Ummugul Bezirhan, Boston College
Conditional dependence between response time and accuracy in cognitive diagnostic models
2022 Dissertation Prize
Computer-based tests are rapidly becoming the common practice in educational and psychological testing, thus a considerable effort has been put into incorporating process data specifically response time (RT) into measurement models. The assumption of conditional independence between response accuracy and RT, given latent ability and speed, is commonly imposed in the joint modelling framework of response and RT. Recently several studies have shown violations of the conditional independence assumption, which prompted various models that accommodate conditional dependence of responses and RTs, especially in the Item Response Theory framework. Despite the widespread usage of Cognitive Diagnostic Models as formative assessment tools, little has been done in exploring the conditional joint modelling of responses and RTs within this framework. This research proposes a conditional joint response and RT model in CDM by using an extended reparametrized higher-order deterministic input, noisy ‘and’ gate (DINA) model for response accuracy. The item-specific effects of residual RT are incorporated into the response accuracy model to capture the conditional dependence. The effect of ignoring the conditional dependence on parameter recovery is explored with a simulation study, and empirical data analysis is conducted to demonstrate the application of the proposed model.
About the Speaker
Ummugul Bezirhan is a Senior Research Specialist at TIMMS and PIRLS International Study Center in Boston College. She holds MS and PhD in Measurement, Evaluation and Applied Statistics from Columbia University. Her research mainly focuses on extending and implementing latent response models utilizing process data, as well as developing novel procedures to improve methodologies used in large scale assessments. She previously gained professional experience as a researcher at National Board of Medical Examiners (NBME) and American Institutes of Research (AIR) and as a data scientist at NBCUniversal.