2002: 69), to the extent that high population areas are also high population density areas within the country, urban areas are represented by high-magnitude districts and rural areas by low-magnitude districts. If partisan sentiment tends to be correlated with urbanerural demographics, then partisan sentiment is also correlated with district magnitude. Roughly speaking, this means that the ideological left should tend to do better in large districts and the right in small districts.4 The empirical evidence from the fourteen Lower-House elections held in Portugal from 1975 to 2011 supports this argument. As can be seen in Fig. 2, in the founding election leftist parties (the PS and the PCP) did better in large districts and rightist parties (the PSD and the CDS) in small districts. However, as we will see later in detail, this pattern of distribution of partisan support across districts does not remain constant over time.
2 Grofman and Selb (2011) also rely on data from PR districted electoral systems in which district magnitude varies (Switzerland and Spain). However, as subnational parties are so strong in both countries, the relationship between turnout and proportionally and the number of parties is undermined considerably. 3 District magnitude goes from 4 to 55 in the 1975 election and from 2 to 47 in the 2011 election. 4 Urbanization is not the only feature that can be correlated with district magnitude. Demographics as employment or average wage, for instance, are not randomly distributed across districts (Monroe and Rose, 2002: 69).
152
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
Fig. 1. Relationship between turnout and district magnitude in the 1975 and 2011 elections in Portugal.
When there is a turnout-related effect on partisan bias in a founding election, the bias should favor rightist parties (or hinder leftist parties) as rightist parties tend to do better in small districts where turnout is lower. However, the slope of the bias will not be flat after the founding election. The intensity of the partisan turnout bias after the founding election will depend on the change in the asymmetry across districts in (i) the distribution of partisan voting strength and (ii) turnout rates. In principle, there are no reasons to expect great changes in the first dimension. However, as we have seen, the asymmetry in turnout rates should increase once the psychological effect of electoral systems comes into play. Minor (mainly leftist) parties will invest their resources in those districts where they have real electoral chances, the large ones, and (some) supporters of minor parties will decide to abstain in those districts where their favorite party does not have real electoral chances, the small ones. Consequently, both the turnout gap between large and small districts and the initial bias in favor of rightist parties will be exacerbated as time goes by. In sum, the partisan turnout-bias after the founding election in PR districted electoral systems emerges as an unintended consequence of political actors when they respond to the incentives provided by the electoral system. Paradoxically, when acting instrumentally to win more seats, leftist voters and party elites are acting against their own interests and in favor of rightist parties. The foregoing discussion generates two hypotheses about the slope of the turnout-related effect on partisan bias: Hypothesis 1. All else equal, the partisan turnout bias should increase as the asymmetry in the distribution of partisan voting strength across districts increases.
Hypothesis 2. All else equal, the pro-rightist bias in the founding election should rise over successive elections; increasingly negative for leftist parties and increasingly positive for rightist parties.
3. Data and methods The causal mechanisms explaining how (the variation in) district magnitude shapes the partisan turnout bias over time are examined through a quantitative analysis of the fourteen Lower-House elections held in Portugal from 1975 to 2011. Portugal fits the ideal institutional, social and political conditions for testing our argument.5 First, following the definitions of Monroe and Rose (2002), Portugal uses a pure ‘districted electoral system’ in which MPs6 are elected in twenty districts ranging from 2 to 47 seats (resident, 2011, see Lobo, 2009). In the sample of seventeen contemporary districted PR systems compiled by Monroe and Rose, Portugal has the highest district magnitude variation. Second, as can be seen in Fig. 3, district magnitude is strongly correlated with urbanization. The indicator for urbanization is the population density at the district level, with density measured as the number of persons per square kilometer.7 Correlations go from 0.9 in the 1975 and 1976 elections to 0.95 in the 2009 and 2011 election. Third,
5 Not surprisingly, Portugal was also selected by Monroe and Rose as their case study when analyzing the partisan political consequences of magnitude variation. 6 An electoral reform in 1991 in Portugal reduced the number of MPs from 250 to 230 thus marginally altering district magnitudes. This change in district magnitudes does not change our results appreciably. 7 The two non-resident districts have not been considered.
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
153
Fig. 2. Relationship between parties' vote shares and district magnitude in the 1975 (left side) and 2011 (right side) elections.
the Portuguese party system is highly nationalized8 (Lago and Montero, 2013) and quite stable and both factors make the empirical analysis relatively easy.9 Since the founding election in 1975, the Portuguese party system has consisted of two major political parties, the centre-left Socialist Party (PS) and the centre-right Social Democratic Party (PSD), and two minor parties, the Communist Portuguese Party (PCP-PEV), and the conservative alliance between the Social Democratic Center (CDS) and the Popular Party (PP) (Lobo and Magalhaes, 2011).10 The empirical analysis proceeds in two steps. The first step consists of showing that the correlation between district magnitude and turnout increases over time as a consequence of the psychological effect of the electoral system. The dependent variable is the level of turnout at the district level and the key independent variable is the (log) of district magnitude. The data source is the National Committee for Elections website (www.cne.pt). To show the increasing asymmetry in turnout rates across districts, several interactions are run between district magnitude and First election ea dummy variable that equals 0 for districts in the founding election and 1 for all remaining districtse and (the log of) Time Trend -a continuous variable than equals 1 for the 1975 election, 2 for the 1976 election … and 14 for the 2011 election) (model 5). The expected
8 Grofman and Selb (2011) rely on data from PR districted electoral systems in which district magnitude varies (Switzerland and Spain). However, as subnational parties are so strong in both countries, the relationship between turnout and proportionally and the number of parties is undermined considerably. 9 Although new viable parties have emerged since the founding election, mainly the Partido Renovador Democr atico (PRD) and the Bloco de Esquerda (BE), they have been presenting candidates in all districts. 10 The PS has not been considered in the 1980 election as its coalition with two minor leftist parties, the UEDS and the ASDI, makes the comparison between this election and the remaining not possible. Similarly, the PSD and the CDS have been excluded in the 1979 and 1980 elections given that they ran the elections together.
sign of both interactions is positive (i.e, the impact of district magnitude should increase as time goes by). The second stage of the analysis consists of showing that the partisan turnout bias depends on the change in the asymmetry across districts in (i) the distribution of partisan voting strength and (ii) turnout rates. To measure the turnout-related effects on partisan bias, we follow the formulation provided by Grofman et al. (1997). As some of the differences in turnout rates across districts may be due to malapportionment or the geographic distribution of partisan support, Grofman et al. use the difference between Ri and Mi as the measure of the aspect of partisan bias than can be taken to be purely turnout-related in nature: Ri is the party i's share of the total national raw vote, i.e., party i's share of the total votes won by party's candidates across all the districts and Mi, a malapportionment-corrected figure, P is Mi ¼ j ðpij dðjÞ Þ, where pij is the proportion of votes P won by party i in the jth district, i.e., pij ¼ vij = i vij , and d(j) is the ratio of raw population in the jth district to total raw national population. The difference between Ri and Mi captures the existence of the bias. If the difference is negative (i.e., Ri Mi < 0), the bias is positive as the party i wins its seats, on average, in districts with lower levels of turnout than other parties and then its seats are cheaper in votes, i.e., each seat requires less votes to be won than the average. If the difference is positive (i.e., Ri Mi > 0), the bias is negative as the seats of the party i in that district are more expensive in votes, i.e., each seat requires more votes to be won than the average. The two independent variables when explaining the determinants of the partisan turnout bias have been operationalized as follows: The correlation between district magnitude and partisan support across districts, Distribution of Support, has been defined as the difference between the share of the vote for a given party in Lisbon and its mean district
154
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
Fig. 3. Correlation between district magnitude and population density in Portugal*. *All the correlations are statistically significant at the 0.01 level.
share of the vote. If the value is positive, this means that the party does better in Lisbon than in the mean of the remaining districts and the other way around when the value is negative. Lisbon is the largest district in Portugal: around 21e22 percent of the total seats are allocated there and around one in five Portuguese people are entitled to vote in the district. The variable has positive values when a party draws its support mainly from urban areas, and negative values when a party draws its support mainly from rural areas. As can be seen in Table 1, the CDS is the party with the most homogeneous electoral results across districts: there are virtually no differences between its support in Lisbon and the rest of the country. On the other hand, the PSD is the party getting the best results in the small districts and the PCP in the largest district. Interestingly, the variable has both positive and negative values for three of the four parties. That is, the PS, the CDS and the PCP do better in urban than in rural districts in some elections, but not in others. The PSD is the only party that always does better in rural districts. Therefore, we expect that only for the PSD should the partisan turnout bias have the same sign across all elections. The correlation between district magnitude and turnout has been defined in three different ways to test the robustness of our findings: (1) the turnout coefficient in every election shown in Fig. 3; (2) the linear and unweighted correlation between district magnitude and turnout in every election; (3) in congruence with the first independent variable, the difference between the turnout in Lisbon and the mean district turnout. If the value is positive, this means that turnout is higher in Lisbon than in the mean of the remaining districts and the other way around when the value is negative.
4. Results A necessary condition for observing a change in the partisan turnout bias is that the correlation between district magnitude and turnout increases after the founding election. This condition is strongly supported by the results in Table 2. According to model 1, the (the log) of district magnitude is positive but not statistically significant when explaining turnout at the district level in the founding election. However, in the second election the variable becomes statistically significant at the 0.01 level (model 2). Table 1 Descriptive statistics. Variables
Mean
Std. Dev.
Max.
N
Turnout (log) District magnitude First election (log) Time trend
71.16 2.14 0.93 1.80
11.90 0.75 0.26 0.75
40.7 0.69 0 0
95.27 4.06 1 2.63
280 280 280 280
PS Bias Distribution of support
0.06 1.53
0.05 1.53
0.11 3.36
0.25 1.58
13 13
PSD Bias Distribution of support
0.17 7.34
0.09 2.74
0.33 14.98
0.04 5.49
12 12
CDS Bias Distribution of support
0.01 0.49
0.02 1.84
0.04 3.69
0.02 2.43
12 12
0.12 4.94 2.27 0.40 2.55
0.12 2.74 0.86 0.09 1.57
0.01 1.47 0.30 0.18 0.17
0.34 8.67 3.55 0.57 6.13
14 14 14 14 14
PCP Bias Distribution of support Turnout coefficient Cor (magnitude, turnout) Turnout in Lisbon
Min.
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
155
Table 2 The impact of district magnitude on turnout over time. Variables
(log)District magnitude First election (log)District magnitude* first election (log)Time trend (log)District magnitude* (log)time trend Constant R-squared N # Clusters
Models Model 1 (first elec.)a
Model 2 (after first elec.)
Model 3 (first-second elec.)
Model 4 (all elections)
Model 5 (all elections)
0.34 (0.61)
3.64*** (0.98)
0.34 (0.62) 13.08*** (1.84) 1.65** (0.74)
0.34 (0.60) 28.94*** (1.81) 3.29*** (0.76)
0.20 (0.90)
16.46*** (0.76) 1.18*** (0.29) 90.79*** (1.59) 0.02 20 e
61.84*** (2.60) 0.06 260 20
90.79*** (1.61) 0.74 40 20
90.79*** (1.55) 0.28 280 20
95.89*** (2.20) 0.81 280 20
Estimation is by ordinary least squares. Standard Errors in parentheses. *p < 0.1; **p < 0.05; ***p < 0.01. a Robust standard errors.
Models 3, 4 and 5 directly test whether the impact of district magnitude on turnout changes over time. The three interactions between (the log) of district magnitude and the dummy variable identifying districts in the founding election versus the second one (model 3) and versus all the remaining elections (model 4), and the time trend (model 5) are positive and statistically significant at the 0.05 level or better. The fit of the model significantly improves when the dummy variable is replaced with the time trend. In sum, the learning argument clearly works: the asymmetry in turnout rates across districts increase as time goes by.11 The results are disaggregated in Fig. 4 where the coefficient of (the log of) magnitude in every election is displayed.12 The increasing impact of district magnitude is already apparent in the second election. In general terms, the correlation between district magnitude and turnout tends to increase as time (elections) goes by: the coefficient is statistically significant in each of the last ten elections with the largest coefficient being found in the 2011 election. According to our argument, all else equal, if the correlation between district magnitude and turnout changes across elections, the partisan turnout bias should also change. The partisan biases in the Lower-House elections in Portugal due to turnout rate differences across districts are displayed in Fig. 5. The magnitude of the bias is not especially large for none of the four parties and is not statistically significant in any of the elections. This is not surprising as the number of observations is low e20 districtsd and the differences in turnout across districts are small (see Table 1).13 For our purpose the most relevant aspect is the slope of the bias, not the level. There are three interesting results in Fig. 5. First, given that in the founding election leftist parties did better in large districts and turnout was
11 The results do not change appreciably if the two largest districts, Lisbon and Porto, are excluded from the analysis. 12 Model 1 in Table 2 has been run for every election. 13 In the founding election, for instance, the average turnout in the 20 districts was 91.56 and the standard deviation only 1.83.
positively correlated with district magnitude, the rightist parties, the CDS and above all the PSD, are the beneficiaries of the bias in the founding election. Second, the bias is far from being constant across elections. The PSD is the only party whose bias always has the same sign, negative values and a positive bias, but show a huge variation; from a value of 0.33 in the 1983 election to 0.04 in the 2002 election. This variability is even higher for the two leftist parties. The PS goes from 0.24 in the 1983 election to 0.23 in the 1991 election and the PCP goes from 0.34 in the 1979 election to 0.01 in the 2002 election. According to our second hypotheses, the initial pro-rightist parties bias should be stronger over time (elections) if the distribution of partisan support across districts does not change. However, as said before, the correlation between parties' vote shares and district magnitude is not constant and even has different signs as time goes by for three of the four parties dthe PSD is the exception. Finally, not surprisingly, for the party with the most homogeneous electoral results across districts, the CDS, the bias is virtually nonexistent; while the parties with the most urban and rural vote, the PCP and the PSD, have the greatest positive and negative values, respectively. In order to test our hypotheses, we have run three specifications in which the partisan turnout bias for every party and election is regressed on different operationalizations of the correlation between turnout and district magnitude, while controlling for the distribution of partisan support. As the slope of the partisan turnout bias should be different depending on the ideology of parties, the models are run separately for rightist and leftist parties. The results of the OLS estimates are displayed in Table 3. Although the observations should be clustered into parties, a regression with robust standard errors instead of a regression with clusters has been run. In that case the degrees of freedom would be reduced to reflect the number of independent groups included in the estimation and then the number of independent observations (degrees of freedom) would only be two in the estimations. Remember that, according to the formulation provided by Grofman et al. (1997) to measure the bias, positive coefficients mean that the bias is negative (i.e., seats are more expensive in votes).
156
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
Fig. 4. How (log) district magnitude affects turnout in Portugal.
Our two hypotheses are strongly supported. Both for rightist and leftist parties, the variable Distribution of Support is positive and statistically significant at the 0.01 level in the six models: the partisan turnout bias is increasingly negative as the correlation between their electoral results
and district magnitude increases. That is, when parties do better in large districts emainly Lisbone than in small districts, differences in turnout create disadvantages for them. Similarly, the impact of the correlation between turnout and district magnitude is statistically significant at the 0.05 level or better in all the models and has the expected signs: negative for rightist parties and positive for leftist parties. In other words, given that the bias was for rightist parties in the founding election, it is increasingly favorable for rightist parties and unfavorable for leftist parties as elections go by eremember that the impact of district magnitude on turnout is bigger over time. The model's fit is very good for such a parsimonious model. To explore the robustness of our findings, in Table 4 the models are clustered by party in order to correct for the non-independence of the data structure. The difference with the previous analysis is that now we are not separating the four parties according to their ideology. As the rightist and leftist parties are taken into account at the same time, the crucial parameter is the interaction between the distribution of partisan support and the different measures of the correlation between district magnitude and turnout: only when parties do better in large districts and turnout is positively correlated with district magnitude should we find a positive and statistically significant coefficient (i.e., a negative bias). This is exactly what can be observed in Table 4. In the three models, the interaction is statistically significant at the 0.05 level. 5. Conclusions
Fig. 5. Partisan turnout bias in Portugal. *Negative values mean that parties are the beneficiaries of the bias.
Partisan turnout bias depends on the asymmetry in the distribution of partisan support and turnout rates across districts. When a party does systematically better in
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
157
Table 3 The sources of partisan turnout bias in Portugal (I). Independent variables
Rightist parties
Distribution of support Turnout coefficient Cor (magnitude, turnout) Turnout in Lisbon Constant R squared N
Leftist parties
1
2
3
4
5
6
0.023*** (0.003) 0.056*** (0.016)
0.023*** (0.003)
0.022*** (0.003)
0.027*** (0.007) 0.082*** (0.027)
0.028*** (0.008)
0.027*** (0.008)
0.433*** (0.113) 0.137*** (0.044) 0.75 24
0.177** (0.049) 0.70 24
0.771*** (0.269) 0.031*** (0.007) 0.081*** (0.026) 0.78 24
0.188** (0.084) 0.46 27
0.312** (0.133) 0.43 27
0.035** (0.015) 0.075 (0.063) 0.40 27
***p < 0.01, **p < 0.05. Estimation is by OLS. Robust standard errors are given in parentheses.
districts with lower turnout, it will be the beneficiary of the bias: it will need fewer votes to pick up the seats than another party doing better in districts with higher turnout. In PR districted electoral systems in which districts are of different sizes, the partisan turnout bias in the founding election is not infrequent. As turnout tends to be higher in large districts and district magnitude is almost always correlated with urbanization, turnout rate differences across districts are linked to the partisan vote shares in those districts: the bias should favor rightist parties (or hinder leftist parties) as rightist parties tend to do better in small districts. In this article, we have used data from Portugal to show that the initial partisan turnout bias is exacerbated across elections once the psychological effect of the electoral system comes into play. Once actors have sufficient information about the relative chances of potential competitors, aggregate campaigning efforts will be more targeted at large districts, while small leftist parties will do worse in small districts and better in large district in comparison with the founding election. The empirical analysis proceeded in two steps. First, we demonstrated the increasing impact of district magnitude variation on turnout. The strongest effect was found in the last election considered, in 2011, and the weakest in the
Table 4 The sources of partisan turnout bias in Portugal (II).
References
Independent variables
1
2
3
Distribution of support
0.008 (0.03) 0.0133 (0.025)
0.003 (0.005)
0.010** (0.003)
Turnout coefficient Cor (magnitude, turnout) Turnout in Lisbon Distribution*turnout coefficient Distribution*Cor (Mag., Turn.) Distribution*turnout in Lisbon Constant R squared N # Clusters
0.132 (0.236) 0.005 (0.011) 0.008** (0.002) 0.073** (0.015) 0.006** (0.001) 0.020 (0.052) 0.71 51 4
**p < 0.05. Estimation is by OLS.
founding election. As expected, actors' strategic behavior leads to an increasing asymmetry as time goes by in turnout rates across districts. Second, the partisan turnout bias was measured. Our hypotheses were strongly supported. The partisan turnout bias after the founding election depends on the change in the asymmetry across districts in (i) the distribution of partisan voting strength and (ii) turnout rates. Finally, our analysis of how PR districted electoral systems generates a partisan turnout bias opens up two important research questions. On the one hand, given that asymmetries in turnout rates across districts are crucial, we expect that the slope of the bias will be greater when variance in district magnitude is high. The higher the variance in district magnitude, the greater the difference the psychological effect will be across districts. A comparison between Portugal and Luxembourg, for instance, would be enlightening. On the other hand, in PR districted electoral systems with a strong regional cleavage (Spain, for instance) the relationship between district magnitude and turnout is undermined. Therefore, when there are regional or local parties the partisan turnout bias should not be particularly important. In sum, there are four possible scenarios for testing our argument when crossing these two variables (variance and regionalization/nationalization). Portugal represents the ideal country for observing the greatest partisan turnout bias.
0.043 (0.087)
0.002 (0.022)
0.70 51 4
0.73 51 4
Blais, A., 2006. What affects turnout? Annu. Rev. Polit. Sci. 9, 111e125. Blais, A., Aarts, K., 2006. Electoral systems and turnout. Acta Polit. 41, 180e196. DeNardo, J., 1980. Turnout and the vote: the Joke's on the Democrats. Am. Polit. Sci. Rev. 74, 406e420. Electoral Studies, 2007. Special symposium: consequences of low turnout. Elect. Stud. 26, 539e632. Fisher, S.D., Lessard-Phillips, L., Hobolt, S.B., Curtice, J., 2008. Disengaging voters: do plurality systems discourage the less knowledgeable from voting? Elect. Stud. 27, 89e104. Gallego, A., Rico, G., Anduiza, E., 2012. Disproportionality and voter turnout in old and new democracies. Elect. Stud. 31, 159e169. Grofman, B., Koetzle, W., Burnell, T., 1997. An integrated perspective on the three potential sources of partisan bias: malapportionment, turnout differences, and the geographic distribution of party vote shares. Elect. Stud. 16, 457e470. Grofman, B., Selb, P., 2011. Turnout and the (effective) number of parties at the national and district levels: a puzzle-solving approach. Party Polit. 17, 93e117. Lago, I., Montero, J.R., 2013. Defining and measuring party system nationalization. Eur. Polit. Sci. Rev. http://dx.doi.org/10.1017/ S1755773913000027.
158
I. Lago, M.C. Lobo / Electoral Studies 35 (2014) 150e158
~o. In: Marina, C. Lobo, Lobo, M.C., 2009. Um sistema partid ario sob pressa ~es, Pedro (Eds.), As eleiço ~ es legislativas e presidenciais, Magalha 2005e2006. ICS, Lisbon, pp. 9e20. Lobo, M.C., Magalhaes, P.C., 2011. Room for Manoeuvre: euroscepticism in the Portuguese parties and electorate 1976e2005. South Eur. Soc. Polit. 16, 81e104.
Monroe, B.L., Rose, A.G., 2002. Electoral systems and unimagined consequences. Partisan effects of districted proportional representation. Am. J. Polit. Sci. 46, 67e89. Tavits, M., Annus, T., 2006. Learning to make votes count: the role of democratic experience. Elect. Stud. 25, 72e90.