Multilevel Analysis of Voting Behavior under the Mixed-Member Majoritarian Systems
Ⅳ. Multilevel Analysis
1. Hypothesis and Data
Since Korea, Taiwan, and Japan share the same election
sys-108 선거연구_제1권2호(2011년 가을호)
tem, this paper is aimed at explaining voting behavior from the perspective of personal vote. Our major hypothesis is that the dif-ference between the SMD and PR districts has a great impact on voting behavior. In other words, the probability of choosing the incumbent party will increase with the difference between the SMD and PR districts, controlling for party identification. The ra-tionale is that the electorate in a given district will cast personal votes if the incumbent party’s candidates perform well regardless of their party identification. Korea, Taiwan, and Japan serve three good cases to test this hypothesis. We use CSES Module 2 (2001-2006) data, the Nation-wide Longitudinal Survey Study on Voting Behavior in the 21st Century (JES III) data, and Taiwan’s Election and Democratization Study (TEDS 2008L) data to exam-ine voting behavior in the 2004 National Assembly election of Korea, the 2008 legislative election of Taiwan, and the 2005 gen-eral election of Japan respectively.
Due to the fact that we are not able to either identify every re-spondent’s SMD district or collect each SMD districts’ PR vote shares, we turn to the cities where the SMD districts are located.
For instance, we can easily find out the election results of Korea’s 16 administrative areas, such as Seoul Special City and Busan Metropolitan City. We then match the election results to each re-spondent’s electoral district according to the city where he/she resides.
In addition to take the difference in vote shares between two tiers into account, we also consider the effects of party
identi-Multilevel Analysis of Voting Behavior under the Mixed-Member Majoritarian Systems 109 fication, ideology or important issues, and political leader’s popularity. In Korea and Japan, the conservative-liberal ideology has been an important long-standing political attitude. As for Taiwan, the independence-unification issue has been critical to voting behavior. Moreover, presidential and prime minister’s pop-ularity is included in the models. We assume that people who like the president or prime minister will vote for his party.
To be sure, voting behavior may be different across different groups of education and age. We also model both age and educa-tion variables as two more levels in addieduca-tion to the level of SMD districts. In doing so, we are able to adjust the nonresponse of voting behavior because the model can generate a variance com-ponent from each factor (Gelman and Hill 2007:301). Due to the fact that we rely on survey data instead of census data, however, there are many nonresponses in the education and age information. Therefore, we may lose many nonresponse cases.
2. Multilevel model
When there is grouped data, we can test if both individual- and group-level indicators can explain the dependent variable. A respondent’s behavior, for instance, could be predicted by a set of individual-level variables such as predisposition, values, and past behavior, and it can vary with the group that he belongs to.
Certainly, one could belong to different groups, such as age,
edu-110 선거연구_제1권2호(2011년 가을호)
cation, membership, or residence. The variation within either group could be explained by another set of indicators, or it fol-lows certain probability distribution. The multilevel model alfol-lows us to fit a logistic regression to the individual data given individual predictors first, then performing a linear regression at the group level. Because group indicators are modeled by a group-level dis-tribution, there is no need to worry about the collinearity.
A multilevel model can be written in many ways. We express our multilevel logistic regression model in the following form:
In the model, β1 and β2 represent the coefficients of partisan-ship and ideology in the individual-level model, and β3 repre-sents the coefficient of the difference between the SMD and PR districts in the group-level model. The group-level distributions for each group-level predictors, α, pull the estimates toward the mean level, β3 and 0. The sample size of each group determines how much information may be in the group-level model. When there are many observations within a group, the multilevel esti-mates will be close to the estiesti-mates from a classical regression model with categorical predictors of each group. In contrast, the multilevel estimates will be like the estimates from a regression
Multilevel Analysis of Voting Behavior under the Mixed-Member Majoritarian Systems 111 model without any group predictor when there are few ob-servations within a group.
3. Results
Is the level of personal vote uniform through all of the districts in each country? Or does it vary from districts to districts? If every district has the same level of personal vote, candidates probably add limited influence to voting behavior. Figure 1 shows the dif-ference in vote shares of the Woori Party between the SMD and PR districts in Korea. It is apparent that in most administrative areas the difference in vote shares between two tiers ranges from -5% to 5%. This result suggests that Korean voters supported the incumbent party consistently; they did not split their votes in most of areas.
Regarding the 2008 legislative election in Taiwan, we draw the bar chart to present the differences in vote shares of the KMT in the SMD and PR tiers. Although the DPP was the presidential par-ty in 2008, the KMT and its allies were the majoripar-ty in the legis-lative body. Therefore, we calculate the KMT’s vote shares in the 23 cities and counties. Figure 2 shows that the differences range from -10% to 10% in most cities and counties. The KMT did not nominate anyone in Penghu County; they waived the race to an incumbent who ran as independent candidate. The difference in Penghu County is thus as high as 50%. This county was not drawn
112 선거연구_제1권2호(2011년 가을호)
Figure 1┃The Differences in the SMD and PR Vote Shares in 14 Major Cities, Korea 2004
in our sampling design, so we may ignore its potential influence.
Following the two graphs above, we draw the bar chart to ob-serve the variation in the SMD-PR gap in Japan’s 47 cities and counties. Figure 3 shows the difference in LDP’s vote shares be-tween two tiers. It is apparent that voters are not loyal to the LDP in both tiers; the difference ranges from -5% to more than 20%. In most districts, the LDP’s SMD vote shares were higher than their PR vote shares. Given that the LDP won 219 SMD seats and 77 PR seats while the DP won 52 SMD seats and 61 PR seats, the SMD-PR gap seemed to reflect the LDP’s unbalanced performance.
One of the explanations of LDP’s overwhelming performance in the SMD districts is that the Prime Minister, Koizumi, successfully
Multilevel Analysis of Voting Behavior under the Mixed-Member Majoritarian Systems 113
Figure 2┃The Differences in the SMD and PR Vote Shares in 23 Cities and Counties, Taiwan 2008
featured himself as well as his privatization policies in this election. Moreover, the LDP recruited celebrities, bureaucrats, and local politicians to run against 37 former LDP parliamentary mem-bers who opposed Koizumi. We suspect that the LDP’s large SMD-PR gap resulted from its candidates’ unusual popularity.
To estimate the effects of individual-level variables and consid-er the group-level variation, multilevel models are estimated for voting behavior in Korea, Taiwan, and Japan. The top part of Table 1 presents the fixed effects and the bottom part presents the random effects of the variables. The fixed effects refer to co-efficients that do not vary by group; the intercept and slopes are
114 선거연구_제1권2호(2011년 가을호)
Figure 3┃The Differences in the SMD and PR Vote Shares in 46 Cities and Counties, Japan 2005
averaging inference over the groups. The random effects mean the variance at the group-level. The larger the group-level var-iance, the influence of groups on the estimates is greater. A square root of the variance is the estimated standard deviation of an intercept. According to the standard deviation, we can tell the ex-tent to which groups may differ from each other.
The first column of Table 1 shows that Korean voters were more likely to vote for the incumbent party if they identified with it. Moreover, they would not vote for the Woori Party if they were closer to ideological conservative. Instead, they would vote for the Woori Party when they liked its performance. The coefficient for
Multilevel Analysis of Voting Behavior under the Mixed-Member Majoritarian Systems 115
District variance 0.240 5.726 0.063
Education variance 0.008 0.186 0.000
Age variance 0.051 0.009 0.000
Log likelihood -504.8 -208.1 -414
N 870 570 945
Note: #Ideology refers to liberal-conservative ideology in Korea and Japan, and it means independence-unification issue position in Taiwan. *p<=0.05, **p<=0.01, *p<=0.001. Standard errors are in the parentheses.
Table 1┃Logistic Models of Predicting Voting Behavior in Korea, Taiwan, and Japan
the difference in vote shares between the SMD and PR districts is not statistical significant, which means that the difference in per-sonal vote would not make any difference in support for the in-cumbent party.
The second column shows the estimates of the model for the 2008 legislative election of Taiwan. Like Korea, the difference of SMD and PR vote shares has no significant impact on voting behavior. Self-identification with the incumbent party (KMT) pos-itively influences the probability of voting for the KMT’s candi-dates, but positive evaluation of the government (DPP) would de-crease it. As for the variance part, 21 intercepts of districts seem to have a large standard deviation ( ).
116 선거연구_제1권2호(2011년 가을호)
The third column of Table 1 presents the M.L.E. estimates and their standard errors. In this model, party identification, left-right ideology, and cabinet’s popularity have significant impacts on the probability of voting for the LDP. More importantly, personal vote measured by the difference in SMD and PR vote shares is a sig-nificant predictor, controlling for other variables. The larger the difference, the higher likelihood of voting for the LDP is.
Compared to Korea and Taiwan, Japan’s MMM system with multi-ple PR districts seems to encourage more personal voting; the in-cumbent party benefits from its candidates’ efforts even though the party as a whole performs not very well.
Ⅴ. Conclusion
We depart from the idea that election systems may constrain voting behavior. Thanks to the concept of personal vote linked to election systems by Carey and Shugart, we explore the dis-trict-level factors of individual’s voting pattern. We assume that personal vote would be more important regarding voting behavior in countries that have more than one party list than in countries that pool all of the PR votes together. To test this hypothesis, we chose Korea, Taiwan, and Japan as three case studies. The results confirm our hypothesis; personal voting measured by the differ-ence in vote shares between SMD and PR tiers was prominent in
Multilevel Analysis of Voting Behavior under the Mixed-Member Majoritarian Systems 117 Japan.
The principal implication of our findings for political parties is that the effect of personal votes depends on election systems.
When PR votes are counted separately in several areas, voters are motivated to split their votes because their PR votes would only influence a portion of PR seats. Therefore, they may focus on the SMD districts in which candidates are expected to cultivate per-sonal votes. The primary election system may reinforce perper-sonal voting because it opens the access to the electorate. Instead, sin-gle party list may lead parties to emphasize more party reputation both in the SMD and PR districts because both votes count for seat allocation in the national level. The electorate may concentrate on their votes on the candidate who is most likely to win the sin-gle-member district and the SMD-PR gap becomes wider (Moser and Scheiner 2005). However, we do not find that strategic voting in the SMD districts contributes to individual’s voting behavior in Taiwan and Korea. Therefore, we suggest that whether there are multiple PR districts conditions voting behavior.
In the future, we should extend this comparative framework to other MM systems, such as New Zealand, Germany, and Russia.
We expect that countries that use PR votes to determine seat allo-cation like Germany may have small SMD-PR gaps than Russia and New Zealand do. However, parties can nominate different party lists in each state may counter their emphasis on PR votes and party reputation. Moser and Scheiner have found that German voters cast ballots to the top candidates to avoid wasting their
118 선거연구_제1권2호(2011년 가을호)
votes. Perhaps it is because separate, localized party lists encour-age the electorate to consider who can best represent their con-stituency interest instead of their party preference.
Finally, we should note that cross-national research with more cases could consider the variation between countries. With coun-tries as the intercept for the districts’ effects, we are able to ob-serve whether difference in election systems partly explained by countries. In doing so, we can control for the influence of factors other than institutions.
References
Bawn, Kathleen, and Michael F. Thies. 2003. “A Comparative Theory of Electoral Incentives.” Journal of Theoretical Politics 15, No. 1, 5-32.
Carey, John. M. and Matthew Soberg Shugart. 1995. “Incentive to cultivate a per-sonal vote: a rank ordering of electoral formulas.” Electoral Studies 14, No.
4, 417-439.
Cox, Gary W. (1997) Making Votes Count: Strategic Coordination in the World's Electoral Systems, Political Economy of Institutions and Decisions. Cambridge, UK, New York: Cambridge University Press.
Crisp, Brian F., Kathryn M. Jensen, and Yael Shomer. 2007. “Magnitude and Vote Seeking.” Electoral Studies 26, 7275-734.
Ferrara, Federico, Erik S. Herron, and Misa Nishikawa. 2005. Mixed Electoral Systems: Contamination and its Consequences. New York: Palgrave.
Grofman, Bernard. 2005. “Comparisons among Electoral Systems: Distinguishing between Localism and Candidate-Centered Politics.” Electoral Studies 24, 735-740.
Multilevel Analysis of Voting Behavior under the Mixed-Member Majoritarian Systems 119
Hicken, Allen and Yuko Kasuya. 2003. “A Guide to the Constitutional Structures and Electoral Systems of East, South and Southeast Asia.” Electoral Studies 22, 121-151.
Katz, Richard S. 1997. Democracy and Elections. New York: Oxford University Press.
Kawado, Sadafumi. 2004. Election System and Party System. Tokyo: Bokutakusha.
Kim, Wang Sik. 2007. “Political Effects of the ‘One Man Two Votes’ Electoral System in Korea.” Japanese Journal of Electoral Studies 22, 107-119.
Kim, HeeMin, Jun Young Choi, and Jinman Cho. 2008. “Changing Cleavage Structure in New Democracies: An Empirical Analysis of Political Cleavages in Korea.” Electoral Studies 27, 136-150.
Kim, Young-rae and Sang-shin Park. 2009. “The Reform of Electoral System and the Change of the Party System in Korea.” Paper presented at the International Conference on Elections in Taiwan, Japan, and Korea under the Mixed-Member Electoral Systems.
Kobayashi, Yoshiaki and Kazuaki Nagatomi. 2009. “Changes in Voting Behavior after the Electoral System Reform in Japan.” Paper presented at the International Conference on Elections in Taiwan, Japan, and Korea under the Mixed-Member Electoral Systems.
Kohno, Takeshi. 2007. “Why and How Political Reform Came to Japan.” Paper presented at the International Symposium on Mixed Electoral Systems in East Asia.
Lin, Jih-wen. 1996. “Consequence of the Single Nontransferable Voting Rule:
Comparing the Japan and Taiwan Experiences.” Ph.D. dissertation, University of California at Los Angles.
Lin, Jih-wen. 2006. “The Politics of Reform in Japan and Taiwan.” Journal of Democracy 17, No. 2, 118-131.
Mori, Tadashi. 2009. “Changes in Political Parties and Party Systems after the Electoral System Reform in Japan.” Paper presented at the International Conference on Elections in Taiwan, Japan, and Korea under the Mixed-Member Electoral Systems.
Moser, Robert G. and Ethan Scheiner. 2004. “Mixed Electoral Systems and
120 선거연구_제1권2호(2011년 가을호)
Electoral System Effects: Controlled Comparison and Cross-National Analysis.” Electoral Studies 23, 575-599.
Moser, Robert G. and Ethan Scheiner. 2005. “Strategic Ticket Splitting and the Personal Vote in Mixed-Member Electoral Systems.” Legislative Studies Quarterly 30, No. 2, 259-276.
Reed, Steven R. 1999. “Strategic Voting in the 1996 Japanese General Election.”
Comparative Political Studies 32, No. 2, 257-270.
__________. 2007. “Duverger’s Law is Working in Japan.” Japanese Journal of Electoral Studies 22, 96-106.
Reed, Steven R. and John M. Bolland. 1999. “The Fragmentation Effect of SNTV in Japan.” Bernard Grofman, Sung-Chull Lee, Edwin A. Winckler, and Brian Woodall, eds. Elections in Japan, Korea, and Taiwan under the Single Non-Transferable Vote. Ann Arbor: The University of Michigan Press.
Samuels, David J. 2002. “Presidentialized Parties: The Separation of Powers and Party Organization and Behavior.” Comparative Political Studies 35, 461- 483.
Shugart, Matthew S. and Martin P. Wattenberg. 2001. Mixed-Member Electoral Systems: the Best of Both Worlds? Oxford: Oxford University Press.
Shvetsova, Olga. 1995. “Electoral Institutional Design.” Ph.D dissertation, California Institution of Technology.
Tsai, Chia-hung. 2003. “Party Voting in Comparative Perspective: the United States, Taiwan, and Japan.” Ph.D dissertation, The Ohio State University.
Multilevel Analysis of Voting Behavior under the Mixed-Member Majoritarian Systems 121