A0814
Title: Bayesian estimation of ordinal cross-lagged panel model
Authors: Vipasha Goyal - University of Amsterdam (Netherlands) [presenting]
Maarten Marsman - University of Amsterdam (Netherlands)
Abstract: The cross-lagged panel model (CLPM) is a widely used statistical method in psychological research for examining reciprocal and dynamic relationships between variables over time and drawing causal inferences. Despite their widespread use, existing implementations of the CLPM are limited in their capacity to support flexible Bayesian estimation and inference, restricting our ability to quantify model uncertainty. Without a formal treatment of model uncertainty, parameter estimates are conditioned on one selected model, which can lead to overconfident and biased inferences when multiple plausible models exist for the data at hand. Moreover, most implementations assume continuous and normally distributed variables, whereas many psychological constructs are measured using ordinal scales. To address these limitations, a Bayesian framework for ordinal CLPM is developed, incorporating Bayesian model averaging techniques for variable selection. This framework enables efficient estimation of dynamic relationships over time through posterior inclusion probabilities and inclusion Bayes factor. The proposed models and procedures are made available in free software packages in R and JASP.
