Voter Confidence and Electoral Participation

The MIT Election Data and Science Lab helps highlight new research and interesting ideas in election science, including through research grants under our ongoing Learning from Elections program.

Our post today was written by Thomas Cao, Susan Athey, and Herman Donner, based on their ongoing research funded by this program. The information and opinions expressed in this column represent their own research, and do not necessarily represent the opinions of the MIT Election Lab or MIT.

Mistrust in electoral outcomes has become an increasingly salient problem in the midst of growing affective polarization among the American public. Unsubstantiated allegations of election fraud have the potential to undermine voter confidence in U.S. election processes, which in turn may impede democratic participation through a negative impact on voter turnout.

However, little existing research has sufficiently explored the causal links between partisan sentiments, voter confidence, and electoral participation. Our research project seeks to address these questions in a large-scale experimental setting over the 2022 midterm elections. In particular, we evaluate 1) whether information on the bipartisan oversight of the electoral process increases voter confidence in election outcomes; and 2) whether this increase in voter confidence leads to additional political participation in the electoral processes, as reflected by voter turnout tendency. 

We reached out to a random sample of registered U.S. voters (each linked to their voter registration record through L2’s voter ID) during the two weeks before the November 8, 2023, mid-term election. We managed to recruit over 13,000 respondents to fill in our online survey. Half of the respondents are randomly assigned to our treatment information on the election process’s bipartisan oversight embedded in the survey, which focuses on alleviating concerns that one party can single-handedly affect substantial decision-making. We have worked with election officials across the U.S. to develop our treatment message, so as to make sure that the wording is not only factually correct but can actually be adopted in future election communications.

Before the treatment message is shown, we ask the treatment group a question on who is involved in election-related decisions (choices including, e.g., only Democrats/Republicans, only the party controlling one's state's executive branch, etc.). Respondents are told that answering the quiz question correctly will provide a chance to win a $20 Amazon gift card, which provides an incentive for them to process the question and the underlying treatment content carefully. The control group is only shown generic information on the 2022 midterm elections’ scope. The post-treatment outcome questions are respondents’ confidence levels in electoral outcomes for their own state, the entire country, and red/blue/swing states, and their self-reported voting tendency, each on a 5-point scale, with 5 representing that one fully trusts the election outcome or that one will definitely vote. 

Our results demonstrate significant and substantial differences in voter confidence, with an over 5 percentage point increase (p < 0.001) in respondents who fully trust their state-level electoral outcomes in the treatment group in comparison to the control group. Similar significant effects are observed for increasing voter confidence in nationwide election outcomes and outcomes in red, blue, and swing states. The differences remain significant after accounting for differential attrition in the treatment group with the nonparametric Lee (2009) bounds. Moreover, in terms of self-reported voting tendency, the treatment group sees an over 4 percentage point difference (p < 0.001) in respondents who said they would definitely vote in the 2022 midterms between treatment and control.

As a sanity check for potential experimenter demand effects, after the treatment/control information, we first ask respondents whether they voted in 2018 and 2020 before showing them the other post-treatment questions. There is no significant difference in the outcomes of these questions, or the proportion of respondents lying about them, between the treatment and control groups. Besides, respondents who answered that they would definitely vote were further asked whether they would vote early or on Election Day and whether in person or by mail, with the option of “I haven’t decided” available for both questions. There is no significant difference between treatment and control in respondents selecting “I haven’t decided.” 

One month after the midterm elections, we sent out a follow-up survey to the respondents who participated in our previous survey. Among them, approximately 1/3 responded again. For those assigned to control in the previous survey, we randomly assign half to the previous treatment information before asking the same outcome questions; the other half are directly asked the outcome questions. For those previously assigned to treatment, we ask the same outcome questions, followed by a question testing whether they still remembered the treatment message. Besides, regardless of their previous treatment status, a random half are shown a notice on the outcomes of the midterm elections at the beginning of the survey.  

Among the respondents previously assigned to control, we recover similar significant treatment effects on confidence in election outcomes between those newly assigned to treatment and control. Moreover, in comparison to respondents who were twice assigned to control, those previously assigned to treatment (and therefore saw the treatment message over a month before) still have significantly higher confidence levels in election outcomes after controlling for demographic and pre-treatment covariates. The difference comes exclusively from the approximately 1/4 of the follow-up respondents previously assigned to treatment who could still remember the treatment message’s theme (bipartisan oversight). Notice of the election outcomes does not interfere with our treatment’s effects.

Overall, the results suggest that our simple, easy-to-implement treatment on electoral bipartisan oversight significantly and substantially increases voter confidence and turnout tendency in the 2022 midterm election outcomes, and the effects on voter trust in electoral outcomes are sufficiently persistent after the election results are announced. This alone has major policy implications for election officials. Next, we will conduct follow-up experiments to better understand the mechanisms of our treatment effects, and to analyze the effects of our treatment on actual voter turnout (and their heterogeneity), after the release of the latest voter files documenting voting behavior for the 2022 midterms.