Purpose: To extract latent topic models from open-ended survey responses, and test the relationship between the resulting models and police officers' motivation to engage in proactive policing.
Methods: The study relies on a corpus of open-ended responses from 396 police officers collected in a survey. Using structural topic modeling, a type of unsupervised machine learning-based text analysis, the study demonstrates how researchers can gain additional insight into police attitudes.
Results: The first stage of analysis results in the construction of three distinct topic models, which are measures of officer sentiments regarding Professionalism, Naivety, and Distortion. In the second stage, the three topics are tested for hypothesis validity in a confirmatory factor structure using structural equation modeling. The model specification includes closed-ended survey responses from the same respondents regarding their willingness to engage in different levels of proactive policing. This analysis establishes topics of Professionalism and Distortion as significant predictors of proactivity sentiments, while Naivety has no significant association.
Conclusions: This study’s contribution is in demonstrating a novel machine-learning method to examine large collections of open-ended survey responses, and providing a method of analysis for researchers to test for a ‘de-policing’ effect in US law enforcement.