Konrad Klotzke

Department of Research Methodology, Measurement and Data Analysis
Faculty of Behavioural Sciences
University of Twente

On 15 January 2024 Konrad Klotzke defended his thesis Structured Dependence Modelling: A Bayesian Framework for Data-Driven Validation of Psychometric Measurement Instruments at the university of Twente

Project
Marginal Joint-Modelling of Response Accuracy and Response Times

Current approaches in psychometrics to integrate response accuracy (RA) and response times (RTs) in a single joint-model are limited by assuming a fixed correlation between a test-taker’s ability and working speed across the whole test. In a new marginal joint-model, the latent ability and speed parameters are integrated out. An explicitly modelled covariance structure is introduced to explain the correlation between a test-taker’s RA, RTs and the cross-correlation between RA and RTs. The cross-correlation between RA and RTs may change over the course of the test, thus allowing to model a variable speed-accuracy trade-off. The joint-model’s covariance structure is extended to allow multidimensionality in the relationship between RA, respectively RTs. Groups of test-takers can be nested in clusters and covariates measured at different hierarchical levels of the model can be included. Parameter estimation is compared to existing means for joint-modelling of RA and RTs. Hypotheses about the (cross-) covariance parameters and the mean structure can be evaluated with Bayes factor testing. For Bayesian model selection, performance of the Bayes factors is compared to the Bayesian information criterion (BIC). Educational data is utilized to demonstrate how speededness, change in ability and multidimensionality can be investigated in the marginal modelling framework.

Supervisors
Dr. Ir. J.P. Fox

Financed by
University of Twente

Period
July 2017 – January 2024