A framework for detecting aberrant responses in multidimensional forced-choice measures: based on the iterative weighted expectation-maximization algorithm

Journal of World Prehistory · Springer Nature · 2026-09-16 · DOI: 10.1007/s10869-026-10148-6

Аннотация

Aberrant responses are a persistent concern in multidimensional forced-choice (MFC) measures, stemming from factors such as test design, respondent motivation, or sample characteristics. Such responses can compromise score accuracy, reduce reliability, and diminish the overall quality of MFC measures, ultimately leading to detrimental consequences in contexts such as personnel selection. However, existing methods for detecting aberrant responses in MFC measures often exhibit poor power due to the masking effect, encounter model identification challenges, or have restricted applicability. To address these issues, this study introduces a detection framework based on the iterative weighted expectation-maximization (IW-EM) algorithm to reduce bias in parameter estimates and to identify respondents exhibiting different forms of aberrant responses. By applying multiple iterations, the IW-EM algorithm progressively reduces the masking effect commonly observed in outlier detection, thereby enhancing the framework’s detection capacity. Monte Carlo simulation results indicated that the proposed method substantially reduced bias in parameter estimates and improved detection power across conditions. An empirical application further highlighted its potential as a practical and robust tool for improving the quality of MFC-based measures. Finally, we provide a step-by-step tutorial to demonstrate how researchers and practitioners can implement the proposed method in real-world MFC measures.

Читать у издателя →

Похожие статьи