Item Response Trees: A recommended method for analyzing categorical data in behavioral studies

Abstract

Behavioral data are notable for presenting challenges to their statistical analysis, often due to the difficulties in measuring behavior on a quantitative scale. Instead, a range of qualitative alternative responses is recorded. These can often be understood as the outcome of a sequence of binary decisions. For example, faced by a predator, an individual may decide to flee or stay. If it stays, it may decide to freeze or display a threat and if it displays a threat, it may choose from several alternative forms of display. Here we argue that instead of being analyzed using traditional nonparametric statistics or a series of separate analyses split by response categories, this kind of data can be more holistically analyzed using a generalized linear mixed model (GLMM) framework extended to binomial response trees. Originally devised for the social sciences to analyze questionnaires with multiple-choice answers, this approach can easily be applied to behavioral data using existing GLMM software. We illustrate its use with 2 representative examples: 1) repeatability in the measurement of antipredator display escalation and 2) the analysis of predator responses to prey appearance.