Extension and Application of Local Structural Equation Modeling to Longitudinal Data

• Applicants: Professorin Dr. Andrea Hildebrandt; Professor Dr. Ulrich Schroeders
• Subject Area: Personality Psychology, Clinical and Medical Psychology, Methodology ; Developmental and Educational Psychology
• Term: from 2017 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 390731750

Studying education as a lifelong process and examining the cumulative and interactive effects of learning in multiple contexts across the lifespan presupposes a comprehensive database and flexible analytical tools for modeling change. The National Educational Panel Study (NEPS) offers such high-quality, nationally representative longitudinal data on educational careers and on developing competencies of students and adults in Germany. In order to understand the underlying conditions of learning and to optimize education, variables concerning the school context or family environment are especially relevant. Whereas Structural Equation Modeling (SEM) of longitudinal data has been rapidly advanced in the last decades, there is still need to develop flexible modeling techniques to study development as a function of continuous context variables. The focus of this project is the extension of a recently developed SEM technique for the analyses of longitudinal data. The aim is to exemplify the novel methodology by answering substantive questions of educational research concerning competence development across the lifespan using NEPS data. Local Structural Equation Models (LSEM) allow one to study the parameters of a SEM as being moderated by continuous context variables such as age or socio-economic status. LSEM avoids the artificial categorization of a naturally continuous moderator variable. Although researchers are often concerned with observed mean structures (i.e., learning trajectories), it is necessary to communicate that such questions are inevitably connected with measurement in general and questions on variances and covariances in particular. This is because studying average trends require measurement instruments that invariantly capture performance across age, time and other context variables. Therefore, research studying the variance-covariance structure of abilities are of particular importance for ensuring the soundness of potential mean effects and any substantive analyses. In a nutshell, LSEM is a non-parametric approach that relies on the idea of local, non-parametric regression analyses based on sample weights. LSEM has hitherto only been used in cross-sectional designs; its extension to the longitudinal case is pending and worthwhile in order to address substantive questions in the educational field. In a series of analyses of NEPS data, we will study the usability and utility of the newly developed method to describe competence development over shorter and longer time spans. We will first examine the influence of socioeconomic status (SES) as a continuous moderator variable on academic performance in a latent growth curve model. Second, we study students' learning gains in math and ICT literacy skills with parents' involvement as a variable of family context. Third, we further extend the LSEM method by considering two context variables simultaneously (SES and years of parental education) to study the development of vocabulary.

• DFG Programme: Infrastructure Priority Programmes
• Subproject of: SPP 1646: Education as a Lifelong Process. Analysing Data of the National Educational Panel Study (NEPS)
• Cooperation Partner: Dr. Alexander Robitzsch