A Caltech Library Service

Computational Methods in the Study of Political Behavior


Kann, Claudia Kenyon (2023) Computational Methods in the Study of Political Behavior. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/0zxf-6b07.


In this thesis, I explore how individual-level actions contribute to aggregate political outcomes. In each chapter, I aim to understand an observed political behavior using data or methodologies previously unused in their contexts. The subject matter ranges from protest activity and vote choice to theoretical opinion models and re-examining how socioeconomic class is understood in quantitative work.

In the first two chapters I employ novel datasets to understand phenomena where popular theories differ from empirical observations. In Chapter 1 I examine protest behavior, which is not the equilibrium prediction of models of collective action. I investigate what aspects of published language can predict protest participation and how these change leading up to and following protests. Specifically, I collect and, using natural language processing methods, analyze 4 million tweets of individuals who participated in the Black Lives Matter protests during the summer of 2020. Using geographical and temporal variation to isolate results, I find evidence that interest in the subject, measured as percentage of online time discussing the matter, is correlated with protest behavior. However, I also find that collective identity, measured through pronoun use, does not have a strong relationship with protest behavior.

Next, in Chapter 2, I use a survey---which I helped to develop and field---to understand the 2020 midterm elections' surprising results. While most accepted models of midterm elections predicted massive Democratic losses (averaging around 40 seats in the House), these predictions were not met. In fact, the Democratic party did well---they did not lose a single state legislature, expanded some majorities, and lost only 9 seats in the House of Representatives. Testing various models of midterm elections, I show that the 2020 midterms were issue-based elections, where views on abortion had a large impact on vote choice.

In the second half of the thesis I focus on methodologies. Specifically, in Chapter 3, I expanded on mathematical models of consensus building to better mimic reality. Bounded confidence models have historically been used to explain convergence of opinions. In this chapter I add a repulsive element, modeling the inclination to differentiate oneself from someone who otherwise has similar beliefs. With this added component, convergence is no longer assumed. I explore both analytical and simulated numerical results to understand the dynamics of opinions in this new context.

Finally, in Chapter 4, I introduce a method for operationalizing socioeconomic class as a latent variable in regression models. While there has been a plethora of research which shows that class affects opinions, views, and actions, the definition of class is nebulous. I argue that this is a result of the nature of class, which is context dependent. Therefore, rather than explicitly determining class, I present using class within a mixture model framework. This allows for the exact definition of class to change within the context being analyzed and enables researchers to use class within their work. Following the theoretical arguments, I present the efficacy of the approach using the American National Election Studies survey from 2020 to show how class differs when related to views of the U.S. Immigration and Customs Enforcement agency and the Black Lives Matter movement.

Item Type:Thesis (Dissertation (Ph.D.))
Subject Keywords:political behavior, social science, methods
Degree Grantor:California Institute of Technology
Division:Humanities and Social Sciences
Major Option:Social Science
Thesis Availability:Public (worldwide access)
Research Advisor(s):
  • Alvarez, R. Michael
Thesis Committee:
  • Katz, Jonathan N. (chair)
  • Alvarez, R. Michael
  • Hoffman, Philip T.
  • Kiewiet, D. Roderick
Defense Date:8 May 2023
Record Number:CaltechTHESIS:05172023-231943782
Persistent URL:
Kann, Claudia Kenyon0000-0002-8318-4890
Default Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:15187
Deposited By: Claudia Kann
Deposited On:22 May 2023 19:48
Last Modified:04 Jun 2024 21:37

Thesis Files

[img] PDF - Final Version
See Usage Policy.


Repository Staff Only: item control page