Measuring the mass-elite preference congruence: findings from a meta-analysis and introduction to the symposium

The extensive scholarship devoted to the congruence of mass-elite policy preferences lacks consensus about the meaning, comparison, and measurement across political settings. This makes comparisons difficult and raises obstacles to advancing the debates. This symposium aims to identify the diversity of methodological choices and to reflect systematically on several key choices of particular importance in understanding the congruence. The contributions to the symposium compare and contrast how several types of measurement fare in diverse political contexts in Eastern Europe, Latin America, North Africa, and East Asia, and what we can learn from those methodological choices.


Introduction
Congruence in policy preferences between voters and elected politicians has been an asset of representative democracy (Pitkin 1967;Dahl 1971;Thomassen et al. 1999). Policy incongruence is, ultimately, a less-than-ideal description of political representation and a phenomenon with a potentially detrimental effect to the existence of democracy. For instance, extensive empirical research has shown that masselite incongruence tends to lower voter turnout and political trust while increasing dissatisfaction with democracy per se (Miller 1974;Ezrow and Xezonakis 2011;Curini et al. 2012). Indirectly, through its negative effects on representation, the absence of mass-elite congruence of policy preferences provided fertile soil for the growth of populism (Kriesi 2014). The mass-elite policy congruence lies at the core of the purpose of political representation through elections, which are "instruments of democracy to the degree that they give the people influence over policy making" (Powell and Powell 2000: 3).
Empirical research focusing on the comparison between mass and elite-level policy preferences has seen a dramatic increase in number and geographical coverage over time (Shim 2019). Starting with Miller and Stokes' (1963) seminal work on comparing the US voters and legislators in the 1960s, key empirical studies have covered developed European democracies either in single-or multi-country form, such as Barnes on Italy (1971), Dalton on Western Europe (1985), Converse and Pierce on France (1986), Granberg and Holmberg on Sweden (1996), Thomassen on the European Union countries (2005), and Belchior on Portugal (2008). Moreover, since the turn of the millennium, there is a number of empirical works covering democracies in Central and Eastern Europe (Rohrschneider and Whitefield 2012) or Latin America, East Asia, and Africa (Bornschier 2019;Jou et al. 2017;Fossati 2019;Kotzé and Steenekamp 2009).
Echoing this growing salience and diversity, the existing scholarship diverges substantially in terms of how it defines, compares, and measures the masses and elites. This has significant implications for whether and to what degree countries display a convergent preference of mass-elite policy preferences (henceforth, masselite congruence) and to what extent we can compare the derived results over time and between countries.
This symposium aims to contribute to this debate by systematically mapping out the diversity of methodological choices and to reflect upon several key choices of particular importance in the global context. Each measurement covered in this symposium is informed by the existing literature and carefully contextualized in light of the political environment in which the students of mass-elite congruence scholarship conduct measurement. Furthermore, in line with the increasing geographical diversity in the scholarship, evidence was gathered from countries in Eastern Europe, Latin America, North Africa, and East Asia.
The four contributions to the symposium present the results which raise several questions about measuring policy preferences at mass and elite levels. In contexts where a large proportion of voters do not have a clear identification with a particular party, how should we compare mass-elite congruence? How can we select specific issue items to capture the key policy dimensions of interest in analysed countries? What alternative measures should we use if we do not have survey results at hand vis-à-vis either the mass or elite level? Although established European democracies were not included in this symposium, addressing questions like these will provide useful insights for them too. Many experience an increase in non-partisan voters, the emergence of new policy dimensions that tend to cut across the existing ones and mixed results in mass-elite congruence evaluations that often lead to an assumed "peaceful coexistence of research results and conclusions" (Müller et al. 2012: 170) without much reflection. In this sense, the special issue is a collective endeavour to provide a springboard from which students of mass-elite congruence can reflect on and discuss the globally relevant methodological choices they make.
The remaining text of the introduction reviews the key measurement choices in the literature based on an original meta-analysis of the latter and situates the four contributions to this symposium within a particular set of those measurement choices.

A meta-analysis of the research on mass-elite congruence
To examine the prominent patterns of measurement choices in the current literature, we conducted a meta-analysis of 100 empirical works that compare the preferences of masses and elites. All these works use quantitative research methods and data sets. The meta-analysis shows how the existing literature has so far defined the masses and elites, compared their preferences in different policy areas using various points of comparison and methods of aggregation, and used different data sources too. Informed by the key patterns revealed through our meta-analysis, four subsequent contributions to the symposium compare and contrast how various types of measurement choice fare in diverse political contexts.
In an attempt to choose widely read and well-qualified empirical contributions to the mass-elite congruence literature, we used the keyword search function 1 from the Web of Science citation database-with the selection parameters confined to English-language academic works in Political Science published up to and including 2018 (for details about the selection process and the full list of empirical works chosen, see "Appendices 1 and 2"). 2 In view of our selection method, the 100 empirical works included here represent important and widely known ones in the extant literature; therefore, the diverse methodological choice patterns outlined in this introductory essay reflect the current state of the art regarding empirical works pertinent to mass-elite convergence.

Defining masses and elites
The most straightforward definition of elites and masses is those who have electoral mandates and those who have the right to give a mandate, respectively-elected legislators and eligible voters. Comparing the preferences of the electorate as a whole and of the parliament as a whole echo the notion that representative democracy should enable every section of society to produce itself in full light (Mill 1861). However, empirical evidence does not necessarily follow this definition. When it comes to defining the masses, 80 per cent of works label them rather loosely as citizen/public/voters without a clear specification demonstrating to what extent the selected sample represent eligible voters, e.g. how research deals with eligible voters who are unlikely to be found in the sample such as overseas citizens (Table 1).
More importantly, the meta-analysis shows that 20 per cent of the analysed texts often confine the scope of masses to partisan voters and compares their preferences with those of the parties that they support. This approach can be problematic in numerous new democracies characterized by high numbers of voters without party identification. Even for old democracies, partisan de-alignment has been an ongoing trend as a consequence of social and political modernization (Dalton 2002); as a result, independent voters nowadays make up a non-negligible portion of eligible voters-ranging from 20 to 50 per cent in many advanced industrial societies (Dalton 2002).
The existing literature clearly shows more variation in how to define elites than how to define the mass. For instance, 40 per cent of works include parties as the point of elite-level preference aggregation (based on party manifestos or the party leader's revealed policy position) and often compare the aggregated value to that of either general or partisan voters. As for the analyses based on individual-level preference aggregation, the extant works select the scope of elites as candidates running for an upcoming election or alternatively as elected legislators. Moreover, the scholarship also suffers from the oft-noted problem arising from the representativeness of the sample. Respondents are not randomly selected, and the survey has very low response rates in most cases making all inferences tenuous (Costello et al. 2012). There is also unclear justification for why research focuses on either candidates or elected legislators. Overall, a specific definition of masses or elites requires more clear justification in light of the disparate contexts of covered countries.

Policy dimensions and issue items
In addition to how the literature defines masses and elites, another challenge that the literature faces-alongside the growing diversity of the measurement contextis which political dimensions to compare, based on which specific issue items. The meta-analysis results demonstrate that three quarters of the literature focus on mass-elite differences in specific policy dimensions, beyond just general left-right differences. This pattern indicates the diminishing ability of a broad left-right distinction to summarize meaningful policy stances between parties and electorates, with specific content often dependent on the country, time period, and respondent in question (Lachat 2018). Nonetheless, more than 80 per cent of the empirical Legislators 37 Ruling party legislators 9 Government policies 3 analyses included in this paper start mass-elite representation gap measurement with the three most well-known policy dimensions: economic (e.g. big vs. small government), sociocultural (e.g. materialist vs. post-materialist), and foreign policy issues (e.g. globalist vs. nationalist). However, the three policy dimensions draw closely from the experiences of old democracies. Once we step beyond existing boundaries to include also new democracies, the potential number of key policy areas, where mass-elite discrepancies exist, expands dramatically. For instance, a survey of the existing party politics literature (Deegan-Krause 2007; Shim 2019) points out that a major axis of party competition revolves, for instance, around democracy in Latin America, ethnicity and regionalism in many African countries, religion and identity in Middle Eastern states, religion, ethnicity, and caste in Southeast and South Asia, or self-determination, defence, and security issues in Northeast Asia.
For countries lacking either sufficient democratic experiences and/or devoid of an institutionalized party system and programmatic party-voter linkages, it is challenging to identify what "key policy dimensions" are to begin with. How do we define which policy dimension is "key" in a specific country at a particular time? Furthermore, should we focus on key policy dimensions at the mass level and examine if a discrepancy exists in the same dimension at the elite one? Or should we instead reverse this sequence? Based on the Tunisian case, Farag's contribution in this symposium suggests using the degree of divisiveness to determine the "key-ness" of a given policy dimension. He also recommends that research should start from the mass level, in view of an established democratic representation theory.
To complicate matters further, even if one does identify which policy dimensions to compare between masses and elites, choosing specific issue items that can meaningfully capture the pertinent policy dimensions requires careful attention being paid by the researcher. For instance, although the most prominent foreign policy dimension across EU countries is one's attitude towards the role of the European Union, the meta-analysis shows that specific issue items employed in given research vary quite substantially concerning currency (to keep one's own or not), borders (to remove national borders or not), social security (to have an EU-wide massive employment program or not), defence (creating an EU-wide army or not), and immigration (responsible entity for hosting immigrants: EU or the nation-state?). Which specific issue items are more salient and relevant might depend on a member state's level of EU integration issues faced at the time of measurement.

Various comparisons: issue saliency, relative congruence, and indirect measurement
The most common empirical approach has been to compare the absolute difference between the masses and elites on their self-identified policy positions. However, as is clear from the meta-analysis patterns shown in Table 2, a non-negligible portion of empirical works focuses on issue-saliency differences, compare relative congruence, and often use indirect methods to estimate mass-or elite-level policy preferences. First, if the issue-position approach asks respondents whether or to what extent they agree/disagree with a particular policy issue, we can characterize the issue-saliency one by its attempt to gauge respondents' priority in particular policy domains (among many others). Comparing mass-elite issue-saliency congruence is becoming increasingly common in the party competition literature, since it is a complementary approach to spatial theories (Alonso 2012)-in addition to its own significant effect on voter turnout or levels of satisfaction with democracy (Reher 2014).
Reflecting this trend, in this symposium McElwain's contribution and Farag's contribution both approach their analyses from both issue-position and issue-saliency perspectives and demonstrate that the representation gap differs substantially between the two. Moreover, by distinguishing incumbent candidates from challengers, McElwain adds important nuance, showing that challengers, in particular, are more likely to strategically prioritize issue saliency or change issue positions vis-àvis their constituencies due to their electoral insecurity.
Second, while we can define absolute congruence as the policy position distance between the masses and elites on a continuous scale, relative congruence-often labelled as "responsiveness" (Wlezien 2017)-frequently takes the form of regressing elites' policy stances over their supporters/constituencies in single or multiple time periods.
Finally, in contrast to "direct" comparisons based on the self-placement of one's policy position, the "indirect" comparison indicates that masses and elites' policy positions are the result of perceptions or assessments. For example, the elites judge voters' policy positions or the voters assess elites' preferences. Alternatively, a third party, i.e. experts, can make an assessment. The meta-analysis reveals that the most common indirect approach in the existing literature has been partisan voters deciding the policy positions of the parties that they support.

Policy preference aggregation methods: averages, distributions, and directions
The meta-analysis informs us that 20 per cent of the research employs a simple uniscale measurement in evaluating the mass-elite policy preference gap, for example by comparing "yes" or "no" answer percentages for questions such as "Do you agree with Brexit?" or "Should the USA take an active role in world affairs rather than stay out?" However, the remaining 80 per cent of empirical works utilize a multiscale measurement scheme where respondents answer given policy preference questions-whose scales often range from 0 to 5, or 0 to 10. On this, the existing literature notes three major factors worth considering: averages, distributions, and directions. First, we can base the degree of discrepancy between the masses and elites on the average (either mean or median) placement of voters and of elites' preferences in a particular policy dimension-such as the two-point gap between elites and masses on a 0-to-10 scale range vis-à-vis the abortion issue. As is clear from Table 3, this method is dominant in measuring the mass-elite representation gap; at the same time, it is easy to interpret too. However, despite the same mean/median placement, the preference of one side can be scattered while the other one can be centred on the mean placement. To address this problem, there have been several systematic attempts to bring both the mass-and elite-level preference distributions into the measuring of the masselite representation gap. The methodological innovation of Golder and Stramski's (2010) measure on cumulative-distribution function forms the basis for further advancement in the field. Other measurements exemplify this technique such as the earth mover's distance by Lupu et al. (2017) or the Kolmogorov-Smirnov Test by Stavrakakis et al. (2017). Finally, the direction of discrepancy examines to what extent the masses and elite agree, either taking positive or negative views on particular policy dimensions (Rohrschneider and Whitefield 2017). This is an underrecognized (but crucial) aspect of convergence since, in the real-world setting, many important issues take binary form at the actual decision stage. Because there can be cases where the mean placement and distribution discrepancy between the masses and elites are the same but differ in terms of alignment in direction, we need to treat this as a separate aspect of any mass-elite discrepancy. The existing literature often calculates alignment of mass-elite direction by comparing "majorities", which is derived by dichotomizing scales and computing the proportion of opinion on each side.

Varieties of data sources and setting analytical equivalence
The meta-analysis results clearly reflect the oft-noted problem of using different data sets and metrics in measuring convergence (Reher 2014). The 100 empirical works approximately cover 60 different data sets at the mass level and 80 at the elite one. Even if we narrow down our focus to 21 empirical works measuring the representation gap related to the EU integration issue across member countries, substantial diversity still exists: six and 12 different data sets cover the mass and elite level, respectively. If we divide the data sources by type, we can observe greater variety at the elite level-ranging from roll-call votes and bill sponsorship, to party Multi-scale measure (direction) 1 manifestos and government policy outputs, to expert judgements and opinion polls (see Table 4).
Considering the increasingly global orientation of the literature, students of masselite congruence use a wide range of diverse data sources. In itself, it is not a problem. This is especially true since new democracies or transition countries often lack reliable and valid cross-national data to measure mass-elite congruence on various policy areas. Country-specific data sources are relevant in such a context. However, not all data sources are equal in terms of the motivations behind data generation, the extent of coverage (e.g. to what extent particular data sets include various issues and diverse political actors), or the measured latent dimensions (e.g. whether the data captures issue positions or issue saliency). In view of the current state of the literature, which witnesses the peaceful coexistence of the research results and conclusions based on different data sources (Müller et al. 2012), what we need is careful scrutiny of the reliability of findings based on various robustness tests-and clear justifications for chosen data.

Content and structure of this symposium
McElwain's contribution to this symposium focuses on surveys of Japanese election candidates and voters on identical policy questions related to one of the key political issues in post-war Japan: constitutional revision. Based on remarkably high levels of elite-level response data, McElwain's findings make clear that, candidates, incumbents, and challengers have different policy priorities and positions on Japan's constitutional revision issue. The author notes that challengers are particularly vulnerable to voter preferences due to their electoral insecurity. The article explains the necessity to differentiate among elites and the masses as well as between policy preference and salience. It calls for a more granular approach in the literature to provide more accurate insights into the nature of democratic representation.
The article written by Bornschier covers four Latin American countries and addresses the context-specific nature of issue items capturing particular policy dimension. He warns against applying a predefined set of issue items-since their saliency and relevance can vary between countries, and over time. Alternatively, he suggests an inductive yet systematic approach-Linear Canonical Discriminant Analysis-that we can utilize to capture mass-elite congruence on key policy dimensions in a scale-free manner. Moreover, the analysis suggests a scale-free relative congruence method that compares to what extent policy positions correlate between partisan supporters and their parties. Considering that the degree to which voters are able to link specific issue items to germane policy dimensions exhibits greater cross-sectional variation outside established democracies (Harbers et al. 2013), this relative congruence method represents a much-awaited toolkit for issue-congruence scholars covering new democracies. Farag's contribution focuses on a new and the only Arab democracy-Tunisia-and measures mass-elite congruence using mass survey, party manifestos, and roll-call votes. The paper takes a systematic approach to derive two key policy dimensions relevant in the Tunisian society since the Arab Spring-democraticauthoritarian and secular-Islamist-and demonstrates that the evolving centrist positions of the two largest Tunisian political parties on both dimensions paves the way to mass-elite incongruence. However, from an issue salience perspective, the article shows that there is congruence between masses and elites about economy being the main policy dimension. The distinction between the level of analysis (mass vs. elite) and the policy perspective (position vs. salience) yields theoretically important insights for the study of mass-elite differences in new democracies.
The article co-authored by Bankov and Gherghina shows how we can measure the mass-elite representation gap with a qualitative data source, i.e. speeches of political leaders. Their analysis on Bulgaria and Romania carefully demonstrates the implications of measurement involving both qualitative and quantitative types of data. Faced with the lack of elite-level surveys and other reliable sources of data on the EU integration issues, they tap into the best possible alternative data source in the two countries. In a transparent step-by-step process, the authors show how we can use qualitative data source to measure representative elite-level policy preferences in a reliable manner. Although the method requires extensive case knowledge and linguistic ability, it demonstrates its potential as an additional methodological toolkit for scholars facing data shortage problems regarding new democracies.
Acknowledgements The authors would like to express their gratitude to Christopher Wratil, Michael Scanlan, Miriam Sorace, and Thomas Lundberg for their careful reading of earlier drafts of several symposium contributions and thoughtful suggestions made to improve them. The suggestions of the anonymous reviewer were particularly helpful. The research also benefited from presenting findings at 2019 ECPR Joint Session, the PETGOV Seminar at the University of Amsterdam, the Research Cluster in Comparative Politics Seminar Series at the University of Glasgow, and the Accountability and Participation Research Programme meeting of the German Institute of Global and Area Studies.
Funding Open access funding provided by Projekt DEAL. This research was supported be Fritz Thyssen Foundation (Grant Number: Az.40.16.0.029PO) and the Japan Society for the Promotion of Science (Grant Number: 16F16308).
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.

Appendix 1: The selection procedure
1 Data source: we selected empirical works based on an established science search engines-Web of Science on December 15, 2018. 2 We used four relevant key terms: "issue congruence"; "opinion congruence"; "issue representation"; and "policy representation." The publication format includes book chapters, articles, and conference papers, and the search parameters included all time periods-but were confined to English-language academic works in Political Science. 3 We listed the search results in order of "relevance" and then manually winnowed out empirical works concerning mass-elite representation gap measurement. Here, empirical works are defined as quantitative ones comparing the preferences of elites and masses on the basis of particular data sets. 4 Since the number of relevant empirical works obtained through the search engine results was only 42, far below the intended target of 100, we employed the snowballing method for the reference lists of the 42 obtained works so as to secure the other 58. 5 To maintain the diversity of the sample, we chose maximum two works from the reference list of each of the 42 empirical works. 6 Limitations: (1) key terms-based selection reliant on "web-built search algorithms" and narrowing down empirical works to those published in English inevitably excludes important parts of the mass-elite congruence literature which does not include four keywords or/and written in languages other than English, for example specific area-based journals; (2) manually winnowing out relevant empirical works and the subsequent application of the snowballing method might have introduced potential bias.