Scholarlink

Purpose <p>To provide guidance on the use of Bayesian mixed-effect modeling for inference in criminological applications involving complex longitudinal data, and a demonstration involving police firearm seizures and gun violence.</p> Methods <p>We outline a Bayesian strategy for specifying and fitting mixed-effects models using the brms package in R, including procedures for prior predictive simulation, model selection, and posterior comparison during sensitivity analyses.</p> Results <p>To demonstrate the utility of the Bayesian approach, we apply it to complex longitudinal data from a police directed patrol intervention in Flint Michigan, estimating the impact of gun seizures on firearm violence using a three-level mixed effects zero-inflated negative binomial regression model. The demonstration suggested a small and fragile reduction in reported gun violence associated with firearm seizures, which was not sufficiently distinguishable from placebo-based sensitivity analyses.</p> Conclusions <p>Bayesian methods are underutilized compared to frequentist approaches in criminological applications. This manuscript outlines an approach to expanding the criminological “toolkit” for mixed-effects modeling that leverages recent advances in software for Bayesian inference.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Using Bayesian Mixed Effect Generalized Linear Models to Evaluate Criminological Interventions: An Application to Firearm Seizures during Directed Patrol

  • 期刊论文

  • 发表日期：

    2025年5月6日

查看全文
  • Jason Rydberg,
  • Emily Ann Greene-Colozzi,
  • Edmund F. McGarrell,
  • Sean Perry

摘要

Purpose

To provide guidance on the use of Bayesian mixed-effect modeling for inference in criminological applications involving complex longitudinal data, and a demonstration involving police firearm seizures and gun violence.

Methods

We outline a Bayesian strategy for specifying and fitting mixed-effects models using the brms package in R, including procedures for prior predictive simulation, model selection, and posterior comparison during sensitivity analyses.

Results

To demonstrate the utility of the Bayesian approach, we apply it to complex longitudinal data from a police directed patrol intervention in Flint Michigan, estimating the impact of gun seizures on firearm violence using a three-level mixed effects zero-inflated negative binomial regression model. The demonstration suggested a small and fragile reduction in reported gun violence associated with firearm seizures, which was not sufficiently distinguishable from placebo-based sensitivity analyses.

Conclusions

Bayesian methods are underutilized compared to frequentist approaches in criminological applications. This manuscript outlines an approach to expanding the criminological “toolkit” for mixed-effects modeling that leverages recent advances in software for Bayesian inference.