A data-analytics framework for exploring regression associations in multivariate categorical data of firefighters' PTSD.

We propose a data-analytics framework for exploratory research that aims for a comprehensive understanding of potential associations in the survey data regarding firefighters' PTSD (Post-Traumatic Stress Disorder). The primary focus is to obtain insights regarding joint, marginal and conditional regression associations between an ordinal response variable, firefighters' PTSD, and a set of categorical risk factors in a comprehensive and integrated manner. To achieve this goal, the proposed framework incorporates two established data-driven methodologies: the recently developed non-model based regression association measure named as SCCRAM (Scaled Checkerboard Copula Regression Association Measure) and resampling (bootstrap/permutation) methods. The former facilitates the identification of subsets of risk factors that more effectively account for the overall regression association with PTSD, while also elucidating the roles of the relevant risk factors in both marginal and conditional aspects. The latter provides valuable information pertaining to uncertainties and statistical significances, as well as potential biases and the credibility of the estimated regression associations in multi-dimensional contingency tables which are often subject to sparseness or imbalance. Utilizing the proposed approach, our empirical findings indicate that disorder/mental health related factors have a more substantial association with PTSD, and the relationship between demographic/job-related factors and PTSD becomes more pronounced when accounting for the disorder/mental health factors.