A New Estimation Method for Multdimensional IRT Models with application to Large-scale Educational Assessments Faculty of Education
EDUCATIONAL AND PSYCHOLOGICAL MEASUREMENT COLLOQUIUM SERIES
A New Estimation Method for Multdimensional IRT Models with application to Large-scale Educational Assessments
Dr. Björn Andersson
Centre for Educational Measurement
University of Oslo
April 4, 2018 (Wednesday)
13:30 – 15:00
Room 526, Meng Wah Complex, HKU
In large-scale educational assessment programs such as PISA and TIMSS, multidimensional latent regression item response theory (IRT) models are used as the underlying models for the reported mean scores in different countries or regions. The prevailing approach in these programs is to conduct a stepwise estimation procedure in which the item parameters and latent regression parameters are estimated separately and where a unidimensional model is assumed for the item parameter estimation. In this talk, we instead propose to estimate multidimensional latent regression IRT models in a single step in the context of large-scale educational assessments. For this purpose, we introduce an efficient estimation method for multidimensional simple structure IRT models and evaluate its properties for binary, ordinal and nominal observed variables. Using data from the PISA study, we illustrate the differences between the stepwise and the simultaneous approach and outline how the proposed method can be incorporated in large-scale educational assessment programs to improve the estimation of population-level parameters.
About the Speaker
Dr. Björn Andersson is an Associate Professor at the Centre for Educational Measurement, University of Oslo. His research interests include large-scale educational assessments, test equating, the use of process data in educational measurement, and estimation methods for latent variable models.
Everyone is welcome to attend!
If interested, please confirm your attendance by sending an email to firstname.lastname@example.org.