Invited Talk: Philip Chalmers | Quantitative

Title: Model-based Measures for Detecting and Quantifying Response Bias

Abstract:
An important research area in psychometrics is the identification and quantification of measurement bias. Measurement bias occurs when one or more items on a psychological test, survey, rating scale, and so on, demonstrate favoritism towards at least one group of individuals, resulting in composite test scores that will ultimately favour one group over another. However, while the identification of measurement bias has been studied using many statistical approaches, particularly under the topic of differential item functioning (DIF), obtaining optimal quantifications of measurement bias in the form of effect sizes has yet to be resolved in the literature. In this talk, I will discuss a set of model-based effect size measures for response bias that (a) boast optimal large-sample statistical properties in terms of efficiency and bias, (b) do not require ad-hoc assumptions after models have been fitted, (c) are applicable to any select item response model in current use, and (d) are capable of quantifying response bias in item bundles of any size.