Conference Objects and Reports

Extending Latent Regression IRT with Response Time

Chen, Yi; Yang, Yi; Lee, Young-Sun

In large-scale assessments (LSAs), item responses and context questionnaires (CQ) are incorporated into the multivariate latent regression IRT for generating the plausible values (PVs) of the population proficiency variables. With the advance of computerized tests, it has become common for test administrators to capture accurate process information (e.g., frequency of using help, number of answer change, and response times). Based on the empirical studies (von der Linden, 2009; Bolsinova & Tijmstra 2016), the positive residual correlation between response accuracy (RA) and response times (RT) are common. Consequently, ignoring process data in the plausible value generation may break the rationale developed by Rubin (1987), that all available information should be used in multiple imputations when data are missing at random (MAR). The purpose of this study is to extend the conventional latent regression IRT model with response time. This approach has been shown to improve the congeniality in data imputation. The proposed model is estimated in a fully Bayesian approach and compared with the traditional treatment of the plausible value generation under varying conditions.

Files

More About This Work

Academic Units
Human Development
Published Here
June 14, 2022

Notes

International Meeting of the Psychometric Society (IMPS), 2020