Dave Lunn and Jon Wakefield
We present preliminary analyses of a study investigating the association between a particular drug and allergic rhinitis. The data consist of four outcome measures, each recorded on a 4-point ordinal scale (none/mild/moderate/severe). We analyse these data using the class of cumulative logit models. We provide a theoretical framework for these (and related) models, and show how they may be interpreted in terms of latent continuous measures. Particular attention is paid to commonly used modelling assumptions, such as ‘proportional odds’; we also assess the predictive performance of the various models.
Reference: PAGE 7 (1998) Abstr 677 [www.page-meeting.org/?abstract=677]
Poster: oral presentation