Performance of a general location model with an ignorable missing-data assumption in a multivariate mental health services study

Abstract

In a study of the impact of case management teams in a publicly funded mental health programme, mental health patients were interviewed about a variety of outcomes suggestive of successful community adaptation, such as support from family and friends and avoidance of legal problems. Because outcome data were missing for a number of patients, a follow-up study was carried out to obtain this information form previous non-responders whenever possible. Because the data of interest were multivariate and included both continuous and categorical variables, a candidate approach for handling incomplete data in the absence of follow-up data would have been to fit a general location model, presumably with log-linear constraints on cell probabilities to avoid overfitting of the data. Here, we use available follow-up data to investigate the performance of a series of general location models with ignorable non-response. We note some problems with this approach and embed the discussion of this example in a broader consideration of the role of ignorable and non-ignorable models in applied research.