Voting for Parties or for Candidates: Do Electoral Institutions Make a Difference?

Voting for Parties or for Candidates: Do Electoral Institutions Make a Difference? Elena Llaudet Department of Government Harvard University April 11, 2015

Abstract Little is known about how electoral institutions systematically affect the level of incumbency advantage in legislative elections. In this paper, I analyze the Comparative Study of Electoral Systems (CSES) data to put the U.S. case in a comparative context and explore how electoral institutions aid or preclude incumbents from gaining an advantage. In particular, I study whether electoral systems affect the likelihood of party defection in lower house elections, a phenomenon closely related to incumbency advantage and defined as voting for a party other than one s own. I find that the U.S. is not exceptional when it comes to party defection. Close to eighteen percent of the respondents in the U.S. claimed to have voted in 1996 for a congressman from a different party than the one they identified as their own, a number that falls midway within the distribution of the sample of countries analyzed. Furthermore, when focusing on countries with mixed electoral systems, I find that voters are more likely to place their votes based on candidate attributes such as incumbency status, as opposed to party affiliation, in elections were they cast a vote for a candidate rather than in proportional representation elections were they cast a vote for a closed party list.

U.S. elections have long been considered unique, with its candidate-centered campaigns and high levels of incumbency advantage. In this paper, I set up to put the U.S. case in a comparative context and explore how electoral institutions aid or preclude incumbents from gaining an advantage. In particular, I study whether electoral systems affect the likelihood of party defection in lower house elections, a phenomenon closely related to incumbency advantage and defined as voting for a party other than ones own. Party defection seems to be a good proxy for incumbency advantage, on the grounds that if citizens are voting for a party other than their own it usually is because they are voting for the incumbent. A cursory look at the trends of incumbency advantage and party defection in the U.S., shows how closely these two phenomenons follow each other. 1 When looking at the 36 countries included in the Comparative Study of Electoral Systems (CSES) dataset, I find that the U.S. is not exceptional when it comes to party defection. Close to eighteen percent of the respondents in the U.S. claimed to have voted in 1996 for a congressman from a different party than the one they identified as their own, a number that falls midway within the distribution of the sample of countries analyzed. 2 However, this list includes countries with very different type of electoral systems. If we only look at elections in which voters cast votes for candidates instead of parties, the U.S. ranks close to the bottom. A quick look at the numbers seems to suggest that electoral systems have an effect on the level of party defection. On average, a vote casted in a single-member district elections has a 25 percent probability of being for a party other than the respondent s own. Multiplemember district elections have a probability of party defection twelve percentage points smaller. Proportional representation elections with open party lists have a probability of party defection 4 percentage points higher than that of single-member district elections, on average. While proportional representation elections with closed party lists have a probability 1 See Figures A.1 and A.2 in the Appendix. 2 For details, see Table A.1. in the Appendix. 3

of party defection 4 percentage points smaller than that of single member district elections, on average. (All differences are found to be statistically significant.) Unfortunately, this type of broad comparisons are not quite valid since there are many differences among these countries other than their electoral systems. The observed differences in party defection might be due to differences in the strength of their parties, for example. Some countries, however, have mixed electoral systems, that is two electoral systems working alongside each other. In these cases, we can observe how the different electoral systems have an effect on voting behavior, while controlling for everything else (e.g., country characteristics, individual preferences,... ). When focusing on the four countries in the CSES dataset where two electoral systems coexist, I find that, on average, the same individual is six percentage points more likely to vote for his or her own party in proportional representation elections with closed party lists than in concurrent single-member district elections. Results vary depending on the respondent s closeness to the party, recollection of candidates names, contact with politicians, and level of political knowledge. This finding suggests that, in general, voters are more likely to place their votes based on candidate attributes such as incumbency status, as opposed to party affiliation, in elections were they cast a vote for a candidate rather than in proportional representation elections were they cast a vote for a closed party list. Data and Methodology The CSES provides information on the attitudes and voting behavior of a representative sample of thousands of citizens in 36 countries immediately following one of their lower house elections. 3 To explore how electoral institutions systematically affect voting behavior, I focus on the four countries in the CSES data set in which two different electoral systems coexist: Germany in 1998, Japan in 1996, Hungary in 1996 and, New Zealand in 1996. 3 I use the August 2003 version of Module 1. 4

In these countries, voters were asked to cast two ballots for the election of the members of the lower house of their legislature, each vote was for a different type of electoral system. They were asked to vote for a closed party list as part of a proportional representation election, and they were asked to vote for a candidate either to be chosen by a plurality (in the case of Germany, Japan, and New Zealand) or by the majority (in the case of Hungary). This type of mixed electoral systems allow us, then, to observe the effect of electoral institutions on the likelihood to vote for a different party than ones own, while everything else remains constant. In order to also explore whether the differences between these two systems are exacerbated by other factors, I also include in the model some interaction terms with other potential determinants of party defection: (1) whether the respondent admitted to feeling close to a party when first asked, (2) whether he was able to recall one or more candidates from the last lower house election, (3) whether he had contact with a member of the legislature in the last year, and (4) whether he had correctly answered one, two, or all three of the political knowledge questions asked. 4 The model I use for the analysis is, thus, as follows: Party Defection ics = β 1 PR Closed Party List cs (1) where: + β 2 Feels Close to a Party ics PR Closed Party List ics + β 3 Recalls Only One Candidate ics PR Closed Party List ics + β 4 Recalls More than One Candidate ics PR Closed Party List ics + β 5 Had Contact with Politician ics PR Closed Party List ics + β 6 Got One Answer Right ics PR Closed Party List ics + β 7 Got Two Answers Right ics PR Closed Party List ics + β 8 Got Three Answers Right ics PR Closed Party List ics + θ ic + ɛ ics Party Defection ics is a dummy variable indicating whether respondent i in country c voted for a party other than his or her own in the elections using electoral system s. 5 4 Unfortunately, the degree of difficulty of these three questions varied widely between countries and only two were asked in the survey following the Japanese elections. 5 To create the PID measure, I used the respondents answers to the following questions: Do you usually think of yourself as close to any particular political party? Which one is that? If more than one was 5

PR Closed Party List cs is a dummy variable indicating whether the electoral system s is PR Closed Party List. Feels Close to a Party PR Closed Party List ics is an interaction variable between a dummy indicating whether respondent i in country c reported to feel close to a party and a dummy indicating whether the electoral system s is PR Closed Party List. Recalls Only One Candidate PR Closed Party List ics is an interaction variable between a dummy indicating whether respondent i in country c was able to recall the name of only one of the candidates in the race and a dummy indicating whether the electoral system s is PR Closed Party List. Recall More than One Candidate PR Closed Party List ics is an interaction variable between a dummy indicating whether respondent i in country c was able to recall the name of more than one of the candidates in the race and a dummy indicating whether the electoral system s is PR Closed Party List. Had Contact with Politician PR Closed Party List ics is an interaction variable between a dummy indicating whether respondent i in country c had had any contact with a member of Parliament/Congress during the past twelve months and a dummy indicating whether the electoral system s is PR Closed Party List. Got One Answer Right PR Closed Party List ics is an interaction variable between a dummy indicating whether respondent i in country c answered only one (out of three) political knowledge questions right and a dummy indicating whether the electoral system s is PR Closed Party List. The other variables related to the political knowledge questions function in the same way. θ ic are fixed effects for each respondent i in each country c. ɛ ics are the usual residuals. The baseline category is the respondent s estimated probability of party defection when voting in the plurality or majority candidate system if he or she doesn t feel close to any party, does not recall any of the names of the candidates, had not had any contact with a member of the legislature in the last year, and did not get any of the political knowledge questions correct. By including respondent s fixed effects, the model estimates how the voting behavior of the same individual changes, on average, as a result of voting in a different electoral system. mentioned: Which party do you feel closest to? If none were mentioned: Do you feel yourself a little closer to one of the political parties than the others? If so, which one? (In New Zealand this last question was not asked, and, thus, it wasn t used in the construction of the PID variable). 6

Presumably, voters are more likely to be loyal to their party in PR systems where they are asked to choose a closed party list than in single-member districts where they are asked to cast a vote for a candidate, especially when they report to feel close to a party. Likewise, voters are probably more likely to vote based on the characteristics of the candidate in systems were they have to chose them directly, especially when they are able to recall one or more of the names of the candidates in the race or have been in contact with them. Finally, higher levels of knowledge of politics might make a respondent more incline to break party loyalties in favor of particular candidates who represent his or her political preferences better, especially in elections were voters cast votes for them directly. As result, we should expect all the coefficients of model (1), to have a negative sign, if any at all. Results Table 1 shows the results of the analyses done of all four countries with mixed electoral systems in the CSES data set. Column (1) shows the model without interactions, column (6) shows the results of the model fully interacted, and all other columns show the results of models with each one of the interactions separately. As shown in column (1) of Table 1, voters are, on average, six percentage points more likely to vote for a party other than one s own in the elections were they choose candidates directly (whether as part of a plurality or majority system) than in proportional representation elections, where they chose closed party lists. About the results of all the other models, the sign of all coefficients (whenever significant) are as expected. Based on the results shown in column (2), for example, voters who feel close to a party are four percentage points more likely to be loyal to their party in the PR elections than in the majority or plurality elections, as compared to those who do not feel close to a party. If we focus on the more complete model shown in column (6), the only significant difference in voting behavior between the two systems is found among respondents who are able to recall one or more than one can- 7

didate in the race. In those cases, voters are between four and ten percentage points more likely to vote for a party other than their own in the majority or plurality candidate system than in the PR closed party list system, depending on whether they recall only one or more than one candidate. Table 1: Analysis of Mixed Electoral Systems Dependent Variable Party Defection Strategic Vote Candidate Vote (1) (2) (3) (4) (5) (6) (7) (8) (9) PR Election -0.062 0.056 0.058-0.059 0.006 0.010 0.006 0.038 0.035 (0.024) (0.191) (0.197) (0.015) (0.131) (0.134) (0.029) (0.100) (0.100) PR Feels -0.039-0.033-0.009-0.003-0.010-0.012 Close to a Party (0.103) (0.100) (0.025) (0.025) (0.096) (0.093) PR Identifies with -0.202-0.205-0.155-0.158-0.046-0.046 Minor Party in SMD (0.323) (0.326) (0.277) (0.279) (0.081) (0.083) PR Identifies with -0.024-0.021 0.264 0.267-0.127-0.127 Minor Party in PR (0.358) (0.353) (0.121) (0.119) (0.362) (0.363) PR Political 0.003 0.010 0.004 0.007-0.003 0.001 Knowledge (0.018) (0.022) (0.011) (0.014) (0.018) (0.017) PR Margin in 0.036 0.033 0.020 0.020 0.014 0.012 SMD Tier (0.031) (0.030) (0.034) (0.032) (0.028) (0.030) PR Name -0.035-0.005-0.030 Recollection (0.047) (0.011) (0.041) PR Contact -0.008-0.033 0.034 with MP (0.036) (0.031) (0.060) Observations 9,190 9,163 9,115 9,190 9,163 9,115 9,190 9,163 9,115 Notes: Respondent clustered standard errors are shown in parentheses. Coefficients statistically significant at the 90 percent level of confidence are shown in bold. All regressions include respondent fixed effects. If we study the data country by country, we find similar results, albeit with some exceptions. (See Tables 2-5). The most notable exception is that in the 1998 german elections to the Bundestag, respondents were, on average, as likely to vote for a different party other than their own in the PR elections as they were in the single member district elections. However, some differences start to come up when we look at some of the interaction terms individually. Based on the model shown in column (2), respondents who claimed to feel close to a party were two percentage points less likely to vote for a party other than their own in the PR 8

system than in the single member district system. Based on the model shown in column (3), respondents who could recall more than one of the candidates were seven percentage points more likely to vote for a party other than their own in the single member district system as they were in the PR system. Similarly, as shown in column (5), respondents with high levels of political knowledge were more likely to vote for a party other than their own in the single member district election than they were in the PR election, by either 2 or 5 percentage points, depending on the level of knowledge. According to the summary statistics shown in Appendix A.1., however, the average voter in the 1998 German elections, did not feel close to a party, recalled none of the candidates in the race, and was not that knowledgeable about politics. Table 2: Effects of the Electoral Institutions in the 1998 German Elections Dependent Variable Party Defection Strategic Vote Candidate Vote (1) (2) (3) (4) (5) (6) (7) (8) (9) PR Election.007.100.101 -.038.010.013.066.106.104 (.020) (.036) (.036) (.013) (.016) (.016) (.018) (.036) (.036) PR Feels -.043 -.041.006.009 -.062 -.061 Close to a Party (.042) (.043) (.025) (.026) (.039) (.040) PR Identifies with -.160 -.160 -.144 -.145.005.005 Minor Party in SMD (.045) (.045) (.031) (.031) (.038) (.038) PR Identifies with.034.028.404.403 -.157 -.160 Minor Party in PR (.155) (.157) (.173) (.172) (.178) (.181) PR Political -.032 -.029 -.017 -.015 -.018 -.016 Knowledge (.021) (.021) (.014) (.014) (.019) (.018) PR Margin in.022.018.016.015 -.000 -.003 SMD Tier (.021) (.021) (.013) (.014) (.017) (.018) PR Name -.028 -.003 -.021 Recollection (.020) (.012) (.018) PR Contact.006 -.031.027 with MP (.058) (.037) (.049) Observations 2,236 2,211 2,205 2,236 2,211 2,205 2,236 2,211 2,205 Notes: Respondent clustered standard errors are shown in parentheses. Coefficients statistically significant at the 90 percent level of confidence are shown in bold. All regressions include respondent fixed effects. 9

Table 3: Effects of the Electoral Institutions in the 1996 Japanese Elections Dependent Variable Party Defection Strategic Vote Candidate Vote (1) (2) (3) (4) (5) (6) (7) (8) (9) PR Election -.069.094.093 -.089 -.011 -.012 -.040.099.097 (.028) (.058) (.057) (.020) (.033) (.033) (.028) (.061) (.060) PR Feels -.078 -.074 -.004 -.004 -.079 -.077 Close to a Party (.063) (.063) (.042) (.043) (.067) (.067) PR Identifies with -.310 -.314 -.228 -.235 -.225 -.224 Minor Party in SMD (.062) (.062) (.048) (.049) (.067) (.067) PR Identifies with.259.241 1.246 1.256 -.835 -.862 Minor Party in PR (.076) (.061) (.050) (.054) (.075) (.068) PR Political.013.014.016.017 -.000.000 Knowledge (.024) (.025) (.017) (.018) (.026) (.027) PR Margin in.009.007.009.008 -.002 -.004 SMD Tier (.026) (.026) (.015) (.016) (.027) (.026) PR Name -.051.006 -.063 Recollection (.034) (.022) (.035) PR Contact.081.019.082 with MP (.080) (.040) (.083) Observations 1,239 1,239 1,220 1,239 1,239 1,220 1,239 1,239 1,220 Notes: Respondent clustered standard errors are shown in parentheses. Coefficients statistically significant at the 90 percent level of confidence are shown in bold. All regressions include respondent fixed effects. Conclusions 10

Table 4: Effects of the Electoral Institutions in the 1998 Hungarian Elections Dependent Variable Party Defection Strategic Vote Candidate Vote (1) (2) (3) (4) (5) (6) (7) (8) (9) PR Election -.095 -.118 -.113 -.011 -.013 -.014 -.099 -.124 -.117 (.018) (.033) (.031) (.014) (.022) (.022) (.019) (.033) (.032) PR Feels.059.066.013.012.064.072 Close to a Party (.038) (.039) (.029) (.029) (.040) (.040) PR Identifies with -.027 -.032 -.034 -.034 -.017 -.024 Minor Party in SMD (.041) (.041) (.028) (.029) (.042) (.042) PR Identifies with -.009 -.007.122.121 -.067 -.063 Minor Party in PR (.084) (.082) (.088) (.088) (.084) (.083) PR Political.002.016 -.011 -.010.000.013 Knowledge (.018) (.018) (.013) (.013) (.019) (.019) PR Margin in -.001 -.002 -.003 -.003 -.007 -.008 SMD Tier (.019) (.019) (.014) (.014) (.020) (.019) PR Name -.063 -.007 -.061 Recollection (.017) (.014) (.018) PR Contact.048.025.023 with MP (.058) (.036) (.058) Observations 1,434 1,432 1,425 1,434 1,432 1,425 1,434 1,432 1,425 Notes: Respondent clustered standard errors are shown in parentheses. Coefficients statistically significant at the 90 percent level of confidence are shown in bold. All regressions include respondent fixed effects. 11

Table 5: Effects of the Electoral Institutions in the 1996 New Zealand Elections Dependent Variable Party Defection Strategic Vote Candidate Vote (1) (2) (3) (4) (5) (6) (7) (8) (9) PR Election -.085.032.043 -.078.001.016.022.040.038 (.017) (.022) (.025) (.010) (.008) (.011) (.016) (.022) (.025) PR Does NOT Feel.061.046 -.034 -.043.077.073 Close to a Party (.076) (.077) (.059) (.058) (.073) (.074) PR Identifies with -.249 -.250 -.176 -.179 -.041 -.040 Minor Party in SMD (.034) (.034) (.021) (.021) (.033) (.033) PR Identifies with -.198 -.187.180.189 -.120 -.117 Minor Party in PR (.114) (.115) (.125) (.125) (.139) (.140) PR Political.019.027.013.017.009.011 Knowledge (.017) (.018) (.010) (.011) (.017) (.018) PR Margin in.061.059.031.031.030.028 SMD Tier (.016) (.016) (.010) (.010) (.016) (.016) PR Name -.025 -.007 -.013 Recollection (.018) (.011) (.019) PR Contact -.031 -.044.007 with MP (.036) (.023) (.034) Observations 4,281 4,281 4,265 4,281 4,281 4,265 4,281 4,281 4,265 Notes: Respondent clustered standard errors are shown in parentheses. Coefficients statistically significant at the 90 percent level of confidence are shown in bold. All regressions include respondent fixed effects. 12

Appendix Figure A.1: Congressmen s Incumbency Figure A.2: Party Defection in US Advantage in the US, 1940-1990 Congressional Elections, 1940-1990 Percentage Points 2 4 6 8 Percent 10 12 14 16 18 20 1940 1950 1960 1970 1980 1990 1940 1950 1960 1970 1980 1990 Decades Decades Source: Ansolabehere and Snyder (2002) Source: Vital Statistics of Congress, 2008. 13

Table A.1: Summary Statistics by Lower House Country Elections (means and sd) Party Electoral Feels Recalls Candidates Had Average # Defection Formula Close None One >One Contact of Correct in Lower & to a with a Political House Voting Political Politician Knowledge Elections Procedure a Party Last Year Answers AUS 1996 0.29 3 0.83 0.16 1.59 0.45 0.37 0.37 0.96 BELF1999 0.27 4 0.95 1.34 0.44 0.22 1.01 CAN 1997 0.51 1 0.50 0.32 0.22 0.46 0.22 1.78 0.50 0.50 0.47 0.41 0.50 0.41 0.97 CHE 1999 0.14 4 0.36 0.54 0.14 0.32 0.19 1.50 0.35 0.48 0.50 0.35 0.47 0.39 1.02 CZE 1996 0.24 4 0.49 0.42 0.21 0.37 0.08 2.03 0.43 0.50 0.49 0.40 0.48 0.26 0.99 DEU 1998 1 0.21 1 0.34 0.58 0.24 0.18 0.11 1.22 0.41 0.47 0.49 0.43 0.38 0.32 0.89 DEU 1998 2 0.21 5 0.34 0.58 0.24 0.18 0.11 1.22 0.41 0.47 0.49 0.43 0.38 0.32 0.89 DNK 1998 0.25 4 0.50 0.23 0.20 0.58 0.20 0.43 0.50 0.42 0.40 0.49 0.40 ESP 1996 0.22 5 0.43 0.74 0.16 0.11 0.03 1.53 0.42 0.49 0.44 0.36 0.31 0.18 1.02 ESP 2000 0.06 5 0.42 0.75 0.14 0.11 0.03 1.53 0.23 0.49 0.43 0.34 0.31 0.18 0.83 GBR 1997 0.15 1 0.46 0.40 0.31 0.29 0.13 1.91 0.36 0.50 0.49 0.46 0.45 0.33 1 HKG 1998 0.25 5 0.08 0.34 0.21 0.44 0.04 1.86 0.44 0.27 0.48 0.41 0.50 0.20 0.92 HKG 2000 0.68 5 0.07 0.38 0.20 0.41 0.06 1.77 0.47 0.26 0.49 0.40 0.49 0.24 0.88 HUN 1998 1 0.17 3 0.35 0.37 0.23 0.40 0.07 1.22 0.38 0.48 0.48 0.42 0.49 0.25 0.98 HUN 1998 2 0.07 5 0.35 0.37 0.23 0.40 0.07 1.22 0.26 0.48 0.48 0.42 0.49 0.25 0.98 ISL 1999 0.23 4 0.51 0.17 0.14 0.70 0.31 0.42 0.50 0.37 0.34 0.46 0.46 ISR 1996 0.45 5 0.64 0.16 1.17 0.50 0.48 0.37 0.99 JPN 1996 1 0.28 1 0.38 0.06 0.14 0.80 0.08 0.52 0.45 0.48 0.24 0.35 0.40 0.27 0.66 a Electoral Formula & Voting Procedure Categories: 1 = SMD-Candidate; 2 = MMD-Candidate: 3 = Majority-Candidate; 4 = PR-Open Party List; 5 = PR-Closed Party List 1 denotes first segment and 2 second segment in mixed systems. 14

Table A.1: Summary Statistics by Lower House Country Elections (Continuation) Party Electoral Feels Recalls Candidates Had Average # Defection Formula Close None One >One Contact of Correct in Lower & to a with a Political House Voting Political Politician Knowledge Elections Procedure a Party Last Year Answers JPN 1996 2 0.20 5 0.38 0.06 0.14 0.80 0.08 0.52 0.40 0.48 0.24 0.35 0.40 0.27 0.66 KOR 2000 1 0.20 1 0.27 0.08 0.51 0.41 0.16 0.40 0.45 0.28 0.50 0.49 0.37 MEX 1997 1 0.16 1 0.44 0.82 0.11 0.07 0.10 1.81 0.37 0.50 0.38 0.32 0.25 0.30 0.97 MEX 2000 1 0.71 1 0.52 0.87 0.11 0.03 0.09 1.52 0.45 0.50 0.34 0.31 0.16 0.29 1.07 NLD 1998 0.18 4 0.28 0.05 1.52 0.39 0.45 0.23 0.95 NOR 1997 0.15 5 0.52 0.31 0.18 0.50 0.15 1.45 0.36 0.50 0.46 0.39 0.50 0.35 0.91 NZL 1996 1 0.21 1 0.56 0.17 0.20 0.62 0.26 1.72 0.49 0.50 0.38 0.40 0.48 0.44 0.97 NZL 1996 2 0.30 5 0.56 0.17 0.20 0.62 0.26 1.72 0.50 0.50 0.38 0.40 0.48 0.44 0.97 PER 2001 0.39 5 0.27 0.27 0.34 0.38 0.09 0.49 0.44 0.45 0.48 0.49 0.28 POL 1997 1 0.43 4 0.53 0.62 0.22 0.16 0.06 1.76 0.49 0.50 0.49 0.41 0.37 0.24 1.03 PRT 2002 0.17 5 0.47 0.81 0.10 0.09 0.06 1.41 0.38 0.50 0.39 0.30 0.29 0.24 0.87 ROU 1996 0.67 5 0.47 0.71 0.19 0.10 0.07 1.46 0.47 0.50 0.46 0.39 0.31 0.26 1.06 RUS 1999 1 0.14 1 0.62 0.31 0.17 0.52 0.03 0.35 0.48 0.46 0.37 0.50 0.16 SVN 1996 1 0.76 4 0.22 0.43 0.41 SWE 1998 0.20 4 0.53 0.67 0.23 0.10 0.11 1.71 0.40 0.50 0.47 0.42 0.31 0.31 1.03 TWN 1996 1 0.14 2 0.34 0.63 0.13 0.24 0.08 1.74 0.34 0.47 0.48 0.33 0.43 0.27 1.21 UKR 1998 1 0.27 3 0.69 0.61 0.18 0.21 0.08 1.87 0.45 0.46 0.49 0.39 0.41 0.26 0.87 USA 1996 0.18 1 0.57 0.48 0.25 0.27 0.14 1.55 0.39 0.50 0.50 0.43 0.45 0.35 0.81 a Electoral Formula & Voting Procedure Categories: 1 = SMD-Candidate; 2 = MMD-Candidate: 3 = Majority-Candidate; 4 = PR-Open Party List; 5 = PR-Closed Party List 1 denotes first segment and 2 second segment in mixed systems. 15