Jennica Petersen, Rebecca Shuel, and Michael Barber, Department of Political Science
Researchers agree that partisanship is the main predictor of how any given individual will vote, but previous researchers have been unable to determine to what exact quantifiable extent party labels determine individuals’ voting choices. This is because previous research has been purely observational in nature, meaning that any inferences about how party labels affect voting behavior could have been impacted by any number of confounding variables. We isolated the independent effect of party labels on determining voters’ choices by conducting a novel survey to measure partisans’ vote choices for two hypothetical presidential candidates in simulations of both nonpartisan and partisan elections. By taking an experimental approach to isolate the effect of party labels on partisans’ vote choices, we eliminated the confounding variables that have plagued previous observational studies. We found that even when faced with the same two hypothetical candidates, individuals chose the candidate from their party at a much higher rate in the hypothetical partisan election than in the hypothetical nonpartisan election. This helps us understand that party-line voting remains rampant in our society, and partisans may rely even more heavily on party labels than on qualifications or issue positions when deciding how to vote.
We used Amazon Mechanical Turk to conduct a survey experiment of 748 adults from across the United States to compare partisans’ vote choices for two hypothetical presidential candidates in a) a simulation of a nonpartisan election to b) a simulation of a partisan election.
We gathered demographic information about each survey participant, namely party identification (based on the typical 7-point scale), race, age, gender, religion, education, ideology, and income. Participants were randomly assigned to either the control or the treatment condition; those assigned to the control condition saw only profiles of two hypothetical candidates (“Candidate A” and “Candidate B”) which detailed the candidates’ experience and a few issue positions, but participants in the treatment condition saw the candidates’ partisan affiliations in addition to their profiles. The candidate profiles that appeared in both the control and treatment conditions were identical and were modeled after the two most moderate members of the U.S. Senate; the sole difference between the two conditions was the indication of the candidates’ partisanship in the treatment condition. After reading about the two candidates, survey takers indicated which candidate they would choose to vote for if this were an actual presidential election.
Since we were interested in seeing how the indication of partisanship on the ballot affected partisans’ vote choices, we restricted our analysis to only those participants who indicated that they self-identified as a Republican or a Democrat to some degree (i.e. we only analyzed data for participants who identified as either “Strong/Weak/Independent-leaning” Democrats, and “Strong/Weak/Independent-leaning” Republicans). This left us with 583 observations, which was more than enough to draw firm conclusions about our results.
The independent variable of interest in our study was whether the individual was assigned to the treatment condition; that is, whether they knew the partisanship of the candidates (this variable was coded 0/1), and the dependent variable was whether they voted for the candidate who shared their partisanship (also coded 0/1). We ran multiple statistical tests, including several linear probability models and a logit regression, to compare the voting behavior of participants in the control condition to the behavior of those in the treatment condition. Each test gave us essentially the same result.
Even when controlling for all demographic factors included in the survey (race, age, gender, religion, education, ideology, and income), being assigned to the treatment condition led to at least a 26 percentage point increase in an individual’s likelihood of choosing to vote along party lines; this is a huge increase on the 100-point scale. In addition to the substantive significance of this number, this was the only variable that was statistically significant at predicting individuals’ vote choices (p<.01). This effect was even stronger in some cases, especially among “strong” partisans. For example, all else equal, “strong Democrats” assigned to the treatment condition were 36.3 percentage points more likely to vote for the Democratic candidate than “strong Democrats” assigned to the control condition, and 100% of “strong Republicans” assigned to the treatment condition chose the Republican candidate. Thus, even though the survey participants in both conditions saw identical candidate profiles, survey participants who clearly knew the candidates’ partisanship were considerably more likely to vote along party lines.
The results confirmed our expectations and quantified the effect of party labels on partisans’ voting behavior. Although participants in both the control and the treatment condition saw identical candidate profiles, partisans who were assigned to the treatment condition had significantly different voting behavior than those assigned to the control condition. Since the only difference between the two conditions was the indication of the candidates’ partisanship in the treatment condition, we can infer that the difference in voting behavior was purely due to the inclusion of partisanship on the ballot. It is also important to note that no other factor (race, age, gender, religion, or income) determined whether or not individuals voted along party lines; this shows that the importance of party labels cannot be understated.
Although this survey greatly simplifies the complexity of real elections, it has important implications because it quantifies the strength of the effect of party labels on partisans’ voting behavior. All else equal, the inclusion of party labels on the ballot caused individuals’ voting behavior to change by at least 26 percentage points, which shows that individuals rely very heavily on party labels.
Again, participants in both the control and the treatment condition saw identical qualifications and issue positions for the two candidates. Because the voting behavior of participants between the two conditions varied so greatly, these results may demonstrate that individuals may choose to vote for a candidate that they may not actually prefer given qualifications and issue positions simply because they share a party label. Although this study is not qualified to answer whether party-line voting is harmful or beneficial for society, it does show that party-line voting remains rampant in society.