Date of Award

Spring 2022

Degree Type

Open Access Dissertation

Degree Name

Political Science, PhD

Program

School of Social Science, Politics, and Evaluation

Advisor/Supervisor/Committee Chair

Melissa Rogers

Dissertation or Thesis Committee Member

Jacek Kugler

Dissertation or Thesis Committee Member

Mark Abdollahian

Terms of Use & License Information

Terms of Use for work posted in Scholarship@Claremont.

Rights Information

© 2022 Glenn-Iain Steinbeck

Keywords

Corruption, Corruption Index, CPI, Extreme Gradient Boosting, Machine Learning, Outcome Indicator

Subject Categories

Political Science | Public Policy

Abstract

The impact of corruption is an increasingly important and visible topic for academics, policy makers, and the public. Yet corruption is exceptionally difficult to directly observe and empirical measurements of corruption remain highly contested. Despite the increasing availability of corruption measures and generally high correlations between them, scholars and practitioners disagree over their applicability, interpretation, and the validity of their methods. With the most frequent complaint being that existing corruption indices are largely based on expert opinion surveys, and therefore potentially open to bias and differences of interpretation. Yet, while corruption itself may be ephemeral its aggregate effects are more concrete. Corruption, especially transactional corruption disproportionately harms the least advantaged, decreases access to services, imposes additional costs, and harms well-being. Consequently, the individual level impacts of corruption impose a measurable collective shadow upon society, and when considered together serve as a latent indicator of its presence.This work applies machine learning to objectively estimate corruption from this shadow, proposing both a novel theoretical approach to assessing national corruption, as well as a new quantitative corruption index. The theoretical approach outlined here begins by identifying measurable consequences, or outcomes of corruption from previous research. It then develops a Python based extreme gradient boosting machine learning model to validate selected corruption outcome measures as predictors of existing corruption indices and estimate predictive factor importance. Finally, a weighted Z-score index of comparative national corruption is produced and tested as a replacement for existing perception based measures of political corruption. The results of this analysis demonstrate that political corruption can be predicted utilizing variation in a combination of readily available socio-economic and public health data. Further, the newly defined Outcome Index of political corruption is correlated with existing perception based corruption measures, and performs comparably as a direct replacement in replication analysis. Additionally, the Outcome Index of corruption is both easier to construct than most existing corruption measures, because it obviates the need to conduct survey analysis, and in contrast to existing approaches it is based on objectively measured and readily available data.

ISBN

9798819308653

Download

Share

COinS