Analyses encumbered by zero inflation - a day in the life of a quantitative psychologist working with clinical data

This talk presents two ongoing studies addressing analytical challenges posed by heavily skewed data. The first study examines how hurdle models, part of the IRTree model family, can improve reliability assessment of highly skewed Likert-scale data. We investigate factors contributing to reliability overestimation when responses contain both structural zeros (true non-responders) and sampling zeros (chance non-responses), and how the hurdle model is better equipped to assess reliability in these data. The second study introduces Bayesian changepoint models as an alternative to semi- and non-parametric methods for identifying nonlinear trends in psychological data. After presenting the methodology, we demonstrate empirical applications examining nonlinear relationships in psychopathology, substance use disorders, and usage frequency patterns.