Identifying Voter Preferences for Politicians’ Personal Attributes: A Conjoint Experiment in Japan∗ Yusaku Horiuchi†

Daniel M. Smith‡

Teppei Yamamoto§

First Full Draft: April 3, 2016 This Draft: May 18, 2018

∗

We thank Ben Ansell, Simon Chauchard, Yoshikuni Ono, Ethan Scheiner, Mike Tomz, anonymous

referees, and participants of workshops and conferences for helpful comments; Ashley Kekona for research assistance; Sean Westwood for writing a PHP script embedded in our experiment; and the Reischauer Institute of Japanese Studies at Harvard University and the Dean of Faculty at Dartmouth College for financial support. †

Professor of Government and Mitsui Professor of Japanese Studies, Department of Government, Dart-

mouth College. Email: [email protected], URL: https://sites.dartmouth.edu/horiuchi/ ‡

Associate Professor,

Department of Government,

Harvard University.

Email:

danielm-

[email protected], URL: https://sites.google.com/site/danielmarkhamsmith §

Associate Professor, Department of Political Science, Massachusetts Institute of Technology. Email:

[email protected], URL: http://web.mit.edu/teppei/www

Abstract Although politicians’ personal attributes are an important component of elections and representation, few studies have rigorously investigated which attributes are most relevant in shaping voters’ preferences for politicians, or whether these preferences vary across different electoral system contexts. We investigate these questions with a conjoint survey experiment using the case of Japan’s mixed-member bicameral system. We find that the attributes preferred by voters are not entirely consistent with the observed attributes of actual politicians. Moreover, voters’ preferences do not vary when asked to consider representation under different electoral system contexts, whereas the observed attributes of politicians do vary across these contexts. These findings point to the role of factors beyond voters’ sincere preferences, such as parties’ recruitment strategies, the effect of electoral rules on the salience of the personal vote, and the availability of different types of politicians, in determining the nature of representation. Keywords: representation, electoral systems, personal vote, conjoint analysis, Japan

1

Introduction

The personal attributes of politicians have long been viewed as an important component in determining the outcomes of elections and the nature of representation in democracies. To voters, personal attributes can serve as an informational shortcut about specific benefits and behavior that can be expected from politicians once they are in office. For example, numerous studies across various contexts have found that an increased presence of women in legislatures is correlated with an increase in public spending on issues like childcare and education (e.g., Bratton and Ray, 2002; Svaleryd, 2009). Other studies have found that politicians from ethnic minority groups do better at serving the interests of their communities (e.g., Banducci, Donovan and Karp, 2004; Gay, 2002), and that politicians with local ties tend to be more likely than non-locals to focus on the needs of their hometown communities in various ways (e.g., Carozzi and Repetto, 2016; Fiva and Halse, 2016). Party leaders seeking to win elections thus also take the personal attributes of potential candidates seriously, in part because the right combination of attributes may help optimize the party’s mobilization efforts, and in part because nominated candidates reflect the collective goals and priorities of the party itself (e.g., Crotty, 1968; Schattschneider, 1942). Yet, despite the important role that personal attributes play in elections and representation, relatively few studies have rigorously investigated which particular attributes — including gender, age, ethnicity, local ties, professional background, and prior political experience — are most relevant in shaping voters’ preferences over politicians. Part of the challenge in investigating this question is that politicians possess more than a single attribute that might appeal to voters, in addition to having a party affiliation (which conveys additional information to voters) and articulating various policy positions. This problem is compounded by the fact that many attributes are correlated with each other, making it difficult to isolate which particular attributes matter the most. For example, although a substantial amount of research finds that candidates tend to receive more votes in their hometowns (e.g., Fiva and

1

Smith, 2017; G´orecki and Marsh, 2012; Hirano, 2006), it is not clear whether this reflects voters’ preferences for a local candidate per se, or the greater mobilizational advantages that local candidates have in their hometowns. Existing research using observational data on the attributes of candidates or elected politicians is usually unable to disentangle the attributes of interest from other attributes possessed by a given candidate. Furthermore, we know relatively little about whether voters hold preferences for distinct types of representation under different electoral system contexts. Conventional wisdom is that electoral systems can shape the nature of representation in democracies by enhancing or downplaying the relative salience of politicians’ personal attributes to voters (e.g., Carey and Shugart, 1995; Shugart, Valdini and Suominen, 2005). Indeed, numerous studies using crossnational observational data have documented that the attributes of politicians tend to differ across systems. Perhaps the most consistent finding is that proportional representation (PR) electoral systems with multimember districts tend to elect more women and ethnic minorities than variations of single-member district (SMD) systems (e.g., Moser, 2008; Salmond, 2006). In addition, electoral systems that structure vote choice around candidates rather than parties are assumed to increase the salience of the personal vote,1 such that candidates with strong “personal-vote-earning attributes” — local ties, prior political experience, name recognition or celebrity status, or a family relationship to a previous politician — are more likely to get elected (e.g., Nemoto and Shugart, 2013; Smith, 2018; Tavits, 2010).2 Given these observational patterns, it is often taken for granted that the types of politicians that emerge in a democracy can, at least in part, be engineered through electoral system design. However, observational data of actual politicians across different democracies are not helpful for drawing inferences about how voters’ preferences for different attributes 1

The personal vote is defined as a “candidate’s electoral support which originates in his or her personal

qualities, qualifications, activities, and record” (Cain, Ferejohn and Fiorina, 1987, p. 9). 2

In a similar vein, the importance of localness may be augmented where electoral constituencies consist

of geographically smaller units (Grofman, 2005).

2

are conditioned by electoral systems, because numerous other confounding variables may be present. Aware of this limitation, a few studies have taken advantage of mixed-member electoral systems to examine observational differences in politicians’ attributes under different electoral rules while holding country-specific determinants constant (e.g., Kostadinova, 2007; Moser and Scheiner, 2012). Nevertheless, as long as our inferences are based on observational data, we still lack clear evidence for whether these differences stem from voters’ differentiated preferences for distinct types of representation, or from the roles that parties and electoral institutions play in translating voters’ preferences into representational outcomes. Moreover, most previous research has focused only on gender or ethnicity, leaving other important attributes of a politician’s background, in particular those related to the personal vote, underexplored. In this study, we address these issues with a conjoint survey experiment in Japan. A growing number of studies have used conjoint analysis to evaluate voters’ preferences for the attributes and policy positions of politicians (e.g., Carnes and Lupu, 2016; Franchino and Zucchini, 2015; Hainmueller, Hopkins and Yamamoto, 2014).3 We aim to contribute to this budding literature by including attributes in our analysis that have been identified in existing research as being particularly relevant to a personal vote, such as local ties, prior experience, dynastic family ties, and celebrity status. Including these personal-voteearning attributes in hypothetical politician profiles — detached from the other qualities and character traits possessed by actual politicians — allows us to examine an important question that is underexplored in the existing literature: how much do voters value these “personal” attributes per se, abstracting from the individuals who might possess them? The case of Japan provides a useful institutional setting to examine whether preferences vary when voters consider politicians who are elected under different electoral rules. Japan’s parliament (Diet) consists of two chambers, each of which uses a mixed-member majoritarian 3

Other studies have investigated voter preferences for various attributes using alternative experimental

survey designs (e.g., Aguilar, Cunow and Desposato, 2015; Campbell and Cowley, 2014; Carlson, 2015).

3

electoral system. The different electoral rules are intended to provide different types of representation to voters — namely, locally-oriented representation in geographically small nominal-vote districts, versus broadly-oriented representation in large districts using variants of list-based PR. We randomly assign our respondents to be primed with information on the electoral contexts of Japan’s system, and investigate whether distinct preferences for certain types of members of parliament (MPs) emerge depending on the context that is primed.4 The results reveal that Japanese voters have strong preferences for and against certain attributes of politicians. In contrast to the observed attributes of actual Japanese politicians, voters do not prefer older politicians or celebrities, and are indifferent with regard to dynastic family ties and gender. Moreover, we find consistent patterns in voters’ preferences regardless of the priming of different representational contexts, while the observed attributes do vary across these contexts. These findings suggest that the observed differences in politicians’ attributes under different electoral rules do not necessarily come from voters’ preferences for distinct types of representation in these contexts. Rather, factors such as parties’ recruitment strategies, the availability of different types of politicians, or the ways in which electoral rules enhance the vote-earning value of name recognition, likely play a greater role in determining the nature of representation in Japan.

2

Case Setting and Observational Data

Japan is a useful setting for our experiment, as the personal attributes of politicians have often been viewed as playing an important role in elections (e.g., Curtis, 1971; Hirano, 2006; Scheiner, 2006), and the mixed-member systems of the Diet provide important intra-country 4

Our goal is to understand voters’ preferences for representation in general, not in the actual context of

elections. For this reason, we focus on voters’ preferences for politicians rather than candidates. Moreover, voters cannot choose from among candidates under closed-list PR, one of Japan’s electoral settings, but can nevertheless have real preferences for which types of politicians serve in this context.

4

variation. We first explain these two contextual backgrounds, and then use observational data from elections in 2013 and 2014 (held prior to the timing of our survey experiment) to describe differences in the attributes of actual candidates and MPs in each chamber of the Diet.

2.1

Parties and the Personal Attributes of Politicians

In the dominant Liberal Democratic Party (LDP), candidate selection has traditionally been decentralized to local party branches, with heavy influence of local elites (Reed, 2009). Roughly a third of all LDP candidates have a background in local politics, and a similar fraction are “legacy” candidates who are related to the previous incumbent by blood or marriage (Smith, 2018).5 The name recognition enjoyed by such candidates is believed to help them perform well in elections (Asako et al., 2015), but they also benefit from long, stable ties to constituents and interest groups (Ishibashi and Reed, 1992). Former national bureaucrats (viewed positively by party leaders for their expertise) were traditionally equally common, but have declined in prominence since the 1990s (by 2014, less than 20% of candidates). Around 70% of LDP candidates run in the prefectures of their birth. Until 2017, the main opposition to the LDP was the Democratic Party of Japan (DPJ), which was formed in 1996 by defectors from the LDP and other parties and expanded through a series of party mergers over time.6 As a result, the backgrounds and personal attributes of DPJ candidates have been more diverse than those of LDP candidates (Smith, Pekkanen and Krauss, 2013). Like the LDP, about a third of the DPJ’s candidates have served in local politics, but former national bureaucrats and members of political dynasties have been less 5

The descriptive statistics on candidate backgrounds in this section come from Smith (2013), Smith

(2018), and original data collected from the 2013 and 2014 elections for this study. 6

The DPJ merged with another party in 2016 and re-branded itself as simply the Democratic Party.

Its caucus in the House of Representatives later dissolved just prior to the 2017 House of Representatives election. At the time of our survey experiment in 2015, it was still known as the DPJ.

5

common (13% and 17%, respectively, in 2014), and the party has been slightly more active in recruiting and nominating women (in 2014, 15% of candidates in the DPJ compared to 12% in the LDP). Two other small parties have consistently contested national elections. Komeito, in coalition with the LDP since 1999, is a religious party affiliated with the Buddhist organization Soka Gakkai. Nearly all Komeito candidates are members of Soka Gakkai, and are often former employees of the party or the religious organization (Smith, 2013). More than 90% of Komeito candidates are male, and only about half run in the area of their birth. The Japanese Communist Party (JCP) is a small leftist party that is the oldest party in Japan. About a fifth of JCP candidates have local political experience, and a quarter are women. The party recruits many of its candidates from among party employees, activists, union leaders, and left-wing lawyers (Curtis, 1979). Dynasties, celebrities, and other types of candidates with high name recognition are rare in both of these small parties. The current electoral system for the House of Representatives (the lower chamber, henceforth HR) has been in use since 1996. Previously, members of the HR were elected entirely under the “hyper-personalistic” (Shugart, 2001) single non-transferable vote (SNTV) system in multimember districts. Under SNTV, intraparty competition within the larger parties, especially the LDP, undermined the value of party label for earning votes. Under the new mixed-member system, intraparty competition has been eliminated, and elections have become relatively more party-centered than in the past (Reed, Scheiner and Thies, 2012). Overall, however, voters still regard politicians’ personal attributes as an important factor in choosing a candidate, regardless of which party they support.7 Intraparty competition is still a feature in the mixed-member system of the House of Councillors (the upper chamber, 7

For example, in a nationwide survey administered by the Association for Promoting Fair Elections

(Akarui Senkyo Suishin Kyokai ) after the 2014 HR election, 29.9% of respondents stated that they cast their vote in the district-level competition based on candidates, rather than parties, and 18.7% stated that their vote was not unconditionally based on either party or candidate.

6

henceforth HC).

2.2

The Mixed-Member Systems of the Diet

For the HR, 295 members are elected by a first-past-the-post (FPTP) system in singlemember districts (SMDs) contained within 47 prefectures. Another set of 176 members are elected through a closed-list PR (CLPR) system in one of eleven regional districts that overlap with the SMDs.8 Voters cast their ballots in the SMD tier by physically writing out the name of a candidate. In the CLPR tier, voters instead write the name of a party. Candidates may be dual-listed in both tiers; if a dual-listed candidate loses the SMD contest, he or she might still be elected through the party’s PR list. The media have sometimes (pejoratively) dubbed these MPs “zombies,” since they “died” in the SMD but were “brought back to life” in PR. The mixed-member system for the HC combines 146 seats in districts that are geographically coterminous with the 47 prefectures, and 96 seats filled through an open-list PR (OLPR) contest in a single, nationwide contest.9 In the prefecture-based races, members are elected through SNTV with the number of seats ranging from 1 to 5.10 Voters cast their ballot for a single candidate, as in the SMD tier of the HR, and seats are allocated in descending order of vote share. In the OLPR tier, voters can choose to write the name of a candidate or a party. Seats are allocated first to parties, then to candidates on those parties’ lists in descending order of preference votes until all party seats are filled. The dual-listing 8

The number of SMDs was 295 when our experiment was conducted in 2015, but was later reduced to

289. District magnitude in the CLPR tier ranged from 6 to 29 at the time of our experiment, and the total number of seats was 180 (later reduced to 176). The seat allocation formula is D’Hondt. 9

Members of the HC are elected to staggered, six-year fixed terms, with half up for re-election every three

years. Prior to the 2016 HC election, four rural prefectures were combined to create two merged districts. 10

For prefectures that return only one member each election, this tier is effectively equivalent to FPTP.

At the time of our survey, the largest prefecture, Tokyo, had 5 seats; this was increased to 6 prior to the 2016 HC election.

7

provision is not used in the HC. Japan’s mixed-member bicameral parliament thus features four different electoral systems (FPTP, SNTV, CLPR, OLPR) with districts of four different geographical sizes: subprefectural area (FPTP in HR), entire prefecture (SNTV in HC), regions made up of multiple prefectures (CLPR in HR), and the entire country (OLPR in HC). Furthermore, voters have different targets of choice: a candidate (FPTP, SNTV), a party (CLPR), or either a candidate or a party (OLPR). In our experimental design, we prime respondents with these distinct variations in the electoral rules.

2.3

Observed Attributes of Politicians

Existing theories would predict that the observed attributes of politicians should vary across Japan’s four electoral contexts (e.g., Moser and Scheiner, 2012; Shugart, Valdini and Suominen, 2005). Specifically, we should expect to observe more women in the OLPR and CLPR tiers, and more politicians with strong personal-vote-earning attributes — such as prior political experience, celebrity status, or dynastic family ties — where vote choice is structured around candidates (FPTP, SNTV, and OLPR) rather than parties (CLPR). Moreover, the importance of these attributes should increase with the intraparty competition produced by the larger district magnitude of SNTV and OLPR, which increases the informational demands on voters, and in turn heightens the salience of personal attributes and name recognition as heuristic devices. To examine these predictions, we collected information on the personal attributes of the 433 candidates who ran in the 2013 HC election and the 1,191 candidates who ran in the 2014 HR election (the two elections immediately preceding the timing of our experiment in 2015).11 For space reasons, here we report just a few notable patterns in the observed 11

For each candidate, we coded party affiliation at the time of the election, previous number of terms

served, incumbency status, age, gender, birthplace, place of education, level of education, prior occupation, and family background in politics, among other common attributes.

8

attributes of candidates and winners across the institutional contexts.12 We first consider gender. Overall, women represent less than a quarter of all candidates and winners across institutional contexts. However, women are more prevalent among MPs in the PR tiers of each chamber than among MPs elected in the smaller districts (although the difference is not statistically significant for the HC); and between SNTV and SMD, the proportion of women is larger for SNTV. These patterns are consistent with theoretical expectations. It is also notable that women are more common among “zombies” than among SMD winners. With regard to age, of note is that Japanese candidates and politicians tend to be on the older side. Only about 10-20% of candidates and politicians are less than 40 years old. Between tiers, PR tends to feature older candidates and winners, particularly in the HC. If we consider age as a proxy for prior experience, which is an important personal-vote-earning attribute, then this pattern would not be consistent with expectations. But it might be due to the fact that terms are longer in the HC and re-election rates are higher in PR systems. This attribute exemplifies the difficulty of untangling the relevance of personal attributes across electoral systems based only on observational data. We next consider two common background careers: local assembly members (including in municipal and prefectural assemblies), and former national bureaucrats. Politicians with local assembly experience — an important measure of local ties as well as experience — are significantly more prevalent in the district tiers of each chamber than in the larger PR tiers. This is consistent with expectations. The exception is that local politicians appear as candidates (though not winners) in the HC at roughly the same rate (about 20% each) across tiers. Former bureaucrats are more common among candidates and winners in the SMD tier than the PR tier in the HR. If we consider experience as a bureaucrat to be a relevant personal-vote-earning attribute, then this finding is also consistent with expectations. But 12

Descriptive statistics for the complete set of attributes are presented graphically in Figures A.1 through

A.5 in the Online Appendix.

9

there are no apparent patterns across tiers in the HC. Finally, we consider differences in the observational data regarding dynastic ties and celebrity status, two attributes that are likely to increase name recognition, which is important for mobilizing support in candidate-centered elections (under SNTV, in particular). While dynastic family ties might matter most in local districts, celebrity status may be most helpful in mobilizing votes from larger constituencies (i.e., OLPR, CLPR). We collected information on whether each politician was related to a previously elected local politician, national-level politician (who never served in cabinet), or former cabinet minister. Overall, such political legacies are much more prevalent in the district tier of the HR (19% of candidates and 35% of winners, compared to 6-7% for either group in PR). Again, this is consistent with theoretical expectations. In the HC, legacies appear as candidates at equivalent rates (about 10%) across tiers but are more likely to appear in the district tier among winners (31%, compared to just 7% for PR). We define a celebrity as any candidate with a high profile from a previous career in television, movies, music, comedy, sports, news, and so on (in Japan often described generally as a “talent”). The observational data show that such candidates are most common in the OLPR tier of the HC (10% of winners), where their national name recognition is presumably an electoral asset; however, the difference across tiers is not statistically significant.

3

Identification Problems

The patterns in the observational data appear to provide support for the theoretical expectations in the literature regarding the relationship between electoral systems and politicians’ personal attributes. However, such observational data are limited in important ways for understanding the actual preferences of voters for different personal attributes in candidates and MPs. First, every politician has multiple attributes, including age, gender, experience, prior

10

occupation, and so on, and usually runs with a party label that also represents a number of substantive policy concerns of voters (Horiuchi, Smith and Yamamoto, 2018). The sheer number of attributes that are highly correlated with one another makes it virtually impossible to identify the effects of those individual attributes from the observational data alone. For example, although incumbents tend to do well in elections, it is often not clear how much of the incumbency advantage is due to incumbents’ personal attributes (such as experience) versus their party labels (e.g., Carson, Engstrom and Roberts, 2007; Fowler and Hall, 2014). The fact that many attributes of interest to voters may be correlated makes it a challenge for researchers, let alone party actors involved in candidate selection, to determine what kinds of attributes voters actually prefer. In our observational data, for example, age and prior incumbency are highly correlated, as are local birth and local political experience. Candidates and MPs from the LDP are more often dynastic than in other parties, but they also tend to be older and are more likely to be born in the district they represent. Similarly, the membership of the DPJ and JCP includes more women. These party differences make it difficult to determine whether voters valued the personal attribute (e.g., gender or dynastic family ties), or whether they supported the party, and the party’s candidate just happened to have these attributes.13 An additional problem with observational data is that the observed attributes of politicians are in part a reflection of which eligible citizens attempt to run for office and who among them are selected by parties. The nature of political life or the structure of political opportunity may be more suited to certain socio-demographic backgrounds than others (e.g., Iversen and Rosenbluth, 2010). As a result, even if we observe that certain attributes are common among elected politicians in particular contexts, it could be because voters preferred candidates with those attributes, or because voters lacked the choice of candidates with alternative attributes. 13

Figure A.6 in the Online Appendix graphically presents correlation matrices for several of the common

personal attributes of candidates and MPs in the HR and HC.

11

A third issue with many of these attributes is the likely correlation with increased name recognition, which helps to win election in candidate-centered systems (e.g., Kam and Zechmeister, 2013), and may help secure higher list positions in party-centered systems. As a result, it is unclear whether voters value these attributes in and of themselves, or are influenced by the heuristic shortcut of a familiar name. Consider dynastic family ties. The name recognition inherent in this attribute may be electorally valuable in candidate-centered elections, but voters may also care about other qualities signaled by dynastic ties in the abstract, such as familiarity with politics or a stable connection to the community, which are related to the kind of representation that might be expected from such a politician. A related, but converse, argument might be made for celebrity status: voters might choose celebrities simply because of the familiar name, or because celebrity status signals something additional, such as an “outsider” perspective. In sum, the existing literature has not yet thoroughly explored the multidimensional preferences of voters for politicians’ personal attributes, nor do we have a good understanding of which attributes are most preferred by voters in Japan. Moreover, despite strong theoretical reasons to expect variation in attributes across electoral contexts, and the observational evidence presented here and in previous empirical studies, the existing literature has produced relatively scant empirical evidence of the mechanisms behind this variation. The observed variation could stem from voters holding differentiated preferences for politicians’ attributes depending on their understanding of the electoral institutions in use and the kind of representation they desire in those contexts. Alternatively, it could stem from the ways in which other factors, such as party recruitment strategies or heuristic shortcuts in voting, shape stable voter preferences into divergent representational outcomes.

12

4

Survey Design

We aim to answer the following two questions: First, which attributes of politicians are most preferred by Japanese voters? Second, do voters prefer different types of politicians depending on their understanding of the representational contexts generated by electoral rules? Before explaining our experimental design, it is important to note that our design is not intended to serve as a realistic simulation of the actual conditions and processes of voting under the different electoral contexts that exist in Japan. Rather, it is specifically designed to activate respondents’ factual understanding of the electoral rules and investigate how such activation affects the way they evaluate politicians’ attributes in different representational contexts. In this regard, it is also important to note that we did not put our experiment in the context of any specific election that happened in the real world. This is advantageous for measuring respondents’ preferences for politicians’ attributes per se, without inadvertently encouraging them to associate the hypothetical politicians presented in our experiments with any actual politicians.

4.1

Random Assignment of Respondents to Groups

Our sample of respondents consists of 2,200 Japanese citizens of eligible voting age (20 and older).14 We randomly assigned respondents to one of four groups: the first group was exposed to information about the SMD tier of HR elections, and then to information about the PR tier; the second group was exposed to the same set of information but in the reverse order; the third group was exposed to information about the SNTV tier of HC elections, 14

Respondents were recruited by Qualtrics Panels between November 10 and December 5, 2015. Our

sample is not a probability sample, but is similar to population demographics in terms of age, gender, prefecture of residence, income, and education. We corrected for slight imbalances using post-stratification weights obtained via entropy balancing (Hainmueller, 2012). The results are similar regardless of whether we use these post-stratification weights. The voting age was lowered to 18 in 2016.

13

and then to information about the PR tier; and the fourth group was exposed to the same set of information but in the reverse order. We prepared two groups of respondents for each chamber (HR or HC) to avoid any potential ordering effect. After the information was presented, each respondent undertook five conjoint tasks, making the total number of tasks per respondent ten overall. For example, respondents in the first group were first asked to read the following sentences that highlight the electoral system, the geographical unit/size of the district, and how voters cast a ballot: The House of Representatives uses two electoral tiers: single-member district and proportional representation. For the single-member district tier, the entire country is divided into 295 districts, and voters write the name of a candidate. The candidate with the most votes wins. On the next five screens, you will see tables featuring hypothetical politicians (or aspiring politicians) in the single-member district tier. Please examine each table carefully before answering the questions that follow. After completing the first set of five conjoint tasks, respondents were asked to read the following sentences: Thank you very much. For the second tier of House of Representatives elections, proportional representation, the entire country is divided into 11 districts, and voters write the name of a party. Seats are allocated to parties based on the party vote. On the next five screens, you will see tables featuring hypothetical politicians (or aspiring politicians) in the proportional representation tier. Please examine each table carefully before answering the questions that follow. Respondents in the second group were given the same set of information about the electoral systems for the HR, but in the reverse order. Respondents in the third and fourth groups 14

were given similar information for the electoral systems used in the HC.15

4.2

Randomized Attributes and Levels

In each of the ten conjoint tasks, respondents compared two hypothetical politicians and answered the following question: “Which of the following two persons do you think is more desirable as a [

]? Even if you are not entirely sure, please indicate which of the two you

would prefer.”16 The phrase within the square brackets was either “single-member district member of the House of Representatives,” “proportional representation member of the House of Representatives,” “district member of the House of Councillors,” or “proportional representation member of the House of Councillors.” Note that we emphasized the words associated with the electoral system so that respondents would be sufficiently primed with the information. Each hypothetical politician’s profile featured nine attributes.17 The order of these attributes was randomized for each respondent to avoid any possible order effect.18 The “levels” (specific attributes) were also randomly assigned for each profile. The randomization of attributes allows us to make causal inferences about which of these attributes are more relevant in respondents’ choice of politicians, and which specific levels on those attributes are more positively or negatively evaluated. Specifically, our design follows the approach proposed by Hainmueller, Hopkins and Yamamoto (2014), whose fully-randomized conjoint analysis non-parametrically identifies the average marginal component effect (AMCE) for each of the attribute levels on the probability of choosing a profile.19 15

The exact wording for these groups is available in Online Appendix B .

16

Our conjoint tasks were therefore forced-choice tasks, which have been found to perform well (Hain-

mueller, Hangartner and Yamamoto, 2015). 17

Table B.1 in the Online Appendix lists the entire set of attributes and levels.

18

To avoid cognitive strain, the order was fixed across all ten exercises for the same respondent.

19

The AMCEs are defined given a set of baseline attributes. For our purposes, the baseline profile represents

a politician who has the least opportunity cost of running for office and the least resources and experiences to

15

Party has five levels corresponding to the four main parties discussed in Section 2.1, along with “Independent.”20 Including party label along with the other personal attributes in our profiles allows us to evaluate the relative value of party for Japanese voters when evaluating politicians. The other attributes in the profiles are composed of relevant personal attributes for the Japanese context, which we will discuss in the next section. Hometown indicates whether the politician was born “Inside” or “Outside” the respondent’s prefecture of residence.21 Highest educational attainment, Prior occupation, and Parental political background also include levels that relate to a politician’s local ties and prior experience. Two additional attributes are more directly relevant to Experience in politics: whether the person is a current or former member of the House of [Representatives/Councillors] and Number of previous terms. Finally, we include two demographic attributes: Age and Gender. As these attributes in the real world are correlated with others, such as experience, it is important to add them in our conjoint analysis.

4.3

Theoretical Expectations

Existing research suggests that voters might view some personal attributes more favorably than others. For example, a considerable amount of survey research has investigated voters’ attitudes toward female candidates (e.g., Aguilar, Cunow and Desposato, 2015; Dolan, 2010; be a legislator. Two of our attributes, gender and hometown, could arguably be set differently given this rule, but since each includes only two levels, any change in the baseline level would not affect our interpretation of the results. 20

Independents cannot win a seat in the PR tier of the HR or HC. Once party-affiliated candidates are

elected, however, they can choose to be independents, or they might be expelled from their party (and become independents). Since we asked respondents to assess the desirability of politicians rather than candidates, it is not unrealistic to use “independent” as a level for Party in the PR context. See also footnote 4. 21

We used the actual name of the respondent’s prefecture of residence as reported in an earlier question

in the survey. This applies also for prefecture-specific prior occupations (assembly member and governor), and parental political background.

16

Schwindt-Bayer, Malecki and Crisp, 2010). Although results across different contexts are mixed, voters are often less informed about women than men as candidates, and may rely on negative stereotypes of how women will perform in office (e.g., Huddy and Terkildsen, 1993; Sanbonmatsu, 2002). As a result, women often must be more qualified and work harder in order to reach the same level of support from voters and parties (e.g., Anzia and Berry, 2011). At first glance, one might presume that Japanese voters have strongly unfavorable attitudes toward female politicians, because Japan ranks last among the OECD countries in gender representation.22 However, it is not clear whether this stems from voters’ negative attitudes toward women in politics, or institutional and structural barriers to recruitment (e.g., Iversen and Rosenbluth, 2010). It may also be possible that gender cues matter less (or more) once bundled together with other relevant attributes and indicators of quality, such as political experience. Experience as an attribute of “quality” is often measured by incumbency or prior service in local assemblies (e.g., Jacobson, 1989). Although a large part of the incumbency advantage is due to name recognition and the deterrence of high-quality challengers (e.g., Carson, Engstrom and Roberts, 2007; Hall and Snyder, 2015), voters may also prefer incumbents for the experience that they bring to representing the needs of the district or area. In other words, voters might value incumbency or length of service in the abstract, setting aside evaluations of the actual individuals who are incumbents. In the context of Japanese politics, in addition to the number of previous terms in office, political experience or “quality” in general could be measured by the types of prior occupations held, such as prior service in the national bureaucracy (Scheiner, 2006), or even by age (given the role seniority plays in Japanese politics). To the extent that voters value local ties, politicians who are born in the voters’ district 22

The Inter-Parliamentary Union, Women in National Parliaments, as of October 1, 2017 (source: http:

//archive.ipu.org/wmn-e/classif.htm).

17

or region, or who have served in local elective office, might be preferred over “outsiders” (e.g., Campbell and Cowley, 2014; Tavits, 2010). In the context of Japanese politics, local ties could be measured by the place of birth, dynastic family ties, or to some extent by education (e.g., a graduate of a local public university). Local assembly experience has been also used in previous studies as a key measure of both local orientation and the personal vote. Voters in Japan may therefore be expected to exhibit some baseline preferences for or against politicians based on gender or several of these indicators of experience and local ties. The conjoint nature of our experiment will in part reveal which particular attributes are most important. But the institutional variation present in Japan might also produce different degrees of relevance or value for these attributes. Given the relationship between PR and gender representation, for example, we might expect to observe more favorable attitudes toward hypothetical female politicians who would represent them under OLPR and CLPR rather than SNTV or FPTP.23 We might also expect that the perceived value of some personal attributes, such as local ties, prior political experience, or a dynastic family relationship to a previous politician, should be higher where voters are primed to consider politicians representing them in contexts where elections are candidate-centered (FPTP, SNTV, and OLPR) rather than party-centered (CLPR). Moreover, where electoral districts are geographically smaller, such as SMDs in the HR and prefectural SNTV districts in the HC, we might expect stronger demand among voters for a politician with local ties to that district. Finally, with regard to the relevance of personal attributes of politicians vis-`a-vis their party affiliation, in the context of Japanese politics, we might expect voters in general to prefer politicians with no party affiliation. This is because the majority of voters in Japan 23

Previous research on how preferential voting (i.e., comparing OLPR to CLPR) affects gender represen-

tation has produced mixed results, and is mostly based on cross-national analyses of observational data or traditional surveys (e.g., Matland and Lilliefeldt, 2014; Thames and Williams, 2010).

18

do not support any party.24 Yet, between different electoral system contexts, party-affiliated politicians might be more preferred under CLPR and OLPR, compared to FPTP and SNTV, where politicians’ personal attributes, in particular local ties, should be more relevant to the kind of representation a voter might expect from a politician.

5

Results

We first show pooled results using all 2,200 respondents. We then show how the estimated effects differ by the priming of different electoral system contexts.

5.1

Overall Results

Figure 1 shows our estimated AMCEs for the pooled sample, where the solid circles represent the point estimates and the horizontal bars represent the 95% confidence intervals robust to clustering at the respondent level. One of the most remarkable patterns is that, as expected, Japanese voters on average strongly prefer independent politicians unaffiliated with any political party over party-affiliated politicians. Compared to the baseline level of Party (“Independent”), the effects of party labels (“DPJ,” “JCP,” “Komeito,” and “LDP”) are all negative and statistically significant. The effects of “JCP” and “Komeito” are particularly large and negative. Although these parties enjoy strong support from their core supporters, neither is popular with the general electorate. Moving beyond party label, which personal attributes do Japanese voters care most about? Although gender representation is comparatively low in Japan, the results of our analysis indicate that gender is, on average, irrelevant when voters consider the desirability of a hypothetical politician. Compared to the baseline category (“Male”), the effect of “Female” is almost zero and not at all significant. In contrast, voters appear to have strong preferences regarding age. Compared to the baseline level of Age (“30”), the effect of “42” is 24

In our survey, 76% of respondents considered themselves to be independents.

19

All Respondents (N=2,200) Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

● ● ● ●

● ● ● ● ● ●

●

● ● ●

●

● ● ●

● ● ● ●

● ● ● ● ● ●

−0.2

0.0

0.2

Change in E[Y]

Figure 1: Average effects of hypothetical politicians’ attributes on respondents’ preference. Note: Each solid circle in the plot represents the estimated average marginal component effect (AMCE) of an attribute-level on a respondent’s probability of choosing a hypothetical politician containing that attribute-level, compared against a politician with the baseline level for the same attribute. The horizontal bars represent 95% confidence intervals robust to clustering at the respondent level. 20

not statistically significant, but the effects of “57”, “64”, and “79” are all negative and statistically significant. The effect of “79” is particularly large. These results are not necessarily consistent with the observed attributes of actual politicians reported in Section 2.3.25 Although younger politicians are generally more preferred, this does not imply that experience is irrelevant. The effects for the various levels of Experience show that politicians who formerly or currently hold office are more positively assessed by voters than the baseline level of “No experience.” The effects tend to be larger for “Formerly in office” than for “Currently in office” and larger as the number of past victories increases. This seemingly contradictory result — that Japanese voters prefer younger but experienced politicians — demonstrates the effectiveness of decomposing the relevance of personal attributes through conjoint analysis. These results may explain, for example, the growing popularity of Shinjiro Koizumi, the son of former prime minister Junichiro Koizumi. Although he was only 36 years old at the time of our experiment, he had already won election four times. We need to be cautious, however, in making such an interpretation. He may be popular not because he is young and experienced but because he is a son of a former prime minister. Dynastic politicians are often able to enter politics at a younger age, bypassing traditional stepping stones to office such as local assembly experience (e.g., Smith, 2018). The estimated effects of Parental Political Background should help to further decompose the effects of these attributes. Compared to the baseline category of “None,” the effects of “Minister,” “National assembly member,” and “Prefectural assembly member” are all small. The effect of “Minister” is negative and barely significant at the 5% level, but the effects of the other two levels are almost zero and insignificant. Contrary to the empirical evidence of a vote advantage for members of Japanese dynasties (Asako et al., 2015), these results suggest that, in the abstract, dynastic ties are not necessarily of inherent value to Japanese voters. What if a politician is a celebrity? Celebrity politicians are increasingly common in 25

See also the Online Appendix.

21

Japanese politics, and some have even taken key leadership roles in parties. Although political parties are increasingly active in nominating such celebrity candidates, the effect of “Celebrity” for Prior Occupation, compared to the baseline level of “Business employee,” is negative and statistically significant. The effect is also quite large. Other prior occupations seem to be irrelevant for Japanese voters. Specifically, compared to the baseline level of “Business employee,” the effects of “Business executive,” “Local government employee,” “National government employee,” “Prefectural assembly member,” and “Prefectural governor” are all small and statistically insignificant. It is possible, however, that some of these occupations that are geographically-specific are relevant under some electoral contexts but not others. We will examine this possibility in the next section. We included two other attributes that are related to politicians’ “localness.” The effect of “Local public university” for Highest Educational Attainment is positive compared to the baseline level of “High school.” The effect, however, is small and barely significant. The effects of “Prestigious private university” and “University of Tokyo” are also positive compared to the baseline, and the effect of “University of Tokyo” is large and statistically significant. Japanese voters appear to consider university graduates as more desirable, but going to a local university is not particularly relevant, at least on average. Again, it is possible that the effect of “Local public university” is larger when respondents are asked to consider the desirability of politicians elected from geographically small areas. Finally, for Hometown, the effect of “Outside” as compared to the baseline of “Inside” is negative and significant. Even when provided with information about other attributes, such as the politicians’ age, experience, and party affiliation, Japanese voters still prefer politicians who are locals.

5.2

Effects by Priming of Electoral Context

In our experiment, each respondent was primed with information on the electoral rules, district sizes, and voting procedures used in Japan’s mixed-member bicameral system. Did 22

this priming of the electoral process affect voters’ preferences for politicians? Figure 2 shows the results of our tests when divided into separate groups by priming treatment. The top set of panels compares SNTV versus PR in the HC; the bottom set compares SMD versus PR in the HR. In each set, the far-right panel shows the point estimates of the differences in AMCEs for each attribute level between the two experimental (electoral tier) conditions, along with their 95% confidence intervals. Since each respondent was given both sets of information in 5 × 2 conjoint exercises, this figure compares the difference in attitudes within respondents. The results are unambiguous: all four of the leftmost panels are very similar to the results shown in Figure 1. In addition, the difference between the priming conditions is statistically significant for only one out of 56 comparisons. Given the large number of hypotheses being tested in this analysis, we interpret our single statistically significant estimate not as evidence of a true difference between the two conditions, but as stochastic noise. We also separately compared the results with respondents pooled across chambers (HR versus HC) and pooled across tier types (small district versus PR); again, we found no statistically significant difference for either of these comparisons.26 We conducted several additional robustness checks to verify that this null finding indeed reflects an absence of a priming effect, rather than a problem with the experiment design or sample.27 First, we estimated the difference between the priming conditions using only the respondents who passed a set of manipulation checks (a knowledge test of the electoral rules we primed) placed immediately after the conjoint tasks in the survey. The results were virtually identical, with none of the 56 comparisons indicating a statistically significant difference between the conditions. This indicates that the lack of difference across the conditions is not due to respondents failing to understand the priming information. Similarly, to see if our result stems from lack of engagement with the survey among respondents, we ran the same 26

These additional results are available in Figure D.1 in the Online Appendix.

27

These robustness checks are presented in Figure D.2, Figure D.3, and Figure D.4 in the Online Appendix.

23

SNTV Tier Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

●

Comparison

●

●

● ● ● ●

●

●

●

●

●

●

● ● ●

● ● ● ●

● ● ● ● ● ●

● ● ●

●

●

●

● ● ●

●

● ● ●

●

● ● ●

●

● ● ●

●

● ● ●

●

● ● ●

●

● ●

● ● ● ●

● ● ●

●

●

●

●

●

● ● ●

● ● ●

● ● ●

● ●

−0.2

Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

PR Tier

● ●

●

0.0

0.2

−0.2

0.0

0.2

−0.2

0.0

Change in E[Y]

Difference in Changes

SMD Tier

PR Tier

Comparison

●

● ●

● ●

● ● ●

●

●

●

●

● ●

●

● ● ● ●

● ● ● ● ●

● ● ● ●

●

● ●

●

● ● ●

●

● ●

● ● ●

●

●

●

● ● ●

●

● ●

● ●

●

●

●

●

● ●

● ● ● ●

● ● ●

●

●

●

●

● ● ● ● ● ●

● ● ● ● ●

−0.2

0.2

Change in E[Y]

0.0

● ● ● ●

0.2

−0.2

Change in E[Y]

0.0

Change in E[Y]

0.2

−0.2

0.0

0.2

Difference in Changes

Figure 2: Effects of politicians’ attributes on respondents’ preference, by priming conditions. Note: Top panels give the results for the HC (SNTV versus PR priming); Bottom panels give the results for the HR (SMD versus PR priming). The third panel for each shows the difference in the estimated AMCEs. 24

analysis on the subsample of respondents who passed a set of screener questions included towards the end of the survey, and found equivalent results, with only two out of the 56 differences estimated to be statistically significant. Finally, we investigated the possibility that those who participate in elections may have distinct and context-contingent preferences, whereas the general voting-age population, which includes non-voters, may not. We find no evidence for this: our null result is replicated when we analyze the subsample of respondents who reported that they had always or almost always voted in past elections. In sum, our analyses consistently show that there is no difference between the experimental conditions using the electoral system priming in any of the AMCEs.28

6

Discussion

The results of our conjoint experiment illustrate that Japanese voters have strong preferences for particular types of politicians. Importantly, despite the fact that there are many elderly, celebrity, dynastic, and male politicians in Japan, voters do not appear to prefer older politicians or celebrities, and are indifferent with regard to dynastic family ties and gender. Furthermore, the clear lack of difference in the AMCEs between the experimental conditions of our survey suggests that Japanese voters prefer the same types of politicians even when asked to consider the kind of representation they desire under different electoral system contexts. This stands in contrast to the patterns in observed attributes described in Section 2.3 and the theoretical expectations discussed in Section 4.3. These discrepancies might be thought of as “representational gaps” between voters’ preferences and the actual attributes of politicians. Yet, caution is warranted in interpreting such gaps. First, the discrepancies do not necessarily imply that voters’ preferences are insufficiently 28

We also note that our sample size is large enough to statistically detect even the smallest difference that

we consider to be substantively meaningful. In other words, our null finding is not due to a lack of statistical power.

25

reflected in real-world electoral processes in Japan. For example, voters may not care about politicians’ gender per se according to our estimates of AMCEs, despite the fact that most Japanese politicians are male. This may be because voters evaluate the attributes of politicians differently depending on the gender of those politicians. More generally, gender — or any other attribute for that matter — might interact in important ways with many other attributes, and such interactions may be one source of the ostensible inconsistency between voters’ preferences and the observed attributes of politicians.29 To investigate this possibility, we estimate the average component interaction effects (ACIEs) as proposed by Hainmueller, Hopkins and Yamamoto (2014). Among many possible interactions, we focus on gender because it is increasingly relevant to parties’ candidate recruitment strategies, even as female politicians remain comparatively underrepresented. The results presented in Figure 3 provide a hint as to how a politician’s gender might condition voter preferences for different attributes. The rightmost panel shows the ACIEs with respect to gender and each of the other attributes, along with their 95% confidence intervals. The key result is that hypothetical male politicians get a significantly larger “boost” from political experience than their female counterparts. That is, the positive effect of having been in office for two or more terms as opposed to no experience in office is significantly smaller for a female politician than a male politician. This may in part explain why women continue to be underrepresented in Japanese politics. Female candidates may win in an election when they are first-timers, particularly when they are young (and depending on their party’s popularity). But at subsequent stages of reelection and promotion, our analysis of ACIEs suggests that women with political experience might not reap the same amount of voter support as men with experience. Japan is known to be a country with a particularly tough environment for working women. This so-called “bamboo ceiling,” which exists not only in business environments but also in politics, may 29

Such interaction effects can be thought of as capturing the idea of “intersectionality” in identities, which

has received increasing attention in sociology and political science (e.g., Hughes, 2013).

26

Female Politician Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

Male Politician

●

Comparison

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

● ●

●

●

● ●

0.0

●

●

●

−0.2

● ●

●

●

●

0.2

−0.2

Change in E[Y]

0.0

Change in E[Y]

●

0.2

−0.2

0.0

0.2

Difference in Changes

Figure 3: Average effects of hypothetical politicians’ attributes on respondents’ preference, given the hypothetical politician’s gender. Note: The solid circles in the left and middle panels represent the estimated AMCEs of a hypothetical politicians’ attributes other than gender, conditional on whether the politician is female (left) or male (middle). The rightmost panel shows the average component interaction effects (ACIEs) with respect to gender and each of the other attributes. The horizontal bars represent 95% confidence intervals robust to clustering at the respondent level.

27

be partly reinforced by voters themselves. Although Figure 3 is suggestive of an important interaction effect, it is unlikely that such interactive effects of attributes can solely account for all of the ostensible gaps between voters’ preferences and the actual attributes of politicians. In fact, the estimates of other ACIEs do not provide clear cues as to why Japanese voters do not prefer elderly and celebrity politicians, or why parental political background is not relevant for voters.30 Rather, the gap between the experimental results and the observational data for the latter two of these variables could be more related to name recognition, which is associated with these attributes in the real world but abstracted out in our experiment. Our survey asked respondents to evaluate hypothetical politicians with a dynastic background or celebrity past, which may signal certain qualities in the abstract — such as “insider” or “outsider” perspectives, respectively. However, our results suggest that voters may not value these potential qualities in and of themselves, and are instead probably drawn to the name recognition these types of politicians tend to enjoy in real-world elections. This explanation would also account for the higher observed prevalence of these types in the electoral contexts (SNTV, SMD, and OLPR) where name recognition can be an important heuristic for voters when choosing between candidates. The gaps may also in part be a function of political parties’ strategies to recruit potential candidates under specific institutional contexts. This points to an important debate in the literature related to different perspectives of representation (Pitkin, 1967). One perspective is that party leaders may want to improve “descriptive representation” and are eager to present voters with candidates representing a microcosm of the socio-demographic variation of society, including gender, race, age, place of residence, level of education, and occupational background. Another perspective, however, is that party leaders may want to improve “substantive representation” and maximize the opportunity to advocate for their policies. In this pursuit, they may not care much about the socio-demographic attributes of politicians, 30

These additional ACIE results are reported in Online Appendix C.

28

and instead focus on strategically recruiting experienced and capable candidates. The discrepancies between the personal attributes of politicians that voters would prefer and the attributes possessed by actual politicians may thus be the result of parties’ placing greater emphasis on their goals of substantive representation. Party choices in nomination may also be constrained by the availability of would-be candidates who seek the nomination. Finally, it is important to note that voters’ preferences may be conditional on what types of politicians are currently in parliament. For example, voters may prefer younger representatives when there are “too many” older ones.31 Voter preferences for politicians’ attributes are unlikely to be fixed (Zaller, 1992). Rather, they may be temporal and contextual. A range of informational contexts, such as the way in which the media portray certain types of politicians at any given time, might also affect voters’ preferences. Our results might thus be interpreted as being contingent on the types of politicians currently in parliament, and the political context at the time we fielded the survey. In sum, although our experimental results point to clear patterns in the attributes that Japanese voters would prefer in the abstract in their elected representatives, the attributes of the politicians who actually emerge and contest elections are likely to be affected by a number of additional factors, including the candidate recruitment processes of parties, the availability of different types of candidates, and the cognitive process of voting under different electoral rules. The lack of any priming effect of the electoral rules at least suggests that voters do not hold differentiated preferences for different types of politicians depending on their understanding of the structure of the electoral system, a potential mechanism behind the observational patterns which has not previously been tested. 31

We thank an anonymous reviewer for suggesting this possibility.

29

7

Conclusion

We have investigated the multidimensional preferences of Japanese voters regarding politicians’ personal attributes via a fully randomized conjoint experiment. Our results reveal that the types of politicians voters would prefer are not necessarily the same as the types of politicians who actually represent them in parliament. Furthermore, these preferences are consistent regardless of whether voters consider the different representational contexts in Japan’s mixed-member bicameral Diet, in contrast to the substantial differences in the observed attributes among the actual candidates and MPs in these contexts. These findings suggest that factors beyond voters’ preferences are likely to play a greater role in determining the nature of representation in Japan. Investigating the various alternative pathways through which representative democracies can aim to reflect voters’ diverse preferences is a promising direction of future research. As we have demonstrated, an effective research design for pursuing these questions is to combine the empirical analysis of observational data of actual politicians with the experimental analysis of voters’ preferences for hypothetical politicians. It should be particularly fruitful to replicate similar research in other democracies with different historical, cultural, and electoral contexts. Another avenue for future research might be to investigate how specific contextual events — such as a high-profile scandal involving a celebrity politician, or the selection of a female politician as party leader — affect the salience of these attributes for voters relative to other attributes. The accumulation of such comparative evidence will be crucial to expanding our understanding of the factors that shape voters’ preferences for politicians, and whether these preferences are effectively represented in modern democracies.

30

References Aguilar, Rosario, Saul Cunow and Scott Desposato. 2015. “Choice sets, gender, and candidate choice in Brazil.” Electoral Studies 39:230–242. Anzia, Sarah F. and Christopher R. Berry. 2011. “The Jackie (and Jill) Robinson Effect: Why Do Congresswomen Outperform Congressmen?” American Journal of Political Science 55(3):478–493. Asako, Yasushi, Takeshi Iida, Tetsuya Matsubayashi and Michiko Ueda. 2015. “Dynastic Politicians: Theory and Evidence from Japan.” Japanese Journal of Political Science 16(1):5–32. Banducci, Susan A., Todd Donovan and Jeffrey A. Karp. 2004. “Minority Representation, Empowerment, and Participation.” The Journal of Politics 66(2):534–556. Bratton, Kathleen A. and Leonard P. Ray. 2002. “Descriptive Representation, Policy Outcomes, and Municipal Day-Care Coverage in Norway.” American Journal of Political Science 46(2):428–437. Cain, Bruce E., John A. Ferejohn and Morris P. Fiorina. 1987. The Personal Vote: Constituency Service and Electoral Independence. Cambridge, MA: Harvard University Press. Campbell, Rosie and Philip Cowley. 2014. “What Voters Want: Reactions to Candidate Characteristics in a Survey Experiment.” Political Studies 62(4):745–765. Carey, John M. and Matthew S. Shugart. 1995. “Incentives to Cultivate a Personal Vote: A Rank Ordering of Electoral Formulas.” Electoral Studies 14:417–440. Carlson, Elizabeth. 2015. “Ethnic Voting and Accountability in Africa: A Choice Experiment in Uganda.” World Politics 67(2):353–385.

31

Carnes, Nicholas and Noam Lupu. 2016. “Do Voters Dislike Working-Class Candidates? Voter Biases and the Descriptive Underrepresentation of the Working Class.” American Political Science Review 110(4):832–844. Carozzi, Felipe and Luca Repetto. 2016. “Sending the Pork Home: Birth Town Bias in Transfers to Italian Municipalities.” Journal of Public Economics 134:42–52. Carson, Jamie L., Erik J. Engstrom and Jason M. Roberts. 2007. “Candidate Quality, the Personal Vote, and the Incumbency Advantage in Congress.” American Political Science Review 101(2):289–301. Crotty, William J. 1968. The Party Organization and its Activities. In Approaches to the Study of Party Organization, ed. William J. Crotty. Boston, MA: Allyn and Bacon pp. 247– 306. Curtis, Gerald L. 1971. Election Campaigning, Japanese Style. New York: Columbia University Press. Curtis, Gerald L. 1979. The Opposition. In A Season of Voting: Japanese Elections of 1976 and 1977, ed. Herbert Passin. Washington, DC: American Enterprise Institute pp. 43–80. Dolan, Kathleen. 2010. “The Impact of Gender Stereotyped Evaluations on Support for Women Candidates.” Political Behavior, Vol. 32, No. 1 (March 2010), pp. 69-88 32(1):69– 88. Fiva, Jon H. and Askill H. Halse. 2016. “Local favoritism in at-large proportional representation systems.” Journal of Public Economics 143:15–26. Fiva, Jon H. and Daniel M. Smith. 2017. “Local Candidates and Voter Mobilization: Evidence from Historical Two-Round Elections in Norway.” Electoral Studies 45:130–140.

32

Fowler, Anthony and Andrew B. Hall. 2014. “Disentangling the Personal and Partisan Incumbency Advantages: Evidence from Close Elections and Term Limits.” Quarterly Journal of Political Science 9(4):501–531. Franchino, Fabio and Francesco Zucchini. 2015. “Voting in a Multi-dimensional Space: A Conjoint Analysis Employing Valence and Ideology Attributes of Candidates.” Political Science Research and Methods 3(2):221–241. Gay, Claudine. 2002. “Spirals of Trust? The Effect of Descriptive Representation on the Relationship Between Citizens and Their Government.” American Journal of Political Science 46(4):717–733. G´orecki, Maciej A. and Michael Marsh. 2012. “Not just ‘friends and neighbours’: Canvassing, geographic proximity and voter choice.” European Journal of Political Research 51(5):563– 582. Grofman, Bernard. 2005. “Comparisons Among Electoral Systems: Distinguishing Between Localism and Candidate-Centered Politics.” Electoral Studies 24(1):735–740. Hainmueller, Jens. 2012. “Entropy Balancing: A Multivariate Reweighting Method to Produce Balanced Samples in Observational Studies.” Political Analysis 20(1):25–46. Hainmueller, Jens, Daniel J. Hopkins and Teppei Yamamoto. 2014. “Causal Inference in Conjoint Analysis: Understanding Multi-dimensional Preferences via Stated Preference Experiments.” Political Analysis 22(1):1–30. Hainmueller, Jens, Dominik Hangartner and Teppei Yamamoto. 2015. “Validating Vignette and Conjoint Survey Experiments against Real-World Behavior.” Proceedings of the National Academy of Sciences of the United States of America 112(8):2395–2400. Hall, Andrew B. and James M. Jr. Snyder. 2015. “How Much of the Incumbency Advantage is Due to Scare-Off?” Political Science Research and Methods 3(3):493–514. 33

Hirano, Shigeo. 2006. “Electoral Institutions, Hometowns, and Favored Minorities: Evidence from Japan’s Electoral Reforms.” World Politics 59(1):51–82. Horiuchi, Yusaku, Daniel M. Smith and Teppei Yamamoto. 2018. “Measuring Voters’ Multidimensional Policy Preferences with Conjoint Analysis: Application to Japan’s 2014 Election.” Political Analysis forthcoming. Huddy, Leonie and Nayda Terkildsen. 1993. “Gender Stereotypes and the Perception of Male and Female Candidates.” American Journal of Political Science 37:119–147. Hughes, Melanie M. 2013. “The Intersection of Gender and Minority Status in National Legislatures: The Minority Women Legislative Index.” Legislative Studies Quarterly 38(4):489–516. Ishibashi, Michihiro and Steven R. Reed. 1992. “Second-Generation Diet Members and Democracy in Japan: Hereditary Seats.” Asian Survey 32(4):366–379. Iversen, Torben and Frances Rosenbluth. 2010. Women, Work, and Politics: The Political Economy of Gender Inequality. New Haven, CT: Yale University Press. Jacobson, Gary C. 1989. “Strategic Politicians and the Dynamics of U.S. House Elections, 1946-86.” The American Political Science Review 83(3):773–793. Kam, Cindy D. and Elizabeth J. Zechmeister. 2013. “Name Recognition and Candidate Support.” American Journal of Political Science 57(4):971–986. Kostadinova, Tatiana. 2007. “Ethnic and Women’s Representation under Mixed Election Systems.” Electoral Studies 26(2):418–431. Matland, Richard E. and Emelie Lilliefeldt. 2014. The Effect of Preferential Voting on Women’s Representation. In Representation: The Case of Women, ed. Maria C. EscobarLemmon and Michelle M. Taylor-Robinson. Oxford, UK: Oxford University Press pp. 79– 102. 34

Moser, Robert G. 2008. “Electoral Systems and the Representation of Ethnic Minorities: Evidence from Russia.” Comparative Politics 40(3):273–292. Moser, Robert G. and Ethan Scheiner. 2012. Electoral Systems and Political Context: How the Effects of Rules Vary Across New and Established Democracies. Cambridge, UK: Cambridge University Press. Nemoto, Kuniaki and Matthew S. Shugart. 2013. “Localism and coordination under three different electoral systems: The national district of the Japanese House of Councillors.” Electoral Studies 32(1):1–12. Pitkin, Hanna F. 1967. The Concept of Representation. Berkeley, CA: University of California Press. Reed, Steven R. 2009. “Party Strategy or Candidate Strategy: How Did the LDP Run the Right Number of Candidates in Japan’s Multi-Member Districts?” Party Politics 15(3):295–314. Reed, Steven R., Ethan Scheiner and Michael F. Thies. 2012. “The End of LDP Dominance and the Rise of Party-Oriented Politics in Japan.” Journal of Japanese Studies 38(2):353– 376. Salmond, Rob. 2006. “Proportional Representation and Female Parliamentarians.” Legislative Studies Quarterly 31(2):175–204. Sanbonmatsu, Kira. 2002. “Gender Stereotypes and Vote Choice.” American Journal of Political Science 46(1):20–34. Schattschneider, E.E. 1942. Party Government. New York, NY: Rinehart & Company. Scheiner, Ethan. 2006. Democracy Without Competition in Japan: Opposition Failure in a One-Party Dominant State. New York, NY: Cambridge University Press.

35

Schwindt-Bayer, Leslie A., Michael Malecki and Brian F. Crisp. 2010. “Candidate Gender and Electoral Success in Single Transferable Vote Systems.” British Journal of Political Science 40(3):693709. Shugart, Matthew Soberg. 2001. “Electoral “Efficiency” and the Move to Mixed-Member Systems.” Electoral Studies 20(1):173–193. Shugart, Matthew Søberg, Melody Ellis Valdini and Kati Suominen. 2005. “Looking for Locals: Voter Information Demands and Personal Vote-Earning Attributes of Legislators under Proportional Representation.” American Journal of Political Science 49(2):437–449. Smith, Daniel M. 2013. Candidate Recruitment for the 2012 Election: New Parties, New Methods. . . Same Old Pool of Candidates? In Japan Decides 2012: The Japanese General Election, ed. Robert Pekkanen, Steven R. Reed and Ethan Scheiner. New York, NY: Palgrave Macmillan pp. 101–122. Smith, Daniel M. 2018. Dynasties and Democracy: The Inherited Incumbency Advantage in Japan. Stanford, CA: Stanford University Press. Smith, Daniel M., Robert J. Pekkanen and Ellis S. Krauss. 2013. Building a Party: Candidate Recruitment in the Democratic Party of Japan, 1996-2012. In Japan Under the DPJ: The Politics of Transition and Governance, ed. Kenji E. Kushida and Phillip Y. Lipscy. Stanford, CA: The Walter H. Shorenstein Asia-Pacific Research Center pp. 157–190. Svaleryd, Helena. 2009. “Women’s Representation and Public Spending.” European Journal of Political Economy 25(2):186–198. Tavits, Margit. 2010. “Effect of Local Ties On Electoral Success and Parliamentary Behaviour The Case of Estonia.” Party Politics 16(2):215–235. Thames, Frank C. and Margaret S. Williams. 2010. “Incentives for Personal Votes and Women’s Representation in Legislatures.” Comparative Political Studies 43(12):1575–1600. 36

Zaller, John. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge University Press.

37

Online Appendix This online appendix of supplementary material presents (A) observational data on candidates and MPs in Japan; (B) additional analyses of the average component interaction effects (ACIEs) for some key attributes; (C) the full text of priming information on electoral system contexts given to each experimental group of respondents, and the full list of attributes and levels; and (D) robustness checks on the main results presented in the text.

A

Observed Attributes

Figures A.1 through A.5 show the observed distributions of politicians’ attributes across electoral system contexts. Data for the House of Representatives (HR) are from the 2014 election; data for the House of Councillors (HC) are from the 2013 election. Attributes are coded based on profiles listed in newspapers and candidate websites during the elections. Each figure is a mosaic plot, which shows the joint distribution of an attribute and electoral system context — SMD (including dual-listed) versus PR for the House of Representatives (HR); SNTV (including SMDs) versus PR for the House of Councillors (HC). For the HR, we separate PR winners who ran as “pure” PR candidates from the dual-listed candidates who lost in SMD but were elected in PR (“zombies”). The area of each rectangular field is proportional to the fraction of observations belonging to the corresponding category. The numbers on the vertical axis represent the cumulative distributions of the variables in the legend conditional on the types of electoral systems shown on the horizontal axis. Figure A.6 presents correlation matrices for several of the common attributes of candidates/MPs in the HR and HC. The top two panels present the correlations among candidates in the HR and HC; the bottom two panels present the corresponding correlations for MPs. In each panel, the cells in the matrix give the numerical values of the correlations. The darker blue (red) cells indicate larger positive (negative) correlations.

A1

(a) Gender HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 11.73 p = 0.001 SMD/Dual

PR

0.25

χ2 = 1.5 p = 0.261

0.00

SNTV

PR Male

HR Winners 1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 10.71 p = 0.006 SMD

Female

HC Winners

Zombie PR

0.25

χ2 = 1.2 p = 0.343

0.00

SNTV

PR

(b) Age HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 14.36 p = 0.005 SMD/Dual

PR

0.25

χ2 = 13.78 p = 0.009

0.00

SNTV

PR

70 or over 60 − 69 50 − 59

HR Winners

HC Winners

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 31.05 p = 0.000 SMD

Zombie PR

40 − 49 Under 40

0.25

χ2 = 9.17 p = 0.055

0.00

SNTV

PR

Figure A.1: Personal attributes across candidates, winners, and institutional contexts (1).

A2

(a) Local Assembly Experience HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 11.66 p = 0.002 SMD/Dual

PR

0.25

χ2 = 0.02 p = 1.000

0.00

SNTV

PR No

HR Winners 1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 6.5 p = 0.035 SMD

Yes

HC Winners

Zombie PR

0.25

χ2 = 7.24 p = 0.006

0.00

SNTV

PR

(b) Former National Bureaucrat HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 7.35 p = 0.008 SMD/Dual

PR

0.25

χ2 = 0.59 p = 0.457

0.00

SNTV

PR No

HR Winners 1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 9.28 p = 0.011 SMD

Zombie PR

Yes

HC Winners

1.00

0.25

χ2 = 0.16 p = 0.794

0.00

SNTV

PR

Figure A.2: Personal attributes across candidates, winners, and institutional contexts (2).

A3

(a) Dynastic Family Ties HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 23.41 p = 0.000 SMD/Dual

PR

0.25

χ2 = 2.01 p = 0.578

0.00

SNTV

None

PR

Loc. Pol.

HR Winners 1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 44.8 p = 0.000 SMD

MP

HC Winners

Zombie PR

Cab. Min.

0.25

χ2 = 8.35 p = 0.031

0.00

SNTV

PR

(b) Celebrity Status HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 0.28 p = 0.635 SMD/Dual

PR

0.25

χ2 = 2.41 p = 0.147

0.00

SNTV

PR No

HR Winners 1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 1.92 p = 0.431 SMD

Zombie PR

Yes

HC Winners

1.00

0.25

χ2 = 0.49 p = 0.504

0.00

SNTV

PR

Figure A.3: Personal attributes across candidates, winners, and institutional contexts (3).

A4

(a) Education HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ = 6.78 p = 0.079 2

SMD/Dual

PR

0.25 0.00

χ2 = 3.54 p = 0.342 SNTV

University of Tokyo

PR

Other public university

HR Winners 1.00

0.75

0.75

0.50

High school

0.50

0.25 0.00

Private university

HC Winners

1.00

χ2 = 28.39 p = 0.000 SMD

Zombie PR

0.25 0.00

χ2 = 6.96 p = 0.078 SNTV

PR

(b) Experience HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ = 46.32 p = 0.000 2

SMD/Dual

PR

0.25 0.00

χ2 = 32.47 p = 0.000 SNTV

PR

Currently in office, 3+ terms Formerly in office, 3+ terms Currently in office, 2 terms Formerly in office, 2 terms

HR Winners

HC Winners

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ2 = 133.04 p = 0.000 SMD

Zombie PR

Formerly in office, 1 term No experience

0.25 0.00

Currently in office, 1 term

χ2 = 11.64 p = 0.052 SNTV

PR

(c) Party HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ = 361.72 p = 0.000 2

SMD/Dual

PR

JCP

0.25 0.00

χ2 = 56.96 p = 0.000 SNTV

SDP DPJ

PR

Komeito

HR Winners

LDP

HC Winners

1.00

1.00

0.75

0.75

0.50

0.50

JIP Other

0.25 0.00

χ2 = NA p = NA SMD

Zombie PR

Independent

0.25 0.00

χ2 = NA p = NA SNTV

PR

Figure A.4: Personal attributes across candidates, winners, and institutional contexts (4).

A5

(a) Secretary for MP

(b) Local Level Bureaucrat

HR Candidates

HC Candidates

HR Candidates

HC Candidates

1.00

1.00

1.00

1.00

0.75

0.75

0.75

0.75

0.50

0.50

0.50

0.25 0.00

χ = 8.2 p = 0.003 2

SMD/Dual

PR

0.25 0.00

χ = 0.39 p = 0.575 SNTV

0.50

0.25

2

0.00

PR

χ = 4.87 p = 0.031 2

SMD/Dual

PR

0.25 0.00

χ2 = 4.5 p = 0.056 SNTV

PR

No

HR Winners

No

Yes

HC Winners

HR Winners

1.00

1.00

1.00

1.00

0.75

0.75

0.75

0.75

0.50

0.50

0.50

0.25 0.00

χ2 = 12.99 p = 0.002 SMD

Zombie PR

0.25 0.00

SNTV

0.50

0.25

χ2 = 3.23 p = 0.106

0.00

PR

(c) Business Employee

χ2 = 7.6 p = 0.021 SMD

Zombie PR

0.25 0.00

HR Candidates

HC Candidates

HR Candidates 1.00

1.00

0.75

0.75

0.75

0.75

0.50

0.50

0.50

0.00

χ = 0.03 p = 0.875 SMD/Dual

PR

PR

HC Candidates

1.00

2

χ2 = 3.54 p = 0.086 SNTV

(d) Business Executive

1.00

0.25

Yes

HC Winners

0.25 0.00

χ = 1.69 p = 0.207 SNTV

0.50

0.25

2

0.00

PR

χ = 1.1 p = 0.342 2

SMD/Dual

PR

0.25 0.00

χ2 = 1.06 p = 0.333 SNTV

PR

No

HR Winners

No

Yes

HC Winners

HR Winners

1.00

1.00

1.00

0.75

0.75

0.75

0.75

0.50

0.50

0.50

0.25 0.00

χ2 = 0.98 p = 0.619 SMD

Zombie PR

0.25 0.00

SNTV

0.50

0.25

χ2 = 0.42 p = 0.653

0.00

PR

Yes

HC Winners

1.00

χ2 = 7.71 p = 0.019 SMD

Zombie PR

0.25 0.00

χ2 = 0.29 p = 0.782 SNTV

PR

(e) Hometown HR Candidates

HC Candidates

1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

χ = 6.25 p = 0.015 2

SMD/Dual

PR

0.25 0.00

χ2 = 77.71 p = 0.000 SNTV

PR Born outside district

HR Winners 1.00

1.00

0.75

0.75

0.50

0.50

0.25 0.00

Born inside district

HC Winners

χ = 2.5 p = 0.281 2

SMD

Zombie PR

0.25 0.00

χ2 = 16.71 p = 0.000 SNTV

PR

Figure A.5: Personal attributes across candidates, winners, and institutional contexts (5).

A6

Gender

−0.14

−0.06

Gender

Any College

−0.1

0.05

−0.03

U of Tokyo

0.08

0.1

−0.09

Local

0.11

0.1

−0.13

0.04

0.03

Local

0.03

0.19

0

0.08

−0.02

0.06

Dynastic

0.11

0.13

−0.14

0.08

0.15

0.16

0.29

DPJ

0.1

0.42

−0.06

−0.01

−0.01

−0.03

0.05

JCP

−0.1

−0.2

0.04

−0.03

−0.12

−0.13

−0.08

Loc. Pol. Exp.

0.17

0.14

−0.11

−0.12

−0.07

0.11

0.16

−0.02

0.27

−0.1

Local

0.1

0.07

−0.07

0

−0.07

Dynastic

0.08

0.34

−0.07

0.14

0.08

0.11

Dynastic

LDP

0.09

0.44

−0.08

0.18

0.1

0.12

0.3

LDP

DPJ

−0.01

0.23

−0.02

0.05

0.09

0

0.01

JCP

0.03

−0.56

0.14

−0.29

−0.19

−0.07

−0.25

Loc. Pol. Exp.

0.15

0.19

−0.06

0.03

−0.12

0.19

0.02

JCP

DPJ

LDP 0.14

0.13

−0.19

0.06

−0.09

0.14

Any College

−0.22

−0.09

0.01

U of Tokyo

0

−0.12

−0.04

Local

0.29

0.1

−0.24

−0.08

−0.03

Local

Dynastic

−0.13

−0.02

0.02

0.13

−0.16

−0.03

Dynastic

−0.04

0.06

−0.04

Local

0.13

−0.02

−0.1

−0.01

−0.12

Dynastic

0.17

0.08

0.03

0.09

−0.02

0.1

LDP

0.07

0.16

−0.04

0.14

−0.03

0.1

0.2

LDP

0.14

−0.13

−0.12

0.02

0.1

0.19

0.17

DPJ

−0.03

0.11

0.04

−0.04

0.07

−0.06

−0.06

DPJ

−0.05

0.18

0.06

0.05

−0.18

−0.07

−0.13

JCP

−0.06

−0.41

0.14

−0.14

−0.03

−0.02

−0.12

JCP

−0.1

0.05

0.13

0.09

−0.12

−0.05

0.07

Loc. Pol. Exp.

0.19

−0.16

−0.12

−0.19

−0.15

0.16

0.1

−0.22

0.23

−0.11

JCP

DPJ

LDP

U of Tokyo −0.08

0.04

0.04

−0.07

JCP

U of Tokyo

DPJ

−0.03

LDP

0.11

U of Tokyo

0

Local

−0.08

Any College

Any College

−0.06

Gender

Gender

−0.13

−0.02

0.43

Exp.

−0.13

Gender

Age Exp.

Gender

0.2

Exp.

Dynastic

0.18

Any College

Exp.

Loc. Pol. Exp.

0.24

HC Winners

Age

HR Winners

JCP

U of Tokyo

DPJ

−0.04

LDP

0.29

U of Tokyo

−0.15

U of Tokyo

Any College

Any College

Exp.

−0.19

0.32

Dynastic

−0.03

Exp.

Local

Gender

Any College

0.11

Gender

Exp.

Exp.

Age

HC Candidates

Age

HR Candidates

0.23

−0.18

−0.08

Figure A.6: Correlation matrix of common attributes of candidates (top) and MPs (bottom) in the Japanese House of Representatives (left) and House of Councillors (right). Note: Each panel presents the numerical values of the correlations along with color coding, with darker blue (red) indicating larger positive (negative) correlations.

A7

B

Priming Information and Attribute Levels

Respondents were randomly divided into four groups. Those in the first group were first asked to read the following sentences that highlight the electoral system, the geographical unit/size of the district, and how voters cast a ballot: The House of Representatives uses two electoral tiers: single-member district and proportional representation. For the single-member district tier, the entire country is divided into 295 districts, and voters write the name of a candidate. The candidate with the most votes wins. On the next five screens, you will see tables featuring hypothetical politicians (or aspiring politicians) in the single-member district tier. Please examine each table carefully before answering the questions that follow. After completing the first set of five conjoint tasks, these respondents were asked to read the following sentences: Thank you very much. For the second tier of House of Representatives elections, proportional representation, the entire country is divided into 11 districts, and voters write the name of a party. Seats are allocated to parties based on the party vote. On the next five screens, you will see tables featuring hypothetical politicians (or aspiring politicians) in the proportional representation tier. Please examine each table carefully before answering the questions that follow. Respondents in the second group were given the same sets of information about the electoral systems for the House of Representatives, but in the reverse order. The respondents in the third group were first asked to read the following sentences: The House of Councillors uses two electoral tiers: district and proportional representation. For the district tier, each prefecture is a district, and voters write A8

the name of a candidate. Candidates are elected in the order of who gets the most votes, up to the number of seats up for grabs. On the next five screens, you will see tables featuring hypothetical politicians (or aspiring politicians) in the district tier. Please examine each table carefully before answering the questions that follow. After completing the first set of five conjoint tasks, these respondents were asked to read the following sentences: Thank you very much. For the second tier of House of Councillors elections, proportional representation, the entire country is the district, and voters write the name of a candidate or the name of a party. Seats are allocated to parties and then to candidate(s) in the order of votes, up to the number of seats that the party wins. On the next five screens, you will see tables featuring hypothetical politicians (or aspiring politicians) in the proportional representation tier. Please examine each table carefully before answering the questions that follow. Again, similar to the second group in the HR scenario, the respondents in the fourth group were given the same sets of information but in reverse order. After reading these sentences, study participants were asked to complete a total of ten conjoint tasks. Figure B.1 shows an example of the conjoint task, and Table B.1 shows a complete list of attributes and levels translated into English.

A9

Figure B.1: Example of conjoint table shown to respondents (in Japanese). Note: Column headers are: “Person 1” and “Person 2.” The text above the table asks: “Which of the following two persons do you think is the more desirable as a single-member district member of the House of Representatives? Even if you are not entirely sure, please indicate which of the two you would prefer.” The first column lists attributes, such as party and former occupation. Full English translations of attributes and levels are in Appendix Table B.1. The text below the table asks: “Which person do you prefer?” and is followed by two buttons for the respondent to use to register his or her choice.

A10

Table B.1: Hypothetical politicians’ attributes in the conjoint experiment Attributes

Levels

Party

Independent DPJ JCP K¯omeit¯o LDP Inside prefecture Outside prefecture High school Local public university Prestigious private university University of Tokyo Business employee Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor None Prefectural assembly member National assembly member Cabinet minister None 1 term 2 terms 3+ terms None Formerly in office Currently in office 30, 42, 57, 64, 79 Male Female

Hometown (Birthplace) Highest Educational Attainment

Prior Occupation

Parental Political Background

Previous terms served

Experience (Incumbency)

Age Gender

Note: Baseline levels are italicized. The University of Tokyo is the national public university that is widely considered to be the most prestigious in Japan. Local public university refers to all other public universities. Levels for hometown are represented as “[name of prefecture of respondent]” or “outside of [name of prefecture of respondent].” Celebrity is represented by the Japanese word “tarento” (a talent from TV, movies, music, comedy, etc.). Local political offices are represented as “[name of prefecture of respondent] assembly member” and “[name of prefecture of respondent] governor.” A11

C

Additional Analyses of ACIEs

Figure 3 in the main text presents estimated AMCEs for the attributes other than gender conditional on a hypothetical politician’s gender, and the estimated ACIEs with respect to gender and each of the remaining attributes. In Figure C.1, C.2, C.3, we show the corresponding conditional AMCEs and ACIEs for three additional attributes: age, dynastic family ties, and celebrity status. In the main unconditional AMCE results, each of these attributes showed a striking divergence from patterns in the observed attributes of politicians in Japan, so are worth exploring in a bit more detail. The differences across two age groups (42 and younger versus 57 and older) reveal few meaningful differences. It appears that young politicians are more positively evaluated if they are graduates of the prestigious University of Tokyo, but are more negatively evaluated if they are also business executives or celebrities. The latter effect is also reflected in the ACIE for celebrity status. Voters may be more willing to tolerate very elderly politicians if they are celebrities — although it is important to note that both attributes are highly unpopular. When it comes to parental political background, there are similarly few meaningful differences. Non-dynastic politicians get a slightly larger boost from a University of Tokyo education, and are slightly less popular than dynastic candidates within Komeito. This latter finding is puzzling since dynastic politicians in Komeito are rare. In Figure C.4, we explore the ACIE of party label. Because the LDP is the largest party, we focus on the difference between LDP politicians and all other parties’ politicians (including independents). There appear to be no substantively meaningful differences in preferences for attributes conditional on party label. This contrasts with the party-level differences reflected in the correlation matrix in Figure A.6. Although actual LDP politicians tend to be more experienced, elderly, male, dynastic, educated, and local, none of these attributes has a different effect for hypothetical LDP politicians compared to hypothetical non-LDP politicians—the significant ACIE for “Formerly in office, 2 (terms)” notwithstanding.

A12

Young Politician (30 or 42) Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

Old Politician (57 or older)

●

●

●

● ●

●

●

●

●

●

●

●

●

● ● ●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

● ●

●

●

● ●

0.0

●

●

●

−0.2

Comparison

●

●

0.2

−0.2

Change in E[Y]

0.0

Change in E[Y]

●

0.2

−0.2

0.0

0.2

Difference in Changes

Figure C.1: Average effects of hypothetical politicians’ attributes on respondents’ preference, given the hypothetical politician’s age. Note: The solid circles in the left and middle panels represent the estimated AMCEs of a hypothetical politicians’ attributes other than age, conditional on the politician’s binary age group (young on the left; old in the middle). The rightmost panel shows the average component interaction effects (ACIE) with respect to age and each of the other attributes. The horizontal bars represent 95% confidence intervals robust to clustering at the respondent level.

A13

Celebrity Politician Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP

Non−Celebrity Politician

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

● ●

● ●

●

●

●

●

●

●

●

●

● ●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

−0.2

Comparison

0.0

● ● ● ●

0.2

−0.2

Change in E[Y]

●

0.0

Change in E[Y]

0.2

−0.2

0.0

0.2

Difference in Changes

Figure C.2: Average effects of hypothetical politicians’ attributes on respondents’ preference, given the hypothetical politician’s celebrity status. Note: The solid circles in the left and middle panels represent the estimated AMCEs of a hypothetical politicians’ attributes other than celebrity status, conditional on whether the politician is a celebrity (left) or not (middle). The rightmost panel shows the average component interaction effects (ACIE) with respect to celebrity and each of the other attributes. The horizontal bars represent 95% confidence intervals robust to clustering at the respondent level.

A14

Dynastic Politician Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

Non−Dynastic Politician

●

Comparison

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

● ● ●

● ●

●

● ●

●

−0.2

●

0.0

● ●

0.2

−0.2

Change in E[Y]

0.0

●

0.2

Change in E[Y]

−0.2

0.0

0.2

Difference in Changes

Figure C.3: Average effects of hypothetical politicians’ attributes on respondents’ preference, given the hypothetical politician’s parental political background. Note: The solid circles in the left and middle panels represent the estimated AMCEs of a hypothetical politicians’ attributes other than dynastic status, conditional on whether the politician is dynastic at any level of politics (left) or not (middle). The rightmost panel shows the average component interaction effects (ACIE) with respect to dynastic status and each of the other attributes. The horizontal bars represent 95% confidence intervals robust to clustering at the respondent level.

A15

LDP Politician Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

Non−LDP Politician

●

Comparison

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

● ●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

● ●

●

0.0

●

●

●

−0.2

● ●

●

●

0.2

−0.2

Change in E[Y]

0.0

●

0.2

Change in E[Y]

−0.2

0.0

0.2

Difference in Changes

Figure C.4: Average effects of hypothetical politicians’ attributes on LDP supporters’ preference, given the hypothetical politician’s party. Note: The solid circles in the left and middle panels represent the estimated AMCEs of a hypothetical politicians’ attributes other than party, conditional on whether the politician is LDP (left) or not (middle). The rightmost panel shows the average component interaction effects (ACIE) with respect to the LDP indicator and each of the other attributes. The horizontal bars represent 95% confidence intervals robust to clustering at the respondent level. Estimates are for the sample of LDP supporters among the respondents.

A16

D

Robustness Checks

Figure D.1 shows the estimated AMCEs from our conjoint experiment for each of the two primed dimensions: HC vs. HR (top) and plurality districts vs. PR (bottom), along with the differences in the AMCEs (rightmost plots). The results confirm our conclusion based on the main analysis reported in Figure 2: there is no discernible difference in the AMCEs between any of the experimental conditions. Figures D.2 and D.3 show the results of our tests for whether the overall null findings are due to either a failure of manipulation or lack of engagement among the respondents. For the manipulation check, we included two knowledge questions about each of the treatment conditions the respondents were shown (specifically, we asked questions about the factual details of the electoral rules), and labeled those who answered both of them correctly as “knowledgeable” for the corresponding condition. Figure D.2 shows the result of the same analysis as in Figure D.1 only on the knowledgeable respondents (N = 604 for the upper house; 706 for the lower house; 622 for the district tier; and 432 for the PR tier). As in the main analysis, none of the differences in the AMCEs are statistically significantly different from zero. For the test of respondents’ engagement, we included a pair of screener questions at the end of our survey, and marked those who passed at least one question as “attentive” respondents. We then repeated our analysis on the attentive respondents (N = 766, of whom 372 were in the upper house condition). The results, reported in Figure D.3, show that only one out of the 28 comparisons shows a statistically significant difference between the two priming conditions on each dimension, replicating the overall null finding even on the attentive subsample. Finally, Figure D.4 presents the results of the same set of analyses on the subset of respondents who stated that they had either always or almost always voted in past elections (N = 687 for the upper house; 697 for the lower house; 692 for each of the tiers). The

A17

differences are statistically significant for only two of the 56 comparisons, again replicating the main finding on this subgroup of the respondents.

A18

Upper House Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

●

Comparison

●

●

● ● ● ●

●

●

●

●

●

● ● ● ● ● ●

●

● ● ● ● ● ●

● ● ● ● ●

●

●

● ● ●

●

● ● ●

●

● ● ●

●

●

● ● ●

● ● ●

●

● ● ●

●

● ●

● ● ● ●

● ● ●

●

●

●

●

● ● ● ● ● ●

● ● ●

● ● ● ●

● ●

−0.2

Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

Lower House

0.0

0.2

−0.2

0.0

0.2

−0.2

0.0

Change in E[Y]

Change in E[Y]

Difference in Changes

District Tier

PR Tier

Comparison

●

●

●

● ● ● ●

●

●

●

●

●

● ● ● ● ● ●

● ● ● ● ● ●

●

● ● ● ● ● ●

●

● ● ●

●

● ● ●

●

● ● ●

●

● ● ●

●

● ●

● ● ●

●

●

●

● ●

● ● ● ●

● ● ●

●

●

●

●

● ● ● ●

● ●

−0.2

● ● ● ● ● ●

● ● ●

0.0

Change in E[Y]

0.2

−0.2

0.2

0.0

Change in E[Y]

0.2

−0.2

0.0

0.2

Difference in Changes

Figure D.1: Effects of politicians’ attributes on respondents’ preference, by priming conditions (alternate comparisons). Note: Top panels give the results for the HC vs. HR, pooling across the tier types. Bottom panels give the results for plurality vs. PR, pooling across the houses. The rightmost panel for each shows the difference in the estimated AMCEs. A19

Upper House (Knowledgeable Only) Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

●

Comparison

●

●

● ●

● ●

●

●

●

●

●

● ●

● ●

● ●

● ● ● ●

●

●

● ● ●

●

● ● ●

●

● ●

●

●

● ● ●

● ●

●

●

●

●

●

● ● ●

● ●

●

● ● ●

●

● ●

● ● ●

● ● ●

●

●

●

●

●

● ● ● ● ●

● ● ●

● ● ● ●

● ●

−0.2

Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

Lower House (Knowledgeable Only)

0.0

0.2

−0.2

●

0.0

0.2

−0.2

0.0

0.2

Change in E[Y]

Change in E[Y]

Difference in Changes

District Tier (Knowledgeable Only)

PR Tier (Knowledgeable Only)

Comparison

●

● ● ●

● ● ●

● ● ● ●

●

●

● ●

●

● ●

● ●

●

● ●

●

● ● ●

●

● ●

●

● ● ●

●

●

● ● ●

●

● ● ●

●

●

● ●

●

●

●

●

●

● ●

●

● ●

● ●

●

●

●

●

●

●

●

● ●

● ●

● ● ● ●

−0.2

●

● ● ●

0.0

●

● ● ●

0.2

−0.2

Change in E[Y]

0.0

Change in E[Y]

● ●

0.2

−0.2

0.0

0.2

Difference in Changes

Figure D.2: Effects of politicians’ attributes on respondents’ preference, by priming conditions (high-knowledge respondents). Note: Estimates of the quantities equivalent to Figure D.1 on the subsample who passed the manipulation check (knowledge tests on the treatment text). A20

Upper House (Attentive Only) Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

●

Comparison

● ●

● ●

● ● ● ●

●

●

●

●

●

●

● ●

●

● ●

●

● ● ●

●

● ●

● ●

●

●

● ● ●

●

●

●

● ● ●

●

● ●

●

●

● ● ●

●

●

●

● ●

●

●

●

● ●

●

● ●

● ●

●

●

●

●

●

●

●

● ● ● ●

● ● ● ● ●

● ● ● ●

●

−0.2

Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

Lower House (Attentive Only)

0.0

0.2

−0.2

0.0

0.2

−0.2

0.0

0.2

Change in E[Y]

Change in E[Y]

Difference in Changes

District Tier (Attentive Only)

PR Tier (Attentive Only)

Comparison

●

●

●

●

● ●

● ●

● ●

●

●

●

● ●

● ● ● ● ●

● ● ●

●

● ● ● ●

●

●

●

● ● ●

● ●

●

● ● ●

●

●

● ● ●

● ●

● ●

●

●

●

●

● ●

● ● ● ●

● ● ●

●

●

●

●

●

● ●

0.0

● ● ●

● ● ●

−0.2

●

● ●

● ● ● ●

0.2

−0.2

Change in E[Y]

0.0

Change in E[Y]

● ●

0.2

−0.2

0.0

0.2

Difference in Changes

Figure D.3: Effects of politicians’ attributes on respondents’ preference, by priming conditions (high-attention respondents). Note: Estimates of the quantities equivalent to Figure D.1 on the subsample who passed the screener questions placed at the end of the survey. A21

Upper House (Voted in the Past) Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

●

Comparison

●

●

●

●

●

●

●

●

● ●

●

● ● ● ● ● ●

● ●

● ●

● ● ● ●

●

● ● ● ●

●

● ●

●

● ● ●

●

●

● ● ●

●

● ● ●

●

● ● ●

●

● ● ●

●

● ●

●

● ● ●

● ● ●

●

●

●

●

● ● ● ● ●

● ● ●

● ● ● ●

● ●

−0.2

Age: (Baseline = 30) 42 57 64 79 Experience: (Baseline = No experience) Formerly in office, 1 Currently in office, 1 Formerly in office, 2 Currently in office, 2 Formerly in office, 3+ Currently in office, 3+ Gender: (Baseline = Male) Female Highest Educational Attainment: (Baseline = High school) Local public university Prestigious private university University of Tokyo Hometown: (Baseline = Inside) Outside Parental Political Background: (Baseline = None) Prefectural assembly member National assembly member Cabinet minister Party: (Baseline = Independent) DPJ JCP Komeito LDP Prior Occupation: (Baseline = Business employee) Business executive Celebrity Local government employee National government employee Prefectural assembly member Prefectural governor

Lower House (Voted in the Past)

0.0

0.2

−0.2

●

0.0

0.2

−0.2

0.0

0.2

Change in E[Y]

Change in E[Y]

Difference in Changes

District Tier (Voted in the Past)

PR Tier (Voted in the Past)

Comparison

●

● ●

● ●

● ● ●

●

●

●

●

● ●

● ● ● ● ● ●

● ● ● ●

●

●

● ●

●

● ● ●

● ● ●

●

●

●

● ● ●

●

● ●

● ●

●

●

●

●

● ●

● ● ● ●

● ● ●

●

●

●

●

● ● ● ● ● ●

● ● ● ● ●

−0.2

● ● ● ● ● ●

0.0

● ● ● ●

0.2

−0.2

Change in E[Y]

0.0

0.2

Change in E[Y]

−0.2

0.0

0.2

Difference in Changes

Figure D.4: Effects of politicians’ attributes on respondents’ preference, by priming conditions (frequent-voter respondents). Note: Estimates of the quantities equivalent to Figure D.1 on the subsample who reported that they had either always or almost always voted in past elections. A22