Why is marriage between relatives common throughout the world? Consanguineous marriage was commonplace throughout human history, but became increasingly rare in the developed world beginning in the 1800’s. For large swaths of the world from Morocco to Indonesia, it is still ubiquitous.

This project investigates the theory that women use consanguineous marriage to mitigate their exposure to Intimate Partner Violence. Marrying relatives, particularly first cousins, provides women with an ex ante information gathering mechanism and an ex post behavior enforcement mechanism, both of which serve to lower the amount of violence women are exposed to.

Observational data taken from USAID’s Demographic and Health Surveys confirms the expected result that consanguinity and Intimate Partner Violence are negatively correlated in three of the four countries studied—Pakistan, Egypt, and Jordan. In India, however, the two are positively linked. Relying on previous literature, I speculate that the reason behind this divergence lies in dowry practices in India.

This dissertation introduces a set of new statistical methods for measuring foundational constructs in political science: ideology, dimensionality and polarization. Using the proposed methods, I offer novel findings on multidimensional ideological characteristics of American Congress. The final chapter provides the first complete description of ideological coalitions in Korean National Assembly from its birth to the present by introducing a new structural model of ideal point estimation for non-voting datasets.

Most social science theories attempt to explain the ways in which social interactions affect human behavior. Inferences therefore depend upon a researchers' ability to measure and characterize these interactions. In this dissertation, we highlight several inferential challenges in measuring social interactions and show how geographical data can be used to address them. One such challenge is posed by exposure to various treatments; many treatment variables of interest are difficult to measure in a coherent and unified way. Chapter 2 illustrates the way in which public records can be used to overcome this challenge. Another challenge is post-treatment sorting. Once social conflict begins, one of the first things that groups do is segregate themselves into smaller sub -groups. Chapter 3 introduces a new instrumental variable to correct for this bias in census data. Lastly, Chapter 4 introduces a novel method for estimating where social media users spend time during their day. These findings and novel methods will contribute to social scientists’ ability to study contextual effects on behavior.

This dissertation considers the topics of prediction and variable selection for the applied political scientist, particularly in the context of high dimensional data.

In Chapter 1, we consider the puzzle of why highly significant variables aren't automatically good predictors. This problem occurs in both simple and complex data. We offer explanations and statistical insights into the puzzle. We suggest shifting the research agenda toward searching for a criterion to locate highly predictive variables rather than highly significant variables. We offer an alternative approach, the Partition Retention (PR) method, which was effective in reducing prediction error from 30% to 8% on a long studied breast cancer data set.

In Chapter 2, we propose approaching prediction from a framework grounded in finding the correct prediction rates of variables. While intuitively obvious, not nearly enough attention has been paid to creating a clear theoretical framework for prediction. We present an objective function for prediction rates and consider but ultimately reject an estimator based on the sample analog of the solution due to its inability to distinguish predictive variables from noisy ones, which leads to an inability to estimate it. We offer an alternative solution and demonstrate that the PR's I-score asymptotically approaches this alternative solution. The I-score for a variable set can be written as an asymptotic lower bound for the correct prediction rate. We offer simulations and applications of the I-score on real data.

In Chapter 3, I propose a new approach to predicting civil war onsets that emphasizes variable selection. A good variable selection approach should search for variables based on a criterion of predictivity and find variable interactions. I suggest the PR method to conduct variable selection and illustrate with simulations and an application to civil wars data, comparing results with alternative approaches. The PR identifies variable sets, some as large as 5 or 6 variables, to predict war onsets. Using these variable sets to predict boosts correct prediction rates on out of sample data from 78.98% to 98.05%. The application demonstrates gains in prediction rates for political phenomena like civil wars when including a research step for variable selection.

Many prominent theories in political science, and social science generally, argue that individuals are imprinted with behaviors and identities over the course of their lives. According to these theories, behaviors and identities persist due to internal, psychological consistency. Here, I argue that persistence in behavior, identity, and social interaction is often maintained by social networks, rather than individuals. Many day-to-day behaviors and identities would be constantly redefined and updated if there were reasons to do so. In other words, imprinting is often highly context specific, consistency in individuals is a property of stable social structures, and individuals can change very quickly when social networks around them allow or encourage it. The first paper in this dissertation ("Widowhood Effects in Voter Participation") studies why people are less likely to vote after their spouse dies. I show that a rapid, permanent drop in turnout is mostly not due to the trauma of the loss or a slowly declining interest in politics. Widows and widowers vote less because their spouse motivated them to vote. The second paper ("Partisan Attachment or Life Stability?") argues that partisan affiliations are stable in the United States electorate because day-to-day lives do not change very much. When lives do change, people reconsider their partisanship and are more likely to change it. A central argument in this paper is that age is associated with increasing partisan stability only because life stability increases with age. The last paper ("Plasticity in Human Social Networks") characterizes social network adaptations after the death of a friend. I show that mutual friends become permanently closer to each other after the shared loss, recovering the same volume of interactions that was lost from the death. Younger people tend to contribute more to recovery than older, but older people contribute equally when a loss was sudden and unexpected. The observed recovery occurs in distinct, overlapping patterns and is mathematically similar to shock responses in small-scale biological networks.

Measurement is foundational to political science research. Theories are only testable to theextent that their abstract concepts can be connected to empirical reality. Political science is a field where many important questions deal with concepts whose measurement is not immediately obvious. Does democracy reduce corruption in a country? Does ideological extremism impact the electoral success of politicians? To answer these questions, valid measurement of the key variables is an essential first step. In this dissertation I propose three projects which help improve measurement in political science. The first is a new method for measuring race/ethnicity when this data is missing. My model uses Bayes’ theorem to predict the posterior probability that an individual identifies with a particular race or ethnicity, given other known attributes. I validate these predictions against voter registration data, and I show that my model is far more accurate compared to previous methods. I also develop an R package to provide easy implementation of my method. The second project is a new model for measuring the political ideology of actors under extreme missing data conditions. Models of political ideology usually use observed actions, such as taking positions on legislation, to infer an actor’s latent ideology. But I show that in contexts where most actors fail to take explicit positions on most pieces of legislation, the measurements from traditional models can quickly degrade. The model I develop directly accounts for these missing signals, thereby generating more accurate measurements of political ideology. I apply the model to data on federal interest group lobbying. The final project is a method for incorporating measurement model uncertainty into an empirical theory testing model. Estimates of ideology from the model in project #2, as well as from any other statistical measurement model, produce more than a single value of the latent variable—they also produce some measure of the error/uncertainty for the true value. I show that failing to account for this measurement error in the theory testing stage can lead to misleading or biased conclusions. The method I propose in this project fixes this source of bias.

Americans are increasingly concerned about where we learn about politics. The reality is that many Americans learn about politics from conversations with friends and family. But how is information transmitted by our peers different from information communicated by the media? More importantly, what are the consequences of relying on our peers for information about politics relative to seeking information from professional news outlets? In the face of fake news, misinformation, and polarization, this dissertation focuses on understanding the content and consequences of interpersonal political communication. Ultimately, I argue that reliance on social information leads to negative informational and attitudinal outcomes, while opening the door for increased political engagement.

Chapter 1 serves as an introduction that outlines my theory about why (and how) information becomes distorted through social transmission and why this could affect political behavior. Chapter 2 introduces a research design called a telephone game experiment that allows researchers to examine how individuals summarize information from the news and how it changes as it flows through communication networks. A subsequent version of this chapter has been published in the Journal of Politics. Chapter 3 uses the design developed in Chapter 2 to examine information transmission about economic performance and investigate the consequences of exposure to a news article or a social summary of that article. The results suggest that individuals can learn just as much from their peers as they can from a news article if they receive information from someone who is more knowledgeable than they are and is a copartisan. However, the information sources lead to divergent patterns in attitudes, such that partisan bias in the social information could influence subjective evaluations. A subsequent version of this chapter is forthcoming in the American Political Science Review.

Chapter 4 uses data from two nationally representative surveys to show that individuals who report relying on conversations with others for information about politics are more likely to believe political rumors. Chapter 5 provides a brief conclusion to the dissertation, highlighting important avenues for future research.