Quantifying Voter Turnout Disparities Using Novel Causal Machine Learning Methods

Do majority groups turn out to vote more than minority groups? Research on race inequalities in voter turnout has consistently found that racial minority groups tend to vote at lower rates than majority groups, which reflects established disparities in political engagement across different racial groups. With regards to voter turnout, a variety of studies point to variables explaining these lower rates of engagement, including socioeconomic status, policy attitudes, and systemic barriers to voting that disproportionately affect minority populations. Yet, much of this research relies on observational data or in some cases on traditional statistical methods that are ill-equipped for identifying complex causal pathways and interactions between these variables. This paper proposes to use the Causal Fairness Analysis (CFA) framework, a novel method with its foundation in causal inference and machine learning, to quantify and model the effect of race on voter turnout and determine through what mediating and confounding variables race influences voter turnout. This framework is, moreover, capable of detecting biases in both observational and experimental studies by mapping observed disparities onto underlying causal mechanisms, allowing for an analysis of disparities at a structural level. In the empirical section of the paper, we first apply the CFA framework to reproduce findings from prior studies, validating its applicability and relevance in election studies. We then apply this framework on recent panel data from the 2024 American National Election Study, to quantify both direct and indirect effects of race on voter turnout. This analysis will also include mediating variables that are new in this survey, such as democratic norms, views on candidate eligibility, opinions on conflicts in Ukraine and Gaza, reactions to shifts in the Democratic ticket, and perspectives on issue importance and ownership. The CFA framework’s ability to parse out the contributions to the total effect, aims to advance our understanding of race-based disparities in voter turnout and improve the valorization of novel causal machine learning techniques in political science.