AbstractsEducation Research & Administration

Contributions to latent variable modeling in educational measurement

by R.J. Zwitser




Institution: Universiteit van Amsterdam
Department:
Year: 2015
Record ID: 1273875
Full text PDF: http://hdl.handle.net/11245/1.473177


Abstract

The aim of an educational assessment is to measure a particular ability. The ability is usually operationalized in separately scored test items or tasks. A question that then emerges is how to summarize these item scores into a final score. In answering this question, latent variable models (LVM) play an important role. These statistical models assume that the relationship between the item scores can be explained by an unobservable scale score, i.e., the latent variable. A common scoring approach is therefore to estimate such a model on the test data, and to report for each candidate a latent variable estimate. This thesis considers three topics regarding LVM in educational measurement. The first is the estimation of LVM on data obtained from an adaptive test. For a particular class of LVM and a particular estimation procedure, it is explained that these models can be estimated on this kind of data. The second topic is the use of the sum score. For practical reason, many people prefer for reporting the use of a simple sum score instead of a latent variable estimate. It is explained under which conditions the latent variable approach and the sum score approach lead to the same ordering of candidates. The third topic is the lack of model fit in international surveys. Due to differences between countries, the same model does not fit in each country. It is explained how the use of different models in different countries can result in comparable scores between countries.