Annica Dominicus, AstraZeneca
	
Modelling variability in longitudinal data using random change point models

Abstract:

Some cognitive functions exhibit multiple phases in old age, which
motivates the use of a change point model for the individual
trajectory from repeated measures data. The change point varies
between individuals and is treated as random. We contrast the random
change point model with linear and quadratic random effects models,
focusing primarily on trait variability over age groups. The methods
are illustrated using Swedish data on cognitive function in old age
and through simulations. We show that the models impose different
restrictions on the trait variance over age groups, and we demonstrate
that the random change point model has favorable properties. The
performance of approximate maximum likelihood estimation based on
first order linearization of the random change point model is
contrasted with a Bayesian approach based on Gibbs sampling.