Author: Matthew Sweeten

Publisher:

Published: 2018

Total Pages: 202

ISBN-13:

"The following dissertation advances methodological aspects of American bureaucratic and judicial politics. Each of the following three papers focus on developing methodological approaches to those subject areas specifically with respect to Natural Language Processing (NLP) and text-as-data. The main contributions of each paper are methodological in nature. First, I create a new text-based preference scaling algorithm for estimating United States Supreme Court preferences using text and network analysis. Despite widespread acceptance of text-based methods for scaling the ideological preferences of political actors, only two text-based models find widespread use: Wordfish and Wordscores. Using network-derived features to identify multiple Supreme Court issue areas and the text of Supreme Court opinions, I develop a new model for estimating the latent ideological preferences, of Supreme Court justices improving and extending Wordfish. This new model integrates new advances in network analysis and community detection algorithms to introduce multi-dimensional preferences into one comprehensive model for Supreme Court decisions. For Freedom of Speech, Search and Seizure, and Commerce Clause cases, I compare the predictive efficacy of estimates derived by the commonly used Wordfish algorithm and my own for Supreme Court cases between 1969-2003. This methods presents a way to estimate preferences from opinions even from cases on which votes are unanimous. Second, I identify key structural and ideological motivations on the U.S. National Labor Relations Board. Bureaucrat and agency preferences, and the policy choices that bureaus make, are crucial for myriad theoretical models and empirical analyses. While measuring such preferences has proven problematic, recently a number of scholars have provided alternatives for estimating these features. However, efforts have generally not included advances in text-based analysis, despite the potential for employing written records to develop rich, dynamic, estimates. To remedy this omission, we utilize and build upon text-based scaling methods to estimate the ideological and partisan preferences of key decision-makers and case decisions for a canonical agency long-studied by policy scholars: the National Labor Relations Board. Specifically, we create a comprehensive, text-based, database of Board decisions from 1947-2015. Using text analysis, we find that member preferences must be separated by case type, i.e., assuming that there we can combine such decisions in generating preference scores is not warranted. We also show that our preferences measures do well compared to alternatives, such as those solely using Item Response Theory, when we estimate models with preferences and forces internal and external to the agency to explain outputs. Finally, we find structural effects on the agency's decision content over time using change-point analysis. Finally, I leverage tools from Natural Language Processing to create an ideological common space between the U.S. Congress, Supreme Court, and Appeals Courts. Comparable ideological preferences across United States Federal institutions are fundamental to understanding the interplay of political preferences and power across those institutions. Two decades of work has encouraged the creation of ideological scores in the bureaucracy, Congress, and the Supreme Court. These scores have largely either been focused narrowly on easily observable and quantifiable institutional behavior such as roll-call votes or focusing on observable behavior that relates to the private political preferences of judges, Supreme Court justices, and members of Congress. To facilitate the creation of a legal common space, I employ textbased information retrieval and sentimental analysis techniques to create a new set of bridge observations between members of Congress, the Supreme Court, and Federal Court of Appeals and District Court judges. Utilizing this new set of bridge observations, I develop a new legal common space using sentiment of legal and judicial citations as votes to estimate the political ideal points of the entire U.S. Federal Judiciary and members of Congress from 1980 to 2012. These scores are then validated by examining the spatial arrangement of these scores relative to political party and comparing to more recent attempts at creating common space scores."--Pages viii-x.