The Bandwagon Effect in an Online Voting Experiment With Real Political Organizations

Mike Farjam, The Bandwagon Effect in an Online Voting Experiment With Real Political Organizations, International Journal of Public Opinion Research, Volume 33, Issue 2, Summer 2021, Pages 412–421, https://doi.org/10.1093/ijpor/edaa008

Since the 1980s, the number of political polls has increased 10-fold, and starting in the last decade, an increasing number of Internet polls have come to be based on convenience samples, no longer even maintaining the appearance of representativeness ( Hillygus, 2011). Today, policymakers are increasingly aware of the dangers to democracy that arise from the spread of disinformation over social media and radicalization within echo chambers ( Bastos, Mercea, & Baronchelli, 2018; Persily, 2017). In this context, a comprehensive understanding of the effects of polls on voters is crucial to avoid extreme polarization within the electorate and to protect against manipulation of public opinion.

Social scientists have long debated how polls affect voters, with many expecting a bandwagon effect ( Hardmeier, 2008; Simon, 1954), according to which popular options become more popular because of a poll. Others expect an underdog effect, whereby options that were previously unpopular attract more votes ( Cloutier, Nadeau, & Guay, 1989). It remains unclear whether changes in vote patterns are the result of (de)mobilization of certain voter groups or of a real shift in opinions ( Barnfield, 2019; Chung, Heo, & Moon, 2018). Some scholars also link shifts in votes to theories on conformity, whereby a poll signals a norm to which individuals conform ( Glynn & McLeod, 1984). Still others explain the shift through voters’ desire to be part of the “winning team” ( Barnfield, 2019). The goal of this paper is not to explain the mechanism behind shifts in opinions, but rather to study such shifts in a controlled, but realistic environment.

The purpose of this study is twofold: First, the study examines precisely how poll results affect voters in an online experiment. To do so, we vary the voting system, control for demographics and political attitudes of individuals and observe how these factors interact with the effect of the poll. Second, the study adopts an experimental design in which votes have real political consequences and which closely mimics real political polling and voting. While experimental research is necessary to disentangle the complex interrelationships between different factors affecting political voting, experiments typically abstract away much of the context in which voting usually occurs ( Baghdasaryan, Iannantuoni, & Maggian, 2019). Meanwhile, field experiments and survey-based studies—which study voting in more realistic, but less controlled, settings—generally measure the effect of polls on opinions, not actual votes ( Arnesen, Johannesson, Linde, & Dahlberg, 2018; Van der Meer, Hakhverdian, & Aaldering, 2015) or, because of the secrecy of the ballot, rely on self-reporting or intentions to vote ( Mehrabian, 1998; Sonck & Loosveldt, 2010). These measures are known to, at best, only loosely correspond to actual voting behavior ( Farjam, Nikolaychuk, & Bravo, 2019; Gabel, 2000; Nordin, 2014) and, at worst, contain systematic biases toward social desirability ( Selb & Munzert, 2013; Van de Mortel, 2008). In short, there is clearly a need for an experimental design that offers a better compromise between realism and experimental control in the study of political voting.

The experimental design of this study is built around typical features of Western representative democracies, namely representation through political parties. While voters in such democracies delegate decision/legislative power to parties or candidates ( Dovi, 2018), in this experiment, voters allocate money (as a proxy for power) to politically active organizations (as proxies for parties). In the experiment, most organizations for which the U.S.-based participants could vote are well known and actively lobby in American politics. These organizations often publicly support candidates or parties during political campaigns. For example, the National Rifle Association often publicly and financially supports candidates from the Republican Party, while organizations such as Greenpeace and teacher unions are often associated with the Democratic Party ( Grossmann & Dominguez, 2009). These organizations were chosen to ensure that political views within the real political landscape carry over to the experiment, that is, to ensure that the study examines voting in the context of interest. Furthermore, the experiment offered options to participants which were less clearly linked to established parties, generally accepted for their contributions to society, and thus considered less politicized or polarizing. Examples of such organizations are those which support single parents or veterans.

The lack of realism in experiments on political voting is often criticized ( Morton & Williams, 2011), and may explain the varying success in reproducing real-world phenomena such as the bandwagon effect. Arnesen et al. (2018, p. 736) recently found that there is “little reason to worry that polls are disruptive to public debate and deliberative processes.” In the researchers’ online experiment, participants received continuously updated feedback about the popularity of opinions among previous participants. They ultimately found no voting shift, either from the beginning or over time, in response to the feedback regarding the opinions of others. A fundamental problem for this type of experiments is that stated opinions often deviate from real voting, especially when topics are highly politicized and answering in a socially desirable way comes without costs. In contrast, this experiment contributes to the existing literature by studying not only shifts in opinions, but also shifts in real votes with real consequences.

So-called “natural experiments” during real elections, although clearly better in terms of external validity than laboratory experiments, are problematic when it comes to the identification of causal relationships ( Dunning, 2012, Chapter 1). For example, Morton, Muller, Page, and Torgler (2015) compared vote results before and after a legislative change. This change made it impossible for voters in French overseas territories to see the results from the national vote that had already ended in mainland France. In their study, they argued that the legislative change also affected voter turnout and it was therefore unclear whether the bandwagon effect was the result of a (de)mobilization of certain voter groups or whether seeing the mainland results led to different voting preferences in overseas voters. The research design present in this paper offers a compromise between studying political voting in the context in which it naturally occurs on the one hand, and studying voting under highly controlled experimental conditions on the other hand.

Unlike previous experiments, we find clear and unequivocal evidence of a bandwagon effect. After participants saw pre-election polls, majority options on average received an additional 7% of the votes. This effect did not depend on the electoral system, political issue at hand, or political attitude of participants.

Method

All 1,113 participants in this study were recruited via Amazon Mechanical Turk (MTurk), and the experiment was programed using oTree ( Chen, Schonger, & Wickens, 2016). Only U.S. residents were allowed to participate (identified via their IP-address), with a minimum age of 18. All data were collected on November 15 and 16, 2018. Participants were compensated with 2$, and the average duration of the experiment was 12 min. To check for participants’ understanding of the instructions and to ensure that software bots did not participate, the participants had to correctly answer at least four out of five comprehension questions.

The experiment consisted of six treatments based on whether poll results were shown to participants or not and on the electoral system (majority representation, full representation, representation with an electoral threshold). The participants were randomly assigned to groups of approximately 190 participants, with each group corresponding to one of the treatments. Participants learned via the instructions that they would vote on four separate political issues and, for each issue, had to choose among three different organizations. Not voting for any of the options was only possible by leaving the experiment and not receiving any reward (which only two participants did). The political issues were abortion/families, environmental issues, firearms, and immigration. For example, in the case of environmental issues, the organizations presented to the participants were Greenpeace (left-wing, accepts climate change), the National Wildlife Federation (neutral, takes care of national parks in the United States), and The Heartland Institute (right-wing, denies climate change). The political issues and the options for each issue were presented in a random order for each participant. Table 1 presents an overview of all organizations.

Overview of the Organizations That Participants had to Vote on and Their Rough Political Orientation Within U.S. Politics