Opinion That Matters: Bringing Behavior Back into the Study of Public Opinion

• In Event: Methodological Innovations in the Study of Public Opinion

Abstract

What people do is more politically consequential than what they say to pollsters. For a variety of historical reasons, however, the concept of public opinion since the 1930s has come to focus narrowly on individualized attitudes elicited through surveys rather than also considering collective behaviors that appear on the streets as protests, demonstrations, and acts of political violence. Declining response rates in phone surveys and the uncertain validity of online panel data have raised pressing questions about the current viability of conceptualizing public opinion as the results of opinion surveys (e.g., Kennedy and Hartig 2019; Leeper 2019). At the same time, new forms of digital trace data (e.g., Guber 2021) as well as new methods for monitoring political behaviors at national and international scale (e.g., Nardulli et al. 2015) invite a reconsideration of what counts as public opinion for purposes of academic research.

In response to these challenges, we revisit an older and more theoretically compelling paradigm for thinking about public opinion that until recently had been seen in many corners of the field as empirically unfeasible: tracking public opinion in the form of manifest actions rather than as latent attitudes. Long before sample surveys were seen as legitimate expressions of public opinion, the “classical tradition” in public opinion research held that public opinion is necessarily expressed in public and organized around groups rather than atomized individuals (e.g., Blumer 1948; Lazarsfeld 1957). The tendency for survey researchers to aggregate with equal weight the individual opinions that sampled individuals expressed in private was seen by many scholars in that classical tradition as misrepresenting a complex and organic process of interactions and deliberation among different groups and individuals which produced the kind of public opinion that “mattered” politically. But by the 1950s, these older sociological assumptions about the nature of public opinion had given way to the new opportunities for systematic national research afforded by random sample surveys, and today this debate has largely been forgotten. It is now rarely questioned among academic researchers conducting quantitative survey research that public opinion is a psychological rather than sociological phenomenon, a property of the mass public rather than organized groups, and something that is expressed privately to pollsters rather than in public by sometimes boisterous groups.

Our paper leverages methodological innovations from computational social science that now allow researchers to rapidly extract data about behavioral events from news reports and other texts as a way to potentially reinvigorate the older sociological paradigm in public opinion. Research using behavioral public opinion data taking the form of protests, demonstrations, and acts of political violence is already being done in sociology, international relations, and comparative politics. There are many available sources of behavioral event data, including the the Social, Political, and Economic Event Database (SPEED) project (Nardulli, Althaus, and Hayes 2015) and the Armed Conflict Location and Event Data Project (ACLED) (Raleigh, et al., 2010). Still, event data remains largely unknown and unexplored in the public opinion literature.

To showcase the potential utility and significance of this behavioral conceptualization of public opinion, we compare opinion dynamics measured over a 60-year period with both opinion surveys and political event data. Our measure of surveyed opinion from 1945 to 2005 comes from 686 surveys that ask some version of the most important problem question (Heffington et al. 2017). Our measure of behavioral opinion over the same period comes from the SPEED global random sample dataset, which contains roughly 10,000 instances of mass political expression or political violence events in the United States that occurred between 1945 and 2005 (Nardulli, Althaus, and Hayes 2015). SPEED event data includes a broad topic origin for each event (e.g., social-cultural animosities, anti-government sentiment, etc.) and these topical categories are aligned to those found in the most important problem data. To illustrate the usefulness of behavioral events as a form of opinion data, we explore whether dynamic changes in surveyed perceptions of most important problems corresponded to dynamic changes in political events associated with the same topics over this six-decade period. Preliminary results indicate only a weak relationship between the topical distributions of surveyed most important problems and the topical distributions of behavioral events. This finding raises questions about the political relevance of conventional forms of opinion data that mass surveys typically capture, and also underscores new opportunities for opinion researchers to leverage novel methods for bringing political behaviors back into the study of public opinion.

Authors

• Scott L. Althaus, University of Illinois at Urbana-Champaign

• Zachary Jablow, University of Illinois at Urbana-Champaign