Hurdle Models in Psychology—A Practical Guide for Inflated Data

Abstract

In psychological research, variables often exhibit point-mass inflation—for example, many zero responses or other boundary lumps—that defy standard regression techniques. Hurdle models address this challenge by separating the zero-generating process from the distribution of nonzero (or non-boundary) observations, thereby allowing for more accurate modelling of behaviour and outcomes. In this paper, I introduce the conceptual basis of Hurdle models and demonstrate how they can be applied to count data as well as other types of data (e.g., continuous variables with excess zeros). Using a step-by-step tutorial in R, I illustrate how the two-part hurdle structure—consisting of a binary component for point-mass observations and a truncated distribution for positive (or above-threshold) values—provides nuanced insights that simpler models often miss. To illustrate this approach, I walk through a fictional dataset examining home-based HIV testing among men who have sex with men, highlighting the Hurdle model's ability to simultaneously handle overdispersion and excess zeros. Emphasising iterative model evaluation, goodness-of-fit checks and a series of practical recommendations, this paper aims to equip psychologists with a robust analytical framework that promotes deeper, theory-aligned interpretations of data—ultimately fostering innovative research in diverse areas of psychological science.