From Causation to Description: Modeling Relationships Between Stress, Negative Emotions, and Criminal Intent

• In Event: Advances in Modeling and Causal Inference

Abstract

We use survey data from adults in Dhaka, Bangladesh to answer questions about the prevalence of subjective stress reports and their relationships with self-reported criminal intentions. We explore heterogeneity across residents of different (rural/urban; low/high SES) communities. We leverage two-wave panel data to estimate separate between-person and within-person change associations. We also present comparable associations for stress and emotions (depressive symptoms; criminogenic emotions) as benchmarks for assessing the magnitudes of stress/criminal intent relationships. Generally, results show respondents reporting more stress also report higher criminal intentions and negative emotions, but changes in stress are reliably associated only with changes in negative emotions and not criminal intentions. Our descriptive approach illustrates benefits of practices that we think should be more widely adopted: (1) Focus on item-specific analysis (vs. multi-item scales) to improve precision of descriptive inferences and permit assessment of between-item heterogeneity in associations; (2) Use of ordinal modeling for both dependent (outcomes) and independent (stress) variables to improve precision and accuracy of descriptive inferences; (3) Transforming and reporting associations as meaningful “practically large” marginal effects on outcome scales (vs. model betas) to aid interpretation of magnitudes; (4) Adoption of reproducible Bayesian workflow for model building, description, effect estimation, and data visualization.

Authors

• Jonathan R. Brauer, Indiana University Bloomington

• Jacob C. Day, University of North Carolina Wilmington

• Marijana Kotlaja, University of Missouri-Kansas City