03/16/2026
By Karen Mullins

The School of Criminology and Justice Studies is proud to announce a Dissertation Defense by Katelyn R. Smith.

Title: Irregular Peacebuilding: Security Force Conduct and Collective Efficacy Against Crime and Conflict in Mexico

Date: Thursday, April 2, 2026
Time: 10 a.m. to noon
Location: Room 431 HSSB and via Zoom

Committee:

  • Arie Perliger, Chair, School of Criminology and Justice Studies, UML
  • Chris Linebarger, School of Criminology and Justice Studies, UML
  • Sheldon Zhang, School of Criminology and Justice Studies, UML
  • Angélica Durán-Martínez, Director of Global Studies, Department of Political Science, UML

Abstract:

Irregular opponents– including criminals, insurgents, and terrorists– pose persistent threats to public safety, stability, and national security across long-term conflicts. This dissertation examines whether population-centric security force conduct, relative to repressive conduct, reduces crime and violence by improving civil-security relations and expanding community resource access. Drawing on Collective Efficacy theories, the study incorporates socioeconomic and cultural control variables to assess the community-level resources available to security forces. To operationalize security force conduct, a custom large language model (LLM) was developed in RStudio to score behavior across state and nonstate actors at the tactical, operational, and strategic levels using qualitatively coded libraries and validated through factor analysis, Latent Dirichlet Allocation (LDA), and BERTopic Modeling.

A Foreign Language Area Study (FLAS) and Ethnographic Content Analysis (ECA) grounded data collection in Chiapas, Mexico. The resulting conduct scores were integrated into a panel dataset covering 215 municipalities across Guerrero, Chiapas, and Mexico City from 2005 to 2024 (N = 4,300 municipality-years) which contains ten categories of security forces. Fixed-effects Poisson Pseudo-Maximum Likelihood (PPML) and OLS estimations with spatial lags tracked both municipal spillover and time-lag effects, with bootstrapped standard errors to account for the generated regressor. Contrary to expectations, population-centric conduct did not significantly reduce violence or crime. Instead, kinetic operations proved the strongest predictor of lethal violence, while structural variables– Market Integrity and State Capacity– outperformed conduct measures as predictors of security and public safety outcomes.