Funded by:
(Login )

Analyzing relations between latent competencies and context information in the NEPS

Minimize
PIs:
Dr. Christian Aßmann, University of Bamberg
Prof. Dr. Claus H. Carstensen, University of Bamberg
Prof. Dr. Steffi Pohl, FU Berlin

Resesarch Staff:
Luise Fischer, University of Bamberg
Christoph Gaasch, University of Bamberg
Theresa Rohm, University of Bamberg
Thorsten Schnapp, University of Bamberg
Daniel Schulze, FU Berlin (since 03/2017)
Eric Stets, FU Berlin (until 12/2016)


Project Summary:
Large-scale studies such as the National Educational Panel Study (NEPS) address the important issue of the acquisition of competencies over the life course and factors influencing this acquisition. Thus, researchers are interested in the relationships between latent competencies and context variables (such as characteristics of parents, teachers, or the school). When estimating these relationships, different methodological issues have to be addressed. One issue are missing values in context variables, which need to be adequately accounted for. Another issue is measurement error in the assessment of competencies, which is usually accounted for by modeling competences as latent variables within an Item Response Theory (IRT) framework. To account for the specific data structure in large-scale assessments, these IRT frameworks have to allow for incorporating hierarchical structures related to both sampling issues and the stratified educational system. Furthermore, in order to investigate competence development–one of the main issues in longitudinal large-scale assessments–, special longitudinal IRT models are needed that allow for linking of test scores from different measurement occasions.

In this project we will develop an approach that allows to investigate relationships of competences and competence development with explaining variables in longitudinal large-scale assessments accounting for these different methodological aspects. In the proposed one-step approach measurement error in competence assessment and missing responses in context variables are simultaneously accounted for. The approach will be extended to incorporate different IRT models, hierarchical data structures, as well as longitudinal data analyses. The project consists of three sub-project in which the different aspects are approached. In the first sub-project (Christian Aßmann, Christoph Gaasch, & Thorsten Schnapp), MCMC based estimation routines for the IRT model framework with missing values in background variables will be developed and their performance will be tested. The second sub-project (Claus Carstensen, Theresa Rohm & Luise Fischer) the hierarchical structure within the NEPS context stemming from sampling as well as the stratified educational system will be modeled. The third sub-project (Steffi Pohl, Eric Stets, & Daniel Schulze) investigates possible linking strategies to allow for modeling competence development over time. Within the course of the project, the different sub-projects will be integrated in one framework. A respective R-package will be developed that allows researchers working with data from large-scale assessment data to use our framework for analyzing their data.