This tutorial demonstrates a streamlined parametric g-formula using iterated conditional expectations for effective causal analysis in longitudinal studies.
Key Points
The aim is to present an accessible method for implementing parametric g-formula using iterated conditional expectations to improve causal inference in longitudinal designs.
Introduced the iterated conditional expectation (ICE) variant of parametric g-formula.
Showed how to use regression models for mean outcomes across time points with lavaan in R.
Provided a running example illustrating social exclusion's effect on depression.
ICE demonstrated reduced susceptibility to model misspecification biases compared to traditional methods.
A doubly robust version of ICE offers additional protection against biases from incorrectly specified outcome models.
Parametric g-formula using ICE was shown to be computationally efficient and user-friendly.