In addition to sociologists, scholars from many different fields, including history, political science, psychology, and geography, have studied elections and voting behavior. In current American sociology, however, these topics are largely neglected. Major advances have been made in related disciplines, yet as of one of the pioneers, the sociologist Rice (1928, p. p.vii) stated: ”The phenomena of politics are functions of group life. The study of group life per se is a task of sociology.” In general terms, despite variations in emphasis between different approaches, the sociological study of voting behavior is concerned with the way individuals obtain, select, and process information related to the political arena; the various forces that shape this process; the relevance individuals attribute to the political sphere; and how they decide to participate in or refrain from specific political actions. Elections provide a convenient focus, a point where the often elusive and latent processing of political information manifests behavioral correlates such as voting or abstaining and supporting one candidate or the other. In contrast, forecasting election returns is not a primary goal of the sociological study of voting behavior, although the general public, parties, and politicians are interested mostly in this aspect. Much applied research served these immediate needs and interests in the past and continues to do so. Still, in the field of voting behavior, pure (academic) and applied research peacefully coexist; cross-fertilization rather than mutual irreverence characterizes their relationship.
The study of voting behavior began in the late eighteenth century (Jensen 1969), although most of the very early work does not meet strict scholarly standards. In the course of its development as an academic discipline, two different strands that are still discernible have emerged. The first strand-aggregate data analysis—is characterized by the use of actual election returns compiled for geopolitical units such as wards, districts, and counties. Those returns are compared with census data, providing a sociodemographic profile of those areal units. Starting in the late nineteenth century, there developed a school of quantitative historiography that made extensive use of maps representing voting and/or census information by using different shades and colors (Frederick Jackson Turner in the United States and Andre Siegfried in France). The mere visual inspection and somewhat subjective interpretation of those maps by the Turner school were supplemented and then replaced by more vigorous statistical techniques, in particular correlation analysis, inspired by the sociologist Franklin Giddins at Columbia. One of Giddins’s students, Rice (1928), demonstrated the utility of quantitative methods in politics. At the University of Chicago, interdisciplinary cooperation in the social sciences produced some of the most outstanding work of that time (e.g., Gosnell 1930). The advent of modern survey research in the 1930s and 1940s, however, obscured the aggregate approach for quite some time.
The second strand in the study of voting behavior—analysis of survey data—also had some early forerunners. Polling individuals about their voting intentions (”straw polls”) or past voting decisions started in the late nineteenth century. In one of the most extensive efforts, more than a quarter million returns from twelve midwestern states were tabulated by a Chicago newspaper for the 1896 presidential contest between McKinley and Bryant. In the 1920s, straw polls conducted by newspapers and other periodicals were quite common and popular. Their reputation was ruined, though, by the failure of the Literary Digest poll to foresee the landslide victory of Franklin D. Roosevelt in the 1936 election. By that time, however, pioneers of modern public opinion research such as George Gallup, Archibald Crossley, and Elmo Roper had started to use more rigorous sampling methods as well as trained interviewers to ensure a proper representation of all strata of the electorate (Gallup [1944] 1948).
Interest in voting and political behavior and concern with mass communication, marketing strategies, and the public’s attitude toward World War II stimulated the rapid development of modern survey research from about the mid-1930s through the 1940s and the establishment of survey research centers in both the academic and commercial sectors (Converse 1986). These centers include the Survey Research Center/Institute for Social Research (ISR) at the University of Michigan and the National Opinion Research Center (NORC) at the University of Chicago on the academic side and Gallup’s American Institute of Public Opinion on the commercial side, to name just a few early organizations that are still leaders in the field.
Modern voting research based on the survey method typically uses small but randomly selected samples of about 1,000 (rarely more than 2,000) eligible voters. Information is collected through the use of standardized questionnaires that are administered by trained interviewers in person or increasingly over the telephone. Advances in modern communication technology such as the Internet are likely to change the face of scholarly survey research even more drastically in the very near future. ”Standardized” means that a question’s wording is predetermined by the researcher and that the interviewer is supposed to read questions exactly as stated and in the prearranged order. For the most part, the response alternatives also are predetermined (”closed questions”); sometimes, for select questions, verbatim answers are recorded (”open questions”) and subsequently sorted into a categorical scheme. In contrast to aggregate-level analysis and the use of official election returns, survey-based research on voting behavior relies on self-reports by individual citizens. Thus, it is subject to bias and distortion resulting from question wording, dishonest answers, memory failure, and unstable attitudes even if the sample is properly drawn. Its major advantage is the unequivocal linkage of demographic traits (e.g., age, sex, ethnicity, and social class) and political attitudes and behavior on the level of the individual.
AGGREGATE DATA ANALYSIS
The use of aggregate data in studying voting behavior poses formidable methodological problems, yet it is the only approach available to study voting behavior before the mid-1930s. For example, the Germans voted Hitler into power in genuinely democratic elections in the late 1920s and early 1930s. The voting behavior of Germans in the Weimar Republic has been subject to much debate and controversy in political sociology. The earlier consensus that Hitler’s support came predominantly from the lower middle classes was challenged by later studies (e.g., Childers 1983; Falter 1991) that contended that his support had a much wider base cutting across all social groups.
Findings based on aggregate data analysis often depend heavily on seemingly technical details of preparing the database and the choice of specific statistical techniques. As a rule, findings are more reliable if the geopolitical units are small. However, even if the greatest care is exercised, there is always the danger of an ”ecological fallacy.” To use a contemporary example, if the vote for a white candidate increases with a rising percentage of white voters across voting districts, it is plausible to assume that people have voted along racial lines, yet this need not be the case. Perhaps ethnic minorities in predominantly white districts are more likely to vote for a white candidate than they are elsewhere. Therefore, they, not the white voters, may be responsible for the increased share of the white candidate.
A solution to the ecological inference problem has been proclaimed (King 1997), but despite some progress, that claim appears to be overstated. In spite of all the remaining shortcomings, though, aggregate data analysis is an indispensable tool for tracing patterns of voting behavior over time in a sociohistorical analysis (e.g., Silbey et al. 1978) or analyzing contemporary voting behavior when sufficiently detailed and reliable survey data are not available. Particularly for local or regional studies, there may not be sufficient funds to conduct appropriate surveys or the research interest may develop only after the elections have taken place.
SURVEY-BASED VOTING RESEARCH
The Columbia School. Four landmark studies connected with the presidential elections of 1940, 1948, 1952, and 1956 mark the establishment of scholarly survey-based research on voting behavior (Rossi 1959). In essence, those studies provided the core concepts and models used in contemporary voting research. Reviewing those studies provides an introduction to present-day theories of voting behavior in U.S. presidential elections, while congressional elections typically follow a very simple pattern: Incumbents are rarely defeated.
The first two studies were conducted by Lazarsfeld and his associates at Columbia University. Their main intention was to ”relate preceding attitudes, expectations, personal contacts, group affiliations and similar data to the final decision” (Berelson et al. 1954, p.viii) and trace changes of opinion over the course of a campaign. Emphasizing the particular set of political and social conditions and its importance for this process, the Columbia group restricted its studies to one community (Erie County, Ohio, in 1940 and Elmira, New York, in 1948) and interviewed the same respondents repeatedly: up to seven times in 1940 and four times in 1948. Repeated interviews, or a ”panel design,” became a standard feature of more sophisticated voting studies, while the major studies to come abandoned the focus on one community in favor of nationwide representation.
Several major findings emerged from the Erie County study (Lazarsfeld et al. [1944] 1968). First, people tend to vote as they always have, in fact as their families have. In the Michigan school of voting behavior (see below), this attitude stability was conceptualized as ”party identification,” a stable inclination toward a particular party that for the most part develops during adolescence and early adulthood.
Second, attitudes are formed and reinforced by individuals’ membership in social groups such as their social class, ethnic group, and religious group and by the associations they belong to. More concretely, the research team found that people of lower social status, people in urban areas, and Catholics tended to be Democrats while people of higher social class, people in rural areas, and Protestants were more likely to be Republicans. Subsequently, the alliance of particular segments of the population with specific parties was amply documented despite modifications in its particular form. More so than in the United States, voting behavior in the major European democracies (notably Britain and West Germany) could be explained largely by the links between social groups and particular parties (Lipset and Rokkan 1967), although those links have been weakening.
Third, change does occur, and people under cross-pressures are the most likely to change. A cross-pressure occurs when the set of different group memberships provides conflicting stimuli. For example, in 1940, Protestant blue-collar workers experienced a pull toward the Republicans on the basis of their religious affiliation and a pull toward the Democrats because of their class position. In the United States today, the impact of religious affiliation is more complicated, but the general notion of cross-pressure remains important.
Fourth, Lazarsfeld and colleagues developed the concept of a ”two-step flow of information.” According to this concept, most people are not directly persuaded by the mass media even if they are susceptible to change. Instead, they tend to follow opinion leaders, who are the informal leaders in the various social networks (family, friends, associates at the workplace) in which individuals are involved. These leaders pay close attention to the media; they redisseminate and validate media messages. With the ever-increasing impact of mass media (television and more recently the Internet) over the last fifty years, this result of the 1940 study may not reflect the situation today. Unfortunately, it is very difficult to prove media effects conclusively, and the cumulative empirical evidence has not been able to settle a long-standing controversy about the extent of media effects.
The 1948 Elmira study was designed to test further and if necessary modify the findings of the earlier study and integrate the results into the body of existing knowledge (see Berelson et al. 1954, pp. 327-347, for a comparative synopsis of several major studies). As a matter of fact, its main contribution lies in the refinement of several aspects that were not covered sufficiently in the Erie County study. However, the Elmira study still failed to show systematically the links between the efforts of the various institutions in the community and the decisions of the voters. The focus on those links was the key rationale for limiting these studies to one community, a feature that invites doubt whether the findings can be generalized to all American voters.
The Michigan School. The sociological approach of the Columbia school was subsequently overshadowed by the social psychological model of the Michigan school that came to dominate survey-based voting research for many years. After a smaller study in 1948 (Campbell and Kahn 1952), the Michigan team, led by Campbell, conducted major studies in 1952 and 1956 (Campbell et al. 1954; Campbell et al. 1960). In contrast to Lazarsfeld and associates, their studies used national samples, thus expanding the geographic coverage, but only two interviews, one shortly before and one shortly after the elections. In addition, the Michigan group introduced far-reaching changes in the conceptualization of the voting process. On the basis of its national study of 1948, those researchers felt that social group memberships have little direct impact on the voting decision. Instead, they focused on ”the psychological variables which intervene between the external events of the voter’s world and his ultimate behavior” (Campbell et al. 1954, pp. 85-86). In particular, they considered three concepts: party identification, issue orientation, and candidate orientation. Party identification refers to the sense of personal attachment an individual feels toward a party irrespective of formal membership or direct involvement in that party’s activities. It is thought of as a stable attitude that develops early in life. In contrast, both issue orientation and candidate orientation depend on the context of a particular election. Issue orientation refers to individuals’ involvement in issues they perceive as being affected by the outcome of an election. For example, if individuals are concerned about the economy and feel that it makes a difference whether the country has a Democratic or a Republican president, this will have an impact on their voting decisions. Similarly, candidate orientation refers to individuals’ interest in the personality of the candidates and to a possible preference that derives from the personal traits of the candidates. For example, Ronald Reagan portrayed himself as a firm and determined leader but also as a caring and understanding father. In that way, he was able to attract many voters otherwise attached to the Democrats.
The Michigan model posits a ”funnel of causality.” The social factors emphasized by the Columbia school are not dismissed outright but are viewed as being at the mouth of the funnel, having an indirect effect only through the three central psychological variables, particularly party identification. Party identification in turn affects issue orientation and candidate orientation as well as having a direct effect on the voting decision. The simplicity of this model is both its strength and its weakness. It clearly marks the shift of emphasis to psychological processes of individual perception and evaluation, but it does not explicitly address the social and political context. However, in The American Voter (Campbell et al. 1960), the Michigan group presents a much more comprehensive analysis of the 1956 elections, addressing topics such as the role of group membership, social class, and the political system without, however, explicitly expanding the basic model.
Additional concepts that have been used widely in subsequent research include the concept of a normal vote and the typology of elections as maintaining deviating, or realigning (Campbell et al. 1966) and an assessment of mass belief systems (Converse 1964). The concept of a normal vote follows directly from the basic model: If all voters follow their long-standing inclinations (vote according to party identification), they produce a normal vote. Comparing actual election returns with the (hypothetical) normal vote allows one to assess the impact of contemporaneous, mostly short-term factors. In a maintaining election, the party with the larger number of partisans wins, but its vote share may be somewhat different from its normal share as a result of short-term factors. If short-term factors lead to the defeat of that party, the elections are considered deviating. Realigning elections mark a major shift in basic allegiances. Such shifts are rare and typically are not accomplished in a single election. In the 1930s, the American electorate shifted toward the Democrats as a consequence of economic depression and Roosevelt’s New Deal, which promised a way out. However, given their long-term nature, processes of dealignment and realignment are difficult to determine in strict empirical terms (Dalton et al. 1984; Lawrence 1996).
With respect to the nature of mass belief systems, Converse’s (1964) article triggered a long-lasting debate that has been settled. Converse asserted that the vast majority of the American people have little interest in politics, that their opinions on issues lack consistency and stability over time, and that those opinions are mostly ”non-attitudes.” Consequently, a large portion of the electorate does not vote at all; if those people do vote, their vote is based mostly on partisanship and/or a candidate’s personality, not on an independent and careful evaluation of the issues.
Critique of The American Voter and subsequent refinements. Like other landmark empirical studies, The American Voter was not exempt from sometimes radical critiques that can be grouped into three categories: challenges to the allegedly derogatory image of the American electorate and its implications for the democratic process, assertions that the findings are valid only for the 1950s, and a methodological critique of operationalization, measurement, and model specification. Most of the methodological critique is too technical to be discussed in this article (see, Asher 1983).
One of the earliest and most vocal critics was Key (1966). Using a reanalysis of Gallup data from 1936 to 1960, he developed a typology of ”standpatters,” ”switchers,” and ”new voters” and asserted that the global outcome of those elections followed a rational pattern derived from an appraisal of past government performance. Hence, as a whole, the electorate acts responsibly despite the fact ”that many individual voters act in odd ways indeed” (Key 1966, p. 7).
The most comprehensive effort to review American voting behavior over time, a critique of the second type, was presented by Nie et al. (1976), based on the series of Michigan election studies from 1952 to 1972. Still working within the framework of the Michigan model, those authors found significant changes in the relative importance of its three central factors: a steady decrease in the level of party identification, particularly among younger groups, and a much stronger relative weight of issue orientation and candidate orientation. In a turbulent period of internal strife and social change (civil rights, the Vietnam War, Watergate), the electorate became more aware of issues and much more critical of parties and the established political process. Nie et al. (1976) found a decomposition of the traditional support bases for both Democrats and Republicans, all adding up to an ”’individuation’ of American political life” (p. 347).
Still largely following the path of the Michigan school, much research in the 1980s was directed toward issue voting, which reflects a continuing decline in stable party attachments through political socialization and/or group memberships. In particular, the impact of economic conditions on electoral outcomes was investigated both in the United States and in other major Western democracies (Eulau and Lewis-Beck 1985; Lewis-Beck 1988; Norpoth et al. 1991). The findings were diverse, contingent on the specification of the research question and the national context, yet one general pattern emerged and was confirmed beyond the United States in many national elections in the 1990s: Political actors perceived as better able to handle economic matters than their competitors have a significant advantage, and perceived economic competence is strongly related to an individual’s voting decision.
Fiorina’s (1981) concept of retrospective voting can be seen as a bridge between the classic Michigan model and the newer directions that have emerged in the last two decades. Fierina posits that both party identification and issue orientation are largely dependent on the evaluation of past government performance. Party identification thus represents a sort of running tally of past experience. It is still a long-term influence, but it is subject to gradual change and is based more on cognition than on affection.
More Recent Approaches. With some simplification, one can discern three major directions in voting research in the last fifteen to twenty years, each anchored in a different discipline: Rooted in economic theory, rational choice models have been applied to voting (as well as to many other forms of social behavior); drawing on more general psychological theories, the subfield of political psychology studies in a comprehensive way how the individual perceives and processes political information, with voting being only one specific aspect; and the focus on reference groups on both the macro level such as class or religion (an important strand in European research on voting) and the micro level (social networks) has reemphasized the sociological perspective.
Rational choice models see the voter as carefully evaluating the pros and cons of each party or candidate, assessing its utility (consumer models as in Himmelweit et al. 1985) and proximity to the voter’s own position (spatial models as in Enelow and Hinich 1984, 1990; directional models as in Rabinowitz and MacDonald 1989), and then voting for the closest or most useful party and/or candidate. It is doubtful, however, whether such models adequately portray the actual process of reaching a voting decision except among the small segment of highly informed and highly motivated citizens. The rational choice approach has been critized more generally as conceptually inadequate for the analysis of mass political behavior (e.g., Green and Shapiro 1994). Other criticism has come from within the rational choice camp, leading, for example, to an ”expressive model of voting” (Brennan and Lomasky 1993) that shifts the focus away from the instrumental aspect of voting but maintains the conceptual and terminological framework.
The literature in political psychology is too extensive to be discussed in detail here; overviews are provided by Lau and Sears (1986), Sniderman et al. (1991), and Mutz et al. (1996), among others. However, the contribution of this approach to the discussion of two long-standing controversial topics must be mentioned explicitly: media effects and political belief systems. In regard to media effects, a number of studies using more refined concepts (agenda setting, framing, priming) have suggested a more direct and stronger impact of media than previously had been assumed (see Kinder 1998; Zaller 1996), but a generally accepted model of how the mass media influence the political process has not emerged. In regard to political belief systems or the lack thereof, the notion of a politically uninformed or ignorant and irrational electorate has strong implications for normative theories of democracy and, on a practically level, campaign strategists. There continues to be strong and largely undisputed empirical evidence that many Americans have rather limited factual knowledge of specific bills and policies, the makeup of political institutions, and even their elected representatives. However, it is controversial whether that lack of specific knowledge matters and whether voters are nevertheless able to make ”correct” or ”reasonable” choices on the basis of more implicit forms of information processing (Delli Carpini and Keeter 1996, Lau and Redlawsk 1997). In both areas, political psychology has not settled the controversy, but it has provided enhanced conceptualizations to guide more productive empirical analyses.
In regard to sociological perspectives, community studies (as in the Columbia school) have attempted to assess the impact of the local context on the decision making of the individual. Context information is gathered by using block-level census data or tracing and interviewing members of the social network of the primary respondents (Huckfeldt 1986; Huckfeldt and Sprague 1995).
On the macro level, the social cleavage approach, long dominant in European voting research (see below), has gained some attention in the United States (see Brooks and Manza 1997).
Voting Behavior in Other National Contexts.
The Michigan school of voting behavior has continued to have a major impact on the emergence of survey-based voting research in other Western democracies (see Beyme and Kaase 1978; Butler and Stokes 1976; Heath et al. 1985; Rose 1974) and electoral research today (e.g., Kaase and Klingemann 1998). A strict replication of the basic model, however, has rarely been feasible because of considerable differences in political systems, party organizations, and electoral rules. In particular, attempts to devise valid measures of the key concept of party identification have produced mixed results at best (e.g., Budge et al. 1976).
Despite considerable variation across European democracies and within specific countries, political parties articulate specific programmatic positions derived from basic ideological beliefs that bind all party members, and parliamentary votes typically follow party lines. Political parties, then, dominate the political contest, and most major parties have their roots in social cleavages relating to class, religion, region, or ethnicity. Consequently, such social group memberships have been powerful determinants of voting behavior (Lipset and Rokkan 1967). Much of the European debate in the last twenty years, however, has focused on evolving changes in the electorate that result in more volatile voters. First, the impact of social (class) origin on the individual life course in general (educational attainment, occupational opportunities, marriage and family, etc.) has declined, leading to a more idiosyncratic definition of self-interest and as a consequence a less predetermined voting pattern. Second, a growing disenchantment with established political parties and politicians in general has weakened the once highly internalized norm of political participation through voting and increased the propensity to vote for new and often extremist parties as a token of protest. Thus, the once stable alliances between parties and certain segments in the electorate along social cleavages have weakened (Crewe and Denver 1985; Franklin et al. 1991; Miller et al. 1990), though there is no consensus whether ”class voting” has ended (e.g., Manza et al. 1995).
The emergence of new democracies after the demise of the Communist bloc in eastern Europe in the late 1980s and the continuing process of European integration, including the creation of European Union as a single political and not just economic entity, have created new challenges and opportunities for voting research in Europe. With the introduction of a common currency in 1999, a continuing reduction of national sovereignty, and a generally strengthened position of the European Parliament in exercising control over the executive branch (European Commission), European elections will lose their stigma as second-order elections. Studies of European elections in 1974, 1979, and 1994 suffered from limited funding and broke no new conceptual ground. However, they did establish the fact that national rather than pan-European issues dominated voting choices (Schmitt and Mannheimer 1990; Van der Eijk and Franklin 1996). This is likely to change, but it is too early to tell the role these elections will play in the further process of European unification and whether voting behavior in those elections will follow patterns different from those in national elections.
The methodological problems of the European election studies in achieving functional equivalence of the instruments (questions) and ensuring uniform quality standards in sampling and interviewing are even more formidable in the new eastern European democracies. Apart from the technical aspects of conducting valid and reliable surveys, there may be inherent limits to exporting the survey method (Bulmer 1998). Rather than relying solely on survey data and taking them at face value, analysts must employ a more comprehensive approach that integrates quantitative and qualitative research, such as the one used by White et al. (1997) in their study of Russian voters. In the absence of a stable party structure and stable political institutions more generally and with the relative novelty of voting in free elections, it is unrealistic to expect the Michigan model or any other ”reductionist” model to provide an adequate description of voting behavior in eastern Europe.
Beyond Europe, it is doubtful whether the concept of voting is functionally equivalent to voting in the United States or western Europe because of systemic differences and traditions. For example, even if one accepts the premise that Japan is a pluralist democracy (Richardson 1997), an intricate net of mutual obligations governs much social behavior, including voting, and for many years relevant electoral competition occurred only within one dominant party. Still, the Michigan model has guided much Western research on Japanese voting behavior (Flanagan et al. 1991).
OUTLOOK
Compared to the United States, research on voting behavior in western Europe has been tied more closely to the study of mass political behavior in general, satisfaction with democracy, the parties, the politicians, and, the stability of the political system despite a continuing orientation toward the Michigan model and its variants. What European research has to offer are not better micro models of voting behavior but detailed trend analyses of cross-national comparative data that put voting into a broader context of political behavior (e.g., Klingemann and Fuchs 1995). As a result of the systemic differences within Europe and the methodological limitations of the database, no unified theory of voting behavior has emerged. For each individual (national) election, there is a plausible explanation, at least in retrospect, drawing on well-recognized factors such as perceived economic competence and leadership image, but the relative weight of each factor varies from one election to the next. As information and communication behavior is undergoing drastic changes as a result of technological advances (Internet) and the ever-increasing presence of the media, it will become even more difficult to determine the relative weight of each factor in a voter’s choice. The Columbia studies dealt with voters in a relatively contained world in which the amount of information and the number of transmission channels were limited; voters in the twenty-first century will be faced with an overload of information and will be as closely connected to the political world (or its competing representations) as they want to be. At best, more knowlegeable voters will be able to make better informed choices; at worst, more perplexed voters will succumb to the most skillful public relations managers. Somehow voters will have to reduce this complexity, and turning to social groups for guidance is a likely strategy. While it will be important to continue to study the processes of cognition and information processing, the real key to voters’ ”rationality” may lie in their social relations. More than ever, the study of voting behavior will have to build bridges between the various disciplines and incorporate the sociological perspective.
