Field Experiments with Robo Calls: The Effect of Treatment Dosage and Social Pressure Messaging on Voter Participation

Daniel Kling

Major Professor: Thomas Stratmann, PhD, Department of Economics

Committee Members: Tim Groseclose, Cesar Martinelli

Carow Hall, Conference Room
August 20, 2018, 10:00 AM to 12:00 PM


In this dissertation, I use data from two field experiments with automated get out the vote (GOTV) telephone calls, or robo calls, to examine behavioral questions related to campaigns and voter mobilization. In Chapters 1 and 2, I use data from a 2014 field experiment to examine how treatment dosage and treatment compliance affect voter participation. In Chapter 3, I use data from a 2016 field experiment to evaluate the effect of different types of social pressure messages on voter participation. These chapters are co-authored with Thomas Stratmann.

 The first chapter, “The Efficacy of Political Advertising: A Voter Participation Field Experiment with Multiple Automated Calls and Controls for Selection Effects,” explores why political campaigns spend money on campaign tools that haven’t been shown to be effective? We document the effectiveness of robo calls for increasing voter participation in contrast to most published research which finds little or no effect from automated calls. We establish this finding in a large field experiment which mimics campaign behavior with a targeted, partisan GOTV campaign. Our experimental design includes a follow-up call, which allows us to control for selection effects. We identify subsets of subjects, for whom the treatment effects are substantially larger than those that are found in previous studies. Our findings show that automated calls can cause up to a one percentage point increase in voter turnout. Additionally, our experimental design allows for testing how the number of calls in a treatment, that is dosage, affects voter turnout. Here, results show that that a few extra calls increase the treatment effect, and that many additional calls decrease that effect.

 The second chapter, “Repeated Treatment in a GOTV Field Experiment: Distinguishing between Intensive and Extensive Margin Effects,” uses the same experimental data to further analyze the effect of repeated treatment. Technological advancements and the increasingly large amounts spent for elections have made it easier for GOTV campaigns to contact potential voters many times prior to elections. However, the effects of repeated exposure to GOTV treatment have received little attention in the academic literature. Multiple treatments can increase the intensive margin of treatment – by increasing the number of times that treated individuals are exposed to the treatment – or the extensive margin of treatment – by increasing the number of people who are exposed to at least one treatment attempt. We analyze the differences between a treatment with one automated GOTV call and a treatment with three GOTV calls to isolate the effect of changes in the extensive and intensive margins of treatment on voting rates. On the intensive margin, we find that answering one additional call is associated with an approximately 0.6 percentage point increased probability to vote. On the extensive margin, we find that answering one additional call is associated with an approximately one percentage point increased probability to vote. This evidence is consistent with the academic literature on commercial advertising, which finds that repeated exposures to commercial advertisements have diminishing marginal effects.

The third chapter, “Observability Without Observation: The Effect of Impersonal Social Pressure Messages in a GOTV Experiment,” investigates the effect of different types of social pressure messaging. Previous studies have shown that social pressure messaging can be used to induce pro-social behavior in a variety of settings, including voter participation. We use a field experiment to show that a generic observability message increases voter turnout above and beyond the effect of a GOTV call without the observability language. This contrasts to previous studies which used messages with personalized voter histories or the appearance of personalized voter history to demonstrate the public nature of voter participation. We also show that including descriptive norm language, which claims that many other people will also be voting, does not significantly increase voter turnout.