Summary. Missing values are common in cohort studies of the elderly. As part of a study of cognition in older age, we implemented a model for the analysis of longitudinal depression data complicated by the presence of non-ignorable missing values, censored measurements and individually varying times of observation. The repeated measures and non-response mechanisms are jointly modelled by assuming that they depend on a common underlying process. The results of our analysis suggest that both depression and non-response increase with age and that women have systematically higher depression scores than men but do not have higher levels of study non-participation.