Title: Investigating Heterogeneity in IRTree Models for Response Styles Using R

Authors: Rudolf Debelak, Thorsten Meiser

Abstract:
IRTree Models provide a flexible framework for the modeling of response styles
and trait-related judgements in rating scales. Conceptually, IRTree Models
consider responses to rating scales as involving multiple processes that can
be represented as nodes in a decision tree. Each node is then parametrized by
an multidimensional item response theory (IRT) model, representing traits and
response styles as latent person parameters. The effectiveness of these models
depends on whether the underlying IRT models are valid and that their item
parameters are stable over respondent populations. We present a novel approach
based on model-based recursive partitioning that aims at detecting and
addressing parameter instabilities in IRTree Models by score-based parameter
invariance tests. Unlike classical approaches that rely on predefined groups
to detect parameter instabilities, our approach allows the flexible detection
of parameter instabilities with regard to categorical and continuous person
covariates. The detection of such instabilities can be further used to obtain
groups of respondents for which the item parameters are stable. In our
presentation, we evaluate the new algorithm using simulated and empirical
datasets. We further discuss and present an implementation of this algorithm
in R, which makes the method widely applicable. The presented method can thus
be considered a valuable, accessible tool for researchers and practitioners
for investigating response styles in a psychometric framework.