The Impact of Descriptive Representation on External Efficacy in Australia

Transcription

1 The Impact of Descriptive Representation on External Efficacy in Australia By Feodor Snagovsky 1, Woo Chang Kang 1, Jill Sheppard 1 and Nicholas Biddle 2 1 School of Politics and International Relations 2 Centre for Social Research and Methods Australian National University Canberra, ACT Australia Presented at the 2018 Meeting of the International Political Science Association July 23, 2018 Abstract How does the descriptive representation of ethnic minorities affect how voters feel about the responsiveness of government? While there are many reasons why we should care about DR from the point of view of equality and justice, there is limited evidence outside the United States that DR actually changes the way citizens feel about the state. We use the case of Australia, a diverse, settler society with high levels of immigration and robust democratic institutions that has yet to elect ethnic minorities to public office in large number. Further, the presence of compulsory voting allows us to rule out voter-mobilization based explanations for this relationship. We use data from a conjoint experiment as well as the Australian Election Study to evaluate whether the presence of ethnic minority candidates changes perceptions of government responsiveness among ethnic minority and ethnic majority voters. We find ethnic minority Australians do not appear to have higher levels of external efficacy when given an opportunity to vote for a fellow minority. By contrast, Anglo respondents tend to feel like minority candidates will be less likely to care about what people like them think and have lower levels of external efficacy when given an opportunity to vote for a non-anglo candidate. Working paper: please do not cite without permission

2 1. Introduction Does the descriptive representation (DR) of ethnic minorities change the way they feel about the responsiveness of government? While there are many reasons why we should care about DR from the point of view of equality and justice, there is limited evidence outside the United States that DR actually changes the way citizens feel about the state. Moreover, the US-based literature we do have is characterized by considerable contradiction. Put simply, we do not yet know how, or even whether, DR changes citizens beliefs or behaviour. This study takes the first step in filling this gap in the literature by considering the case of Australia a diverse, settler society with high levels of immigration and robust democratic institutions that has yet to elect ethnic minorities to public office in large number. To our knowledge, this is the first analysis of how descriptive representation affects political attitudes in Australia. Australia is useful as a case study for another reason; it is one of the few advanced, Westminsterstyle democracies in the world with compulsory voting. As such, the relationship between DR and external efficacy in Australia is unlikely to be mediated through differential voter turnout or mobilization, since almost everyone votes. This allows us to investigate other parts of the causal mechanism in greater detail. As will be described below, however, the Australian case presents a challenge for another reason: very few ethnic minorities have been elected to federal office, making it difficult to evaluate how electing minority representatives affects minority constituents. Instead, this analysis will evaluate how political candidates can shape the political attitudes of those whom they seek to represent. This paper combines a conjoint experiment on a probability-based sample of Australian voters with an observational analysis of the Australian Election Studies to evaluate the relationship between candidate ethnicity and voter external efficacy. We find descriptive representation has no effect on the external efficacy of ethnic minority respondents, but decreases the external efficacy of Anglo Australian respondents. This paper will first discuss the history of multiculturalism and existing work on minority representation in Australia. Next, it will discuss the research designs for the two components (observational and experimental) of this analysis. Finally, it will explain the findings of these analyses, and then discuss their implications. 2. Descriptive Representation and External Efficacy Descriptive representation (DR) refers to a shared identity between a representative and their constituents, most commonly thought of in terms of gender or ethnicity (Mansbridge 1999; Casellas and Wallace 2015). Several arguments have been advanced in favour of DR in the decades since Pitkin (1967) first described it, including on the grounds of pure justice, as a symbol to historically excluded groups, giving minorities legislators more control over political agendas, and making sure minority interests are represented (Childs and Cowley 2011). 1

3 Research on how descriptive representation affects political attitudes is limited. Measuring changes in external efficacy, in particular, offers an opportunity to evaluate some of the most logically foreseeable impacts of descriptive representation on political attitudes. External efficacy is a measure of government responsiveness, representing the belief that the authorities or regime are responsive to influence attempts (Balch 1974, 24). External efficacy is an important indicator of democratic health (Craig, Niemi, and Silver 1990, 289); if descriptive representation strengthens the bond between minority citizens and their government, these citizens should be more likely to think government is responsive to their demands. As such, measuring external efficacy offers a very specific and testable measure of government legitimacy that may provide empirical support to arguments about the value of descriptive representation in elected bodies of government (Atkeson and Carrillo 2007, 80). Past work on how the descriptive representation of ethnic minorities affects their external efficacy mostly focuses on the case of the United States, where the polarization of race in public discourse may limit the generalizability of those findings to other, less polarized, democracies. Even there, evidence is mixed; some studies suggest the representation of ethnic minorities in the US increases their external efficacy (Pantoja and Segura 2003; West 2016; Merolla, Sellers, and Fowler 2013), while others suggest it does not (Banducci, Donovan, and Karp 2004). Scherer and Curry (2010) find descriptive representation may have a trade-off increasing the diffuse support of ethnic minorities, but decreasing the diffuse support of ethnic majorities. 2. Descriptive Representation in Australia While candidates in the 2016 election described Australia as the most successful multicultural country in the world (Jupp and Pietsch 2018, 663), Australia lags far behind other advanced democracies in descriptive representation. 28% of Australians were born overseas, while 49% of Australians were either themselves born overseas or had a parent who was (Australian Bureau of Statistics 2017b). By contrast, only 11 out of 150 Members of the current (45 th ) House of Representatives (7%) were born outside Australia, and only three of those Members (2%) were born outside of Europe (Parliament of Australia 2016). Pietsch (2017) finds only two individuals with Euro-Asian heritage and two with Middle Eastern heritage have been elected to the House of Representatives from In addition, there are only two Indigenous Australians (1%) in the House, despite the fact 3% of Australians identify as Aboriginal or Torres Strait Islanders (Parliament of Australia 2016; Australian Bureau of Statistics 2017a). By way of comparison, 14% of Canadian MPs are visible minorities (compared to 22% of the population) and 3% are Indigenous (compared to 5% of the population) (Tolley 2015; Statistics Canada 2017b, 2017a). Despite how few minorities have been elected to federal office, there is evidence ethnic minority candidates for the Labor party receive an electoral bonus from ethnic minority voters (Zingher and Farrer 2016). However, while high levels of district-level ethnic diversity makes centre-left parties more likely to nominate ethnic minority candidates, few electorates in Australia have very high levels of ethnic minority concentration (Farrer and Zingher 2018; Pietsch 2017). Further, living in areas with high levels of ethnic diversity seems to lead to a backlash from Anglo Australians, resulting in more negative attitudes towards immigration (Bilodeau and Fadol 2011). Evidence to this effect indicates the outsider status of Muslims, as well as Australians of Middle-Eastern and, less so, Asian origin (K. M. Dunn et al. 2004, 414). Negative attitudes 2

4 towards Muslim Australians, in particular, have many parallels with the old-fashioned racism of the White Australia era, a fact which does not appear to have changed over two decades of research (McAllister and Moore 1991; K. Dunn, Klocker, and Salabay 2007). 3. Data and Methods This study uses a two-pronged approach to examine how descriptive representation affects external efficacy. The first component uses observational data to evaluate how these variables interact in real-life Australian elections. The second component uses experimental data to dig deeper into the causal mechanism behind this relationship. Together, both approaches shed light on the underlying link between descriptive representation and external efficacy as it concerns Australian politics. 3.1 Observational Analysis The observational component of this analysis evaluates the extent to which political candidates from have changed the political attitudes of Australians in their electorates. Accessing data on culture or ethnicity is very difficult in Australia. The most complete dataset on candidate ethnicity in Australia comes from Zingher and Farrer (2016), who handcoded each candidate in the 2004, 2007 and 2010 Australian elections according to whether or not they are of non-australian descent. They use the electorate as the unit of analysis, indicating whether Coalition (Liberal or Nationals) and Labor/Greens candidates (which are combined for sample-size reasons) were of Australian or immigrant descent. As a result, their dataset has 450 observations one observation for each of the 150 federal electorates in Australia for each of the three election years mentioned above. As far as measuring ethnicity is concerned, voter-level data is somewhat more limited. For the most part, Australian surveys do not ask about culture or ancestry, instead relying on birthplace to measure ethnic diversity. In addition to this measure, the Australian Election Study (AES) asks respondents to indicate their parents birthplaces, allowing this analysis to consider not only first-generation, but also second-generation immigrants who responded to each survey. Further, the AES intentionally samples all federal electorates, ensuring that at least some respondents are drawn from each electorate where a minority candidate may have competed. This makes the AES particularly useful for this analysis. The final dataset uses the individual voter as the unit of analysis. Table 1 shows each variable considered. The dependent variable is external efficacy, which is measured in each AES using two survey questions, known in the external efficacy literature as carethink and knowthink. The first, carethink, corresponds to the survey question, Some people say that political parties in Australia care what ordinary people think. Others say that political parties in Australia don't care what ordinary people think. Where would you place your view on this scale from 1 to 5? Similarly, knowthink corresponds to the survey question, Some people say that Federal politicians know what ordinary people think. Others say that Federal politicians don't know much about what ordinary people think. Where would you place your view on this scale from 1 to 5? The results of each question were combined into an index to create a single variable, exteff, with a Cronbach s Alpha value of 0.67, indicating an acceptable scale consistency. 3

5 [Table 1 about here] The independent variables measure respondent and candidate ethnicity. For the former, ethvoter is coded as 1 if the respondent or either of their parents was born in a non-english speaking or non-european country, and 0 otherwise. labeth and coaleth are dummy variables from the Zingher and Farrer (2016) data that indicate whether the Labor/Greens and Coalition candidate, respectively, in the respondent s electorate comes from a non-australian background. Finally, ethcan is a combination of these two variables and indicates whether either party fielded a minority candidate in the electorate during the election. Following Zingher and Farrer (2016), this analysis is ultimately interested in descriptive representation; thus, the primary measure of interest is the interaction between ethvoter and each of the measures of candidate ethnicity discussed above. According to most theories of descriptive representation, minority voters should feel like the political system is more responsive when they see themselves represented in the political contest in their electorate. This leads to the expectation that: H1 (co-minority efficacy): Ethnic minority respondents will have higher levels of external efficacy when a minority candidate runs in their electorate. While minority respondents may update their political attitudes in the context of descriptive representation, it is unlikely white (Anglo) Australians will experience the same effect. Almost all candidates for the House of Representatives have an Anglo background; co-ethnic representation for Anglo Australians is the rule, not the exception. However, it is possible Anglo Australians may push back when this rule is violated and exhibit lower levels of external efficacy when a minority candidate runs for either major political party in their electorate. We may therefore expect that: H2 (majority disillusionment): Non-minority respondents will have lower levels of external efficacy when a minority candidate runs in their electorate. The analysis also uses a number of control variables to account for alternative explanations of external efficacy. External efficacy may have demographic determinants; therefore, the models control for age (both as a raw measure of years and as age squared, to account for the possibility of a non-linear relationship), female, which indicates the gender of the respondent, income, which measures the household income of the respondent, and highqual, which measures education through the highest level of educational attainment of each respondent. We might expect that older, wealthier, more educated and male respondents should be more likely to think government is responsive to their demands, all else equal. Finally, foreignborn measures the percent of the electorate that was born outside of Australia, in order to evaluate any possible ecological effects a respondent may derive from their community. The analysis also controls for political determinants of external efficacy; votewinner is a binary variable that is coded as 1 if the respondent voted for the winning candidate, and 0 otherwise. In theory, a respondent should think the political system is more responsive if their preferred 4

6 candidate wins. Further, polint and pidstrength measure each respondent s level of political interest and partisanship, respectively, since those who are interested in politics and have strong partisan preferences should be more likely to think the political system is responsive. Since the dependent variable is continuous, ordinary least squares (OLS) regression is most appropriate for this analysis. Further, data are pooled from three separate elections; thus, clustering by election year is necessary to account for intra-election homogeneity. 3.2 Experimental Analysis While analyses using observational data have high external validity, they must often sacrifice internal validity. Not only must the above methodological description disregard differences between different types of minority groups, it is also unable to pinpoint which part of a candidate s profile drives potential changes in political attitudes. For example, if a minority respondent is reacting to a minority candidate in their electorate, do they feel differently about them compared to an Anglo candidate because of the potential for shared ancestry, because of shared cultural understanding, or rather due to the ability to speak in a different language? By contrast, do Anglo, Australian-born respondents feel differently about minority candidates because they are prejudiced against those born outside of Australia or because they fear the candidate s cultural values do not align with their own? An experimental approach has obvious advantages in answering these types of questions. Especially in Australia, where low numbers of minority candidates have contested federal elections, the observational approach has limited statistical power to explore variation. A survey experiment is well suited for this task since researchers are able to control all aspects of a candidate s profile. However, traditional survey experiments that rely on vignettes are limited in what they can accomplish in this regard. For example, if a researcher wanted to test whether voters prefer white-collar or blue-collar candidates, presenting respondents with a choice between a banker or a tradesperson would not reveal whether a respondent prefers one candidate over another because type of job they have, their likely educational background, or perhaps the level of wealth they have accumulated. Systematically testing each of these mechanisms through a one dimensional design is both expensive and time consuming (Hainmueller, Hopkins, and Yamamoto 2014). Conjoint designs are a relatively recent tool in political science to address the above-described problems, though they have been applied widely in other disciplines (Hainmueller, Hopkins, and Yamamoto 2014). A relatively common conjoint design presents respondents with a series of hypothetical candidate profiles in pairs and asks respondents to choose between them (Franchino and Zucchini 2015; Carnes and Lupu 2016). These candidates attributes vary randomly, allowing the researcher to determine the relative influence of each trait. Moreover, they increase an experiment s external validity compared to traditional vignette experiments and reduce social desirability bias by creating many plausible reasons (from the point of view of a respondent) for why a candidate might be chosen over another (Hainmueller, Hangartner, and Yamamoto 2015). These attribute combinations are usually constrained to logical bounds; for example, a hypothetical lawyer with only a high-school education might confuse a respondent. Thus, it is 5

7 necessary to limit certain potential combinations that could be created from random assortment. Further, more specific candidate profiles may help prevent respondents from making unintended inferences about certain candidates. For example, Koch (2000) shows that voters make inferences about the ideology of women candidates based only on their gender. This makes it difficult to understand whether a voter chooses (or does not choose) a woman candidate because of her gender, or rather her perceived ideological position. This study drew upon a probability-based online panel that asked respondents to choose between five hypothetical sets of candidates. With a sample of 943 respondents, this design provides information on 9430 candidates (943 respondents, with five sets of candidates each, and two candidates per set). Our hypothetical candidates varied across twelve dimensions a list of these attributes is shown in Table 2. The list of dimensions represents personal attributes voters might care about when choosing their elected representative, such as their gender, age, family background, job and education. It also includes political characteristics, such as the candidate s political party. The candidate s ideology was specified in order to prevent respondents from making inferences about ideology based on the candidate s other personal attributes. Finally, we specified three dimensions related to descriptive representation country of birth, cultural background and other languages spoken. These dimensions were included to determine which aspects of a descriptive representative s identity drove potential changes in political attitudes. [Table 2 about here] The dependent variable of interest is chosen, which corresponds to a respondent choosing one of the two candidates according to the survey question, Which candidate do you think is the most likely to care what people like you think? This question is distinct from asking a respondent who they might vote for; indeed, there might be many reasons why someone would vote for one candidate but indicate the alternative cares more about what they think (such as the first candidate representing their preferred political party). Following Hainmueller et al., the relationship between the dependent variable and the independent variables is expressed as the Average Marginal Component Effect (AMCE), which equals the increase in the population probability that a profile would be chosen if the value of its lth component were changed from t0 to t1, averaged over all the possible values of the other components given the joint distribution of the profile attributes (2014, 11). Put otherwise, the ACME of a candidate attribute, such as country of birth, on the probability they will be chosen is the weighted average of the differences of all possible combinations of otherwise-identical candidate profiles that include different values of country of birth. In the context of candidate income, for example, the AMCE of income represents the average effect of income on the probability that the candidate will be chosen, where the average is defined over the distribution of the attributes (except for the candidate s own income) across repeated samples (Hainmueller, Hopkins, and Yamamoto 2014, 11). The other variables considered in this analysis are shown in Table 3. The variable rparty describes the respondent s political party preference (Coalition, Labor/Greens, or Other). The variables rculture and rcob are intended to capture different aspects of descriptive 6

8 representation, corresponding to the respondent s cultural background (Caucasian, Asian or Other), and country of birth (Australia, English-Speaking Country, Non-English-Speaking Country). The fate_yes variable is binary, and corresponds to the question, Agree or disagree: What happens to others in your ethnic group in this country will have some effect on your own life. This question is intended to measure whether respondents have a sense of linked-fate, a potential moderator in the relationship between descriptive representation and external efficacy. Finally, seifa_high is another binary variable which captures the respondent s socio-economic background according to the Socio-Economic Indexes for Areas (SEIFA) score for the area in which they reside. [Table 3 about here] The experimental design allows us to test several individual-level mechanisms. Given samplesize constraints, the observational study can only test hypotheses regarding minority status (i.e. whether the respondent is a visible minority or not). The experimental component is slightly more detailed, testing hypotheses at the level of the respondent s country group. Therefore, the analogous hypothesis to H1 (co-minority efficacy) is: H3 (co-ethnic efficacy): Asian respondents will be more likely to indicate Chinese and Indian candidates care about what people like them think, while those of Caucasian origins will have no preference. However, since politics in Australia is not especially polarized along racial lines, it could be that ethnic background matters less than immigrant background. Perhaps immigrants think other immigrants will better understand their experiences and needs: H4 (immigrant-based efficacy): Overseas-born respondents will be more likely to indicate overseas-born candidates care about what people like them think. While immigrants face unique challenges to integration in Australian public life, immigrants from non-english speaking backgrounds (ESB) face even tougher barriers. Since the ethnicities of our hypothetical candidates primarily imply non-esb origins (with the exception of Indian candidates, though Indians speak a plethora of languages), it could be that only NESB immigrants care about shared immigrant experience: H4a (NESB immigrants): Only overseas-born respondents from NESB countries and not those from ESB countries or those born in Australia, will be more likely to indicate overseas-born candidates care about what people like them think. The extant literature suggests a sense of linked fate can motivate ethnic-based preferences in certain contexts (Gay 2004; Simien 2005; McConnaughy et al. 2010). While not all members of an ethnic minority will have a sense of linked fate, whether one thinks their future prospects are tied to their ethnic group may be one reason why a constituent thinks a descriptive representative will be more responsive: 7

9 H5 (linked fate): Asian respondents with a sense of linked fate will be more likely to indicate Chinese and Indian candidates care about what people like them think. The following section summarizes the results of each study and is followed by a discussion of their implications in the context of descriptive representation in Australia. 4. Results 4.1 Observational Analysis The results of the regression for external efficacy and descriptive representation are shown in Table 4. Model 1 presents the overall results. Model 2 breaks the results down by candidate party. Models 3 and 4 further decompose the results according to respondent vote choice. [Table 4 about here] Age, both measured in years and as age-squared, has a statistically significant impact on external efficacy for all models save age squared for Coalition voters. However, the direction of the impact is different between age and age-squared, suggesting that the impact of age on external efficacy may be parabolic increasing to a point and then decreasing. Gender and income are not significant in any of the models. Political interest and strength of partisanship are mostly significant and in the expected direction. Education does not appear to be strongly linked to external efficacy when accounting for other factors, save for the strong, statistically significant bonus for those with post-graduate degrees. While all the above results are displayed for the sake of completeness, following Brambor et al. (2005) and Zingher and Farrer (2016), the results of the interactions between ethvoter and each of ethcan, labeth and coaleth are also presented and interpreted through marginal effects and marginal differences. The marginal effects present the means for each group, while the marginal differences present the differences between them, and evaluate whether those differences are statistically significant. The results of these analyses are shown in Tables [Tables 5 10 about here] Turning first to the overall differences between Anglo and minority voters (Tables 5 and 6), it appears the significant results for the interaction in Table 4, model 1 are driven by differences between minority and Anglo candidates among Anglo voters, leading to a decrease in external efficacy when a minority candidate is present. The presence of a minority candidate makes no difference for minority respondents. When these differences are broken down according to candidate party (Tables 7 and 8), we find some evidence that the backlash against minority candidates is limited to Anglo respondents assessing Coalition candidates. No other coefficients are statistically significant at the 0.05 level. However, when the results are also broken down by respondent vote choice, it appears that Anglo Labor/Greens voters also have lower levels of external efficacy when a minority 8

10 Labor/Greens candidate runs in their electorate (Tables 9 10). Critically, minority respondents do not have statistically significant differences in external efficacy when given a choice between Anglo and minority candidates in any of the models. 4.2 Experimental Analysis The results of the conjoint experiment for external efficacy are shown in Figures 1 8. Figure 1 shows the overall results; Figures 2 6 break the respondents down according to their partisan preferences, culture, country of birth, social advantage status, and sense of linked fate, respectively. Figure 7 breaks respondents down by both culture and linked fate, while Figure 8 shows results for country of birth and linked fate. [Figures 1 8 about here] Looking first at the effect of candidate gender, respondents seem to think female candidates are more likely to care about what people like them think in most models. The results hold across all cultures, for Australian and NESB-respondents, and all socio-economic statuses (Figures 1, 3 5). When broken down by party, however, it appears only Labor/Greens identifiers and Other identifiers, but not Coalition identifiers, think women are more responsive to their interests (Figure 2). Older candidates do worse than younger candidates in almost all models, while the effects for marital status, and incumbency are generally not significant. Party officials and candidates working in agriculture tend to do better than lawyers and professionals, but again the results are often not significant. Candidates with a postgraduate degree do better among socioeconomically advantages respondents and Labor/Greens respondents, but otherwise candidate education tends not to be important. Some of the strongest results in the model appear for candidate ideology. In almost every model, extremist candidates ( Hard Left or Hard Right ) are seen as less responsive to respondents concerns than moderate candidates. While Coalition-identifying respondents seem to think centre-left candidates and hard-right candidates are about as likely to care about their concerns (that is, not very likely), Labor/Greens respondents worry about hard left candidates in addition to all right-wing candidates. Turning to the question of descriptive representation, Caucasian respondents and those born in Australia or an English-speaking country consistently indicate Australian-born candidates are more likely to be responsive to their concerns. Asian and NESB respondents also indicate a bias towards Australian-born candidates, though the results are not significant at the 0.05 level. Candidate language generally does not have a statistically-significant impact on respondent evaluations of responsiveness, though there is some evidence voters think Arabic-speaking candidates will be less receptive to their concerns. Australian-born and Caucasian respondents, as well as those respondents who live in socially-disadvantaged areas, also think Middle Eastern candidates will do a worse job of representing their interests. There is very little evidence Asian or foreign-born respondents think non-caucasian candidates will be more receptive to their concerns in any of the models. 9

11 Finally, there is little evidence liked-fate considerations mediate respondents perceptions of candidate responsiveness. Paradoxically, the only exception appears to be Asian respondents without a sense of linked fate, who think Indian candidates will be more responsive to their concerns. The effect does not hold for any other group. 5. Discussion Overall, the results for the observational and experimental analyses paint a remarkably similar picture of the effects of descriptive representation. Of all the hypotheses, only H2 (majority disillusionment) is supported by both analyses; Caucasian and Australian-born respondents have lower levels of observed or predicted external efficacy when evaluating a foreign-born or non- Anglo candidate. In other words, Anglo respondents are less likely to think non-anglo candidates will be responsive to people like them. This finding is consistent with prior work in other jurisdictions, which shows the descriptive representation of ethnic minorities sometimes reduces the diffuse support of ethnic majorities (Scherer and Curry 2010). By contrast, there is little evidence the opposite is true; minority respondents do not seem to think non-anglo or foreign-born candidates will be more interested in their perspectives on political issues. This finding is at odds with some of the existing literature, which holds that the descriptive representation increases diffuse support (Banducci, Donovan, and Karp 2004; Pantoja and Segura 2003). There are several possible explanations for why this might be the case. First, most of the existing literature on the relationship between DR and external efficacy is based on the United States, a country that is much more polarized along racial lines than Australia. Race and racial relations are a major focus of American political discourse, which is not surprising given the United States long and turbulent history of minority oppression including slavery, laws that both explicitly and implicitly target ethnic minorities, and modern-day tension between minorities and state actors (such as the police). By contrast, while Australia also has a long history of discrimination and bias against immigrants, race-relations are not a major subject in Australian political discourse. Simply put, when ethnic tension is lower, it is possible the power of descriptive representation may be reduced. This explanation supports another of the important findings of this study that linked fate does not explain the relationship between DR and external efficacy, since we should expect linked fate to matter less when ethnic divisions are less pronounced. Second, most existing literature on DR and political attitudes measures how voters respond to ethnic minority legislators. By contrast, this study evaluated how voters respond to minority candidates. Australian legislators are overwhelmingly white; it is possible that as more ethnic minorities win elected office, members of these groups will see more benefits to being represented by one of their own. Third, most voters are not likely to know or care how many ethnic candidates are running in other electorates. Thus, a candidate-based study is necessarily dyadic in nature that is, focusing on the relationship between voters and candidates in their electorate. Many of the findings of a positive DR effect are related to collective (legislature-level) and elite descriptive representation. 10

12 As discussed above, this may change as more minority legislators win elected office. It is possible the political attitudes of ethnic minorities may change in response to a certain minimum threshold of minority legislators, beyond which there is a greater likelihood of minority-friendly policy change. 6. Conclusion Very little research has examined how descriptive representation affects the political attitudes of ethnic minorities outside of the United States. While this not only limits our understanding of other democratic contexts, it also limits our understanding of the mechanism behind this relationship; racial tension in the United States is high, and racial relations are a very salient issue in political discourse. As such, it is difficult to tell whether the two variables are related in contexts where ethnicity is less salient. Since very few non-white Australians have been elected to the House of Representatives, this analysis turned to the next best option for descriptive representation in Australian federal politics how political candidates with an ethnic minority background influence those whom they seek to represent. While later future work should consider how electing ethnic minorities changes political attitudes, Australia provides a unique case of a developed, Western democracy that has very low levels of minority representation. As such, Australia is an instructive example of what a baseline might look like a very diverse society where parties field minority candidates, but in which they have yet to win federal office in large number. Alternatively, other unobserved elements of Australian political culture may have led to this result, which in turn mediates the relationship between descriptive representation and diffuse support. As this analysis demonstrated, minority Australians do not appear to have higher levels of external efficacy when given an opportunity to vote for a fellow minority. This could be due to the unit of analysis perhaps it actually takes a minority being elected to office for a voter to update their political attitudes. Alternatively, minority representation might be best understood as a collective phenomenon perhaps it is not just a question of being elected to office, but rather being elected in large number. By contrast, Anglo respondents tend to feel like minority candidates will be less likely to care about what people like them think and have lower levels of external efficacy when given an opportunity to vote for a non-anglo candidate. This finding supports prior work that shows Australian voters are biased against non-anglo candidates. 11

13 7. References Atkeson, Lonna Rae, and Nancy Carrillo More Is Better: The Influence of Collective Female Descriptive Representation on External Efficacy. Politics & Gender null (01): Australian Bureau of Statistics. 2017a. Aboriginal and Torres Strait Islander Population Census of Population and Housing: Reflecting Australia - Stories from the Census, Canberra, Australia: Australian Bureau of Statistics b. Cultural Diversity in Australia, Census of Population and Housing: Reflecting Australia - Stories from the Census, Canberra, Australia: Australian Bureau of Statistics. 0Features~Cultural%20Diversity%20Article~60. Balch, George I Multiple Indicators in Survey Research: The Concept Sense of Political Efficacy. Political Methodology 1 (2): Banducci, Susan A., Todd Donovan, and Jeffrey A. Karp Minority Representation, Empowerment, and Participation. Journal of Politics 66 (2): Bilodeau, Antoine, and Nada Fadol The Roots of Contemporary Attitudes toward Immigration in Australia: Contextual and Individual-Level InfluencesÞ. Ethnic and Racial Studies 34 (6): Brambor, Thomas, William Roberts Clark, and Matt Golder Understanding Interaction Models: Improving Empirical Analyses. Political Analysis 14 (1): Carnes, Nicholas, and Noam Lupu Do Voters Dislike Working-Class Candidates? Voter Biases and the Descriptive Underrepresentation of the Working Class. American Political Science Review 110 (4): Casellas, Jason P., and Sophia J. Wallace The Role of Race, Ethnicity, and Party on Attitudes Toward Descriptive Representation. American Politics Research 43 (1): Childs, Sarah, and Philip Cowley The Politics of Local Presence: Is There a Case for Descriptive Representation? Political Studies 59 (1): Craig, Stephen C., Richard G. Niemi, and Glenn E. Silver Political Efficacy and Trust: A Report on the NES Pilot Study Items. Political Behavior 12 (3): Dunn, Kevin, Natascha Klocker, and Tanya Salabay Contemporary Racism and Islamaphobia in Australia: Racializing Religion. Ethnicities 7 (4): Dunn, Kevin M., James Forrest, Ian Burnley, and Amy McDonald Constructing Racism in Australia. Australian Journal of Social Issues 39 (4): Farrer, Benjamin David, and Joshua N. Zingher Explaining the Nomination of Ethnic Minority Candidates: How Party-Level Factors and District-Level Factors Interact. Journal of Elections, Public Opinion and Parties 0 (0): Franchino, Fabio, and Francesco Zucchini Voting in a Multi-Dimensional Space: A Conjoint Analysis Employing Valence and Ideology Attributes of Candidates. Political Science Research and Methods 3 (2):

14 Gay, Claudine Putting Race in Context: Identifying the Environmental Determinants of Black Racial Attitudes. American Political Science Review 98 (4): Hainmueller, Jens, Dominik Hangartner, and Teppei Yamamoto Validating Vignette and Conjoint Survey Experiments against Real-World Behavior. Proceedings of the National Academy of Sciences 112 (8): Hainmueller, Jens, Daniel J. Hopkins, and Teppei Yamamoto Causal Inference in Conjoint Analysis: Understanding Multidimensional Choices via Stated Preference Experiments. Political Analysis 22 (1): Jupp, James, and Juliet Pietsch Migrant and Ethnic Politics in the 2016 Election. In Double Disillusion : The 2016 Australian Federal Election, edited by Anika Gauja, Peter Chen, Jennifer Curtin, and Juliet Pietsch. Canberra, Australia: ANU Press. doi.org/ /dd Koch, Jeffrey W Do Citizens Apply Gender Stereotypes to Infer Candidates Ideological Orientations? The Journal of Politics 62 (2): Mansbridge, Jane Should Blacks Represent Blacks and Women Represent Women? A Contingent Yes. The Journal of Politics 61 (03): McAllister, Ian, and Rhonda Moore The Development of Ethnic Prejudice: An Analysis of Australian Immigrants. Ethnic and Racial Studies 14 (2): McConnaughy, Corrine M., Ismail K. White, David L. Leal, and Jason P. Casellas A Latino on the Ballot: Explaining Coethnic Voting among Latinos and the Response of White Americans. The Journal of Politics 72 (4): Merolla, Jennifer L., Abbylin H. Sellers, and Derek J. Fowler Descriptive Representation, Political Efficacy, and African Americans in the 2008 Presidential Election. Political Psychology 34 (6): Pantoja, Adrian D., and Gary M. Segura Does Ethnicity Matter? Descriptive Representation in Legislatures and Political Alienation among Latinos. Social Science Quarterly 84 (2): Parliament of Australia Part 2 - The Forty-Fifth Parliament : Statistical Information as at : Overseas Born. Canberra, Australia: Parliament of Australia. book%2fnewhandbook%2f %2F0028;query=Id%3A%22handbook%2Fnewhandbook%2F %2F0009%22. Pietsch, Juliet Explaining the Political Under-Representation of Asian Australians: Geographical Concentration and Voting Patterns. Political Science 69 (2): Pitkin, Hanna Fenichel The Concept of Representation. Univ of California Press. q=pitkin+the+concept+of+representation&ots=15vslp_mn4&sig=1osyg1chbltbnrgkutfieezfjs. Scherer, Nancy, and Brett Curry Does Descriptive Race Representation Enhance Institutional Legitimacy? The Case of the US Courts. The Journal of Politics 72 (01):

15 Simien, Evelyn M Race, Gender, and Linked Fate. Journal of Black Studies 35 (5): Statistics Canada. 2017a. The Daily Aboriginal Peoples in Canada: Key Results from the 2016 Census. Census Ottawa, Canada: Statistics Canada b. The Daily Immigration and Ethnocultural Diversity: Key Results from the 2016 Census. Census Ottawa, Canada: Statistics Canada. Tolley, Erin Visible Minority and Indigenous Members of Parliament. The Samara Centre for Democracy (blog). 26 November West, Emily A Descriptive Representation and Political Efficacy: Evidence from Obama and Clinton. The Journal of Politics 79 (1): Zingher, Joshua N., and Benjamin Farrer The Electoral Effects of the Descriptive Representation of Ethnic Minority Groups in Australia and the UK. Party Politics 22 (6):

16 Appendix 1: Tables and Figures Table 1: Summary Statistics and Variable Descriptions for Observational Analysis Variable Observations Mean Std. Dev. Min Max Description exteff External Efficacy age Age (years) age Age (squared) female Whether respondent is female income Gross Annual Income (by category) foreignborn Percent of electorate that is foreign-born votewinner Voted for winner polint Interest in politics pidstrength Strength of party identification highqual Highest qualification since leaving school ethvoter Ethnic minority voter ethcan Race has an ethnic minority candidate labeth Race has ethnic minority Labor candidate coaleth Race has ethnic minority Liberal/Nationals candidate 15

17 Table 2: Candidate Attributes for Experimental Vignette Attribute Permutation 1 Permutation 2 Permutation 3 Permutation 4 Permutation 5 Gender Male Female Age Aged years Aged years Aged years Party Labor Liberal Currently seated Yes, currently seated No, not currently seated Political ideology Hard <left/right> Centre <left/right> Moderate Marital status Single Married Divorced Widowed Children No children One child Two children Three children Four children Country of birth Born in Australia Not born in Australia Cultural background Anglo-Caucasian Middle Eastern Chinese Indian Other languages None Mandarin Hindi Arabic Previous occupation Lawyer Professional Agriculture Party official Highest education High school certificate Bachelor degree Postgraduate degree 16

18 Table 3: Summary Statistics and Variable Descriptions for Experimental Analysis Variable Observations Mean Std. Dev. Min Max Description chosen Binary variable indicating whether candidate was chosen rparty Respondent s political party preference (Coalition, Labor/Greens, or Other) rculture Respondent s cultural background (Caucasian, Asian or Other) rcob Respondent s country of birth (Australian, ESB, NESB) fate_yes Binary variable indicating whether respondent has a high sense of linked fate seifa_high Binary variable indicating whether respondent lives in socio-economically advantaged area 17

19 Table 4: Effect of Minority Candidates on External Efficacy (Observational) (1) (2) (3) (4) Overall By Candidate Party Vote for Coalition Vote for Labor/Greens Demographics Age (years) * * * * ( ) ( ) ( ) ( ) Age (squared) * * * ( ) ( ) ( ) ( ) Female (0.0301) (0.0295) (0.0347) (0.0310) Income (0.0264) (0.0251) (0.0361) (0.0190) Percent electorate foreign-born * * (0.0730) (0.0618) (0.110) (0.197) Political Characteristics Voted for winner (0.0839) (0.0831) (0.0749) (0.0325) Political interest * * ** (0.0232) (0.0232) (0.0515) (0.0100) Party ID Strength * * (0.0217) (0.0212) (0.0520) (0.0476) Education Postgrad Degree/Diploma * * * (0.0545) (0.0543) (0.0438) (0.0839) Bachelor Degree * (0.0837) (0.0861) (0.135) (0.0350) Undergraduate Diploma (0.0956) (0.0923) (0.136) (0.0851) 18

20 Associate Diploma (0.0876) (0.0892) (0.153) (0.0801) Trade qualification (0.0422) (0.0452) (0.0642) (0.104) Non-trade qualification (0.0496) (0.0492) (0.0333) (0.140) Descriptive Representation: Minority Voter (0.0282) (0.0377) (0.0513) (0.0258) Minority Candidate ** (0.0134) Minority Voter # Minority Candidate * (0.0484) Minority Labor Candidate * (0.0611) (0.208) (0.0370) Minority Voter # Minority Labor Candidate (0.0889) (0.574) (0.334) Minority Coalition Candidate * (0.0460) (0.166) (0.0458) Minority Voter # Minority Coalition Candidate (0.0854) (0.262) (0.133) Constant *** *** ** ** (0.0233) (0.0276) (0.0872) (0.138) Observations Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p <

21 Table 5: Marginal Effects for Minority Candidates on External Efficacy (Observational) Interaction Estimate Anglo Voter # Anglo Candidate [2.641,3.056] Anglo Voter # Minority Candidate [2.447,2.823] Minority Voter # Anglo Candidate [2.547,3.224] Minority Voter # Minority Candidate [2.754,3.220] 95% confidence intervals in brackets 20

22 Table 6: Marginal Differences for Minority Candidates on External Efficacy (Observational) Comparison Difference (Minority vs Anglo Candidate) for Anglo Voters ** [-0.271,-0.156] (Minority vs Anglo Candidate) for Minority Voters [ ,0.299] 95% confidence intervals in brackets * p < 0.05, ** p < 0.01, *** p <

23 Table 7: Marginal Effects by Candidate Party (Observational) Interaction Estimate Labor/Greens Candidate Coalition Candidate Anglo Voter # Anglo Candidate [2.444,3.048] [2.655,2.884] Anglo Voter # Minority Candidate [2.417,2.770] [2.263,2.878] Minority Voter # Anglo Candidate [2.561,3.214] [2.818,3.185] Minority Voter # Minority Candidate [2.905,3.307] [2.842,3.142] 95% confidence intervals in brackets 22

24 Table 8: Marginal Differences by Candidate Party (Observational) Comparison Difference Labor/Greens Candidate Coalition Candidate (Minority vs Anglo Candidate) for Anglo Voters * [-0.415,0.111] [-0.397, ] (Minority vs Anglo Candidate) for Minority Voters [-0.308,0.744] [-0.319,0.300] 95% confidence intervals in brackets * p < 0.05, ** p < 0.01, *** p <

25 Table 9: Marginal Effects by Candidate Party and Vote Choice (Observational) Interaction Estimate Coalition Candidate, Voted for Coalition Labor Candidate, Voted for Coalition Coalition Candidate, Voted for Labor Labor Candidate, Voted for Labor Anglo Voter # Anglo Candidate [2.963,3.013] [2.045,3.638] [2.121,3.080] [2.299,3.082] Anglo Voter # Minority Candidate [1.955,3.404] [2.634,3.019] [2.219,2.852] [2.034,2.858] Minority Voter # Anglo Candidate [1.604,4.501] [2.419,3.430] [2.165,3.694] [2.242,3.579] Minority Voter # Minority Candidate [1.791,4.016] [0.974,5.089] [2.386,3.796] [1.763,4.457] 95% confidence intervals in brackets 24

26 Table 10: Marginal Differences by Candidate Party and Vote Choice (Observational) Comparison (Minority vs Anglo Candidate) for Anglo Voters (Minority vs Anglo Candidate) for Minority Voters Coalition Candidate, Voted for Coalition Labor Candidate, Voted for Coalition Difference Coalition Candidate, Voted for Labor Labor Candidate, Voted for Labor * [-1.023,0.406] [-0.911,0.881] [-0.262,0.132] [-0.404, ] [-0.560,0.262] [-1.466,1.679] [-0.259,0.582] [-1.394,1.792] 95% confidence intervals in brackets * p < 0.05, ** p < 0.01, *** p <

27 Figure 1: Unconditional AMCE for Candidate Characteristics (Experimental) 26

28 Figure 2: AMCE for Candidate Characteristics by Respondent Party ID (Experimental) 27

29 Figure 3: AMCE for Candidate Characteristics by Respondent Culture (Experimental) 28

30 Figure 4: AMCE for Candidate Characteristics by Respondent Country of Birth (Experimental) 29

31 Figure 5: AMCE for Candidate Characteristics by Respondent Socio-Economic Advantage (Experimental) 30

32 Figure 6: AMCE for Candidate Characteristics by Respondent Linked-Fate (Experimental) 31

33 Figure 7: AMCE for Candidate Characteristics by Respondent Culture and Linked-Fate (Experimental) 32

34 Figure 8: AMCE for Candidate Characteristics by Respondent Country of Birth and Linked-Fate (Experimental) 33