In my dissertation, I study how the adoption of ranked-choice voting (RCV) – an increasingly popular type of election reform in the U.S. – affects racial minority representation. More specifically, I address the following substantive questions:

1. Does switching from first-past-the-post (FPTP) to RCV induce moderation in ethnic party/candidate competition and racially polarized voting?
2. How does strategic ballot concentration by minority voters impact voters’ aggregate-level preferences in RCV?
3. How do heterogeneous ballot order effects by race and ethnicity shape minority representation under RCV?

Despite the theoretical and practical importance of these questions, political scientists have not fully understood the mechanisms through which adopting RCV affects minority representation due to the lack of rigorous formal theories, appropriate data to test hypotheses drawn from such theories, and tailored statistical methods for analyzing data from RCV elections. To overcome these challenges and to answer the above questions, I propose formal models for ethnic multiparty competition along with their algorithmic solutions, develop a set of novel statistical methods for analyzing and modeling rank data, and collect novel data on FPTP and RCV elections in the U.S. My dissertation is financially supported by the Electoral Reform Research Group (New America) and is composed of three essays. Below, I describe the abstract for each of the three essays.

Does Ranked-Choice Voting Reduce Racially Polarized Voting?” (with Theo Landsman)
Does adopting ranked-choice voting (RCV) induce moderation in racial and ethnic party competition? Despite the importance of this question, which lies at the intersection of two vast literatures on electoral systems and party competition, previous research has only offered insufficient explanations as to why switching from first-past-the-post (FPTP) to RCV may reduce racial and ethnic conflicts, if any, and reported empirical findings with mixed conclusions. This article offers a neo-Downsian spatial model of ethnic party competition in multiparty systems under FPTP and RCV. Unlike existing spatial models, our model accounts for multidimensional policy space, multiple ethnic groups, and moderate and extreme parties within each ethnic group, and different party strategies in the two electoral systems. To solve for the distribution of Nash equilibria and estimate the degree of ethnic party moderation in FPTP and RCV, we offer a set of novel algorithms by drawing from the literature on agent-based modeling. Moreover, we provide evidence for several observable implications of our theory by using newly collected data and clustering methods for rank data. By applying cluster analysis and ecological inference to more than 5.5 million ranked preferences and 13,635 precinct-level data from Bay Area mayoral elections from 1990 to 2020, we quantify the degree of racially polarized voting in a manner that is comparable across electoral systems. We show that switching from first-past-the-post to RCV election does not seem to reduce the level of racially polarized voting for the overall electorate and all pairs of racial and ethnic groups examined. Finally, we discuss several implications of our research for minority representation, which not only contribute to the ongoing scholarly debate but also offer practical insights for election reform in the U.S.

Statistical Methods for Partially Ranked Ballot Data
Partially ranked data, in which some items remain unranked, are common in many rank elicitation processes, including ranked-choice voting (RCV) elections. Existing methods for partially ranked data, however, do not properly account for a set of underlying institutional and behavioral reasons for why voters offer partially ranked ballots in elections, leaving researchers a non-optimal option to use existing tools to analyze ranked ballot data while making unreasonable assumptions. To remedy this problem, I introduce a novel framework for analyzing and modeling partially ranked data collected in both single and multi-winner RCV elections. Unlike existing methods, the proposed model first decomposes observed partial rankings into (1) structural partial rankings controlled by election law (voters are only allowed to rank top-K candidates); (2) incomplete rankings caused by random errors (voters fail to rank lower-choice candidates); and (3) strategic ballot concentration also known as plumping (voters intentionally choose only a single or two candidates) and model observed partial rankings as a mixture of the three types of data. Moreover, I develop efficient estimation and inferential methods for rank aggregation under the proposed model and investigate their finite-sample properties in simulation studies. To illustrate the proposed method, I analyze individual-level ranked ballot data from more than 100 RCV elections in the U.S. Finally, I examine several implications of strategic ballot concentration in RCV for minority representation.

Causal Inference with Rankings as Generalized Discrete Outcomes
Rankings lie at the center of a multitude of research in social sciences over the past century. More recently, researchers use rankings as outcome variables in experimental studies to gauge the causal effects of treatments on rankings as generalized discrete outcomes. Despite the promise of such studies both in the substantive and methodological literature, very little is known about how to analyze ranked outcomes in experimental studies. In this article, I propose a novel framework to perform causal inferences when the outcome variables of interest consist of rankings. Given the structured and high-dimensional nature of rank data, I nonparametrically identify three estimands based on the well-known ideas of average rankings, pairwise rankings, and rank correlations. These quantities of interest are not only intuitive but also degenerate into a more conventional average treatment effect for binary outcomes. Additionally, I develop nonparametric sharp bounds for partial identification, offer simple estimators for point identification, propose inferential methods for constructing confidence intervals and performing multiple hypothesis testing, and clarify formal connections between the proposed method and the Plackett-Luce model. I demonstrate the proposed methodology by reanalyzing experimental data from a recent study that addresses the effect of external information on people’s responses to police violence. Finally, I demonstrate how the proposed method can be used to gauge the heterogeneous ballot order effects in ranked-choice voting elections and discuss their implications for minority representation.