Structural Equation Modeling for Description, Prediction, and Causation

Abstract

Structural equation modeling (``SEM'' for short) is a widely applicable statistical analysis framework that is popular among psychological scientists (and related disciplines). Valid use of this technique requires researchers to clearly distinguish the type of research they are interested in. Scientific research can roughly be divided into descriptive, predictive, and causal research, and each type of research has different implications for the analysis strategy. One of the problems, however, is that this distinction is often implicit in psychological research. Therefore, it can be unclear whether research results actually answer the research question. In Chapters 2 through 5, I collaborate with applied researchers on both descriptive and predictive research projects, and I describe extensions of one specific popular longitudinal SEM model. Moreover, many alternative analytical techniques have been in disciplines such as epidemiology and biostatistics for causal research. These methods are still largely unknown among psychological researchers. In Chapters 6 and 7, I therefore compare popular SEM models with analytical techniques from biostatistics for longitudinal and observational causal research: which method performs better and under what conditions? Furthermore, I introduce psychological researchers to these alternative causal methods, and discuss if they can be applied in the context of psychological research.