Understanding SEM: where do the numbers come from?

Half day short course (Monday, July 11; 2:30PM-5:30PM)

This workshop will attempt to demistify SEM software. In particular, this workshop will explain how SEM software finds the values that are shown in a typical output of an SEM analysis, both in theory and in code. The workshop will first provide the necessary theoretical background needed for point estimation (maximum likelihood estimation, optimization, matrix representation of SEMs) and will then illustrate how this can be translated to R code. The workshop will demonstrate how one can construct a simple function in R that takes a vector of free parameters as input, and returns the value of the so-called ML discrepancy function. By exploiting a built-in optimizer in R (nlminb), this will allow us to find the maximum likelihood estimates for these free parameters, for a given model and a given dataset. Next, the workshop will explain the concepts of information matrices, and demonstrate how (robust) standard errors can be computed. If time permits, the workshop will also demonstrate how robust (Satorra-Bentler) test statistics can be computed. Simple and easy-to-understand R code will be provided for all computations. This workshop aims to deepen your understanding of how structural equation modeling works, but the knowledge gained is equally applicable to other psychometric models.

About the Instructor

Yves Rosseel

Yves Rosseel is a full professor at the Department of Data Analysis, Ghent University (Belgium). His research interests include computational statistics, psychometrics, and structural equation modeling. He is a strong advocate of open source software for scientific purposes, and contributed code to several projects. He is the main developer of lavaan, an R package for structural equation modeling.