Mokken Scaling Analysis: Scale Development the Nirt way

Abstract

Many questionnaires in Health Psychology consist of ordered items (e.g. behaviors of varying difficulty, beliefs or attitudes of varying intensity) and are thus appropriately examined via IRT methods. However in some studies researchers only aim for ordinal level measurement, as some psychological constructs only claim to differentiate between respondents in terms of degree, not quantity. In these studies nonparametric IRT methods (NIRT) such as Mokken Scaling Analysis (MSA) allow more flexible modeling than parametric IRT, as they require less strict assumptions and are applicable to smaller item pools and sample sizes. This talk aims to demonstrate the flexibility of MSA in scale development in three different research contexts: assessing patient preferences (treatment beliefs in low back pain), health communication processes (HIV status disclosure), and quality of care (routine asthma management in primary care). I will give examples of both confirmatory and exploratory MSA using the mokken R package, and illustrate how it can be applied to examine the dimensionality of item sets and important item properties such as homogeneity, monotonicity and invariant item ordering. I also discuss its limitations and some ways in which NIRT results can be complemented with other analyses.