A latent variable approach to responses observed with structured error

Abstract:

We consider responses for which the measurement or observation error is substantial but to some extent predictable. Knowledge of the mechanism which perturbs the response leads to models in which the true response is considered a latent variable, and the observed response a stochastic version of it. These models allow us to separate out the effects of response perturbation and covariates on the parameters of the latent distribution. Examples given are the amount of blood transfused after transplant, and reported age that subjects stopped smoking.

About the speaker: Gillian Heller is Associate Professor and Director of Postgraduate Studies at the Department of Statistics, Macquarie University. Her research interests include Generalized linear modelling and Functional data analysis, with applications in Biostatistics and Actuarial Sciences.