Definition and estimation of causal effects in latent state trait models (CaST)

点击翻译按钮获取中文摘要

Latent State Trait (LST) theory is one of the most influential theories for studying processes of change and variability in psychological traits. LST models allow researchers to study trait change and variability processes in a flexible and versatile way. Central but as yet unresolved questions concern the conditions under which (the parameters in) LST models can be causally interpreted and estimated from observational data. The project considers the issue of causal inference in LST models from three perspectives: a) statistical versus causal models, b) discrete-time versus continuous-time models), and c) stochastic versus graph-based (Steyer vs. Pearl) causal theory. The contribution of the project lies in a systematic and comprehensive investigation of how causal effects of time-varying variables, time-stable variables, and their interactions on trait changes and variability parameters can be defined and estimated in discrete-time and continuous-time latent state-trait models with autoregressive effects. In addition to the theoretical developments, the newly developed LST models are extensively investigated in simulation studies and real data applications.