REMINDER CENTRE FOR MEDICAL STATISTICS SEMINAR SERIES on BIOSTATISTICS 2000/2001 NO 2: Wednesday, December 13th, 2000, at 2:30 pm Multilevel Modelling of Nominal Data and Rankings by Dr Sophia Rabe-Hesketh (Institute of Psychiatry, London) Unordered categorical responses are of interest in many disciplines, examples including party voted for in an election or treatment selected for a patient. Here the response is the category 'selected' by an individual. Rankings arise when the individual does not just select one category, but orders the categories according to some criterion. Such data can be modelled by assuming that each individual assigns 'utilities' to the categories and 'selects' the category with the greatest utility (nominal data) or ranks the categories according to the utilities (rankings). For nominal data and independent utilities, certain distributional assumptions for the utilities lead to the well-known multinominal logit or polychotomous logistic regression model. I will discuss extensions to this model for rankings and for correlated utilities. If the data are multilevel, the utilities are not just correlated within individuals but also between the individuals in the same higher level unit or cluster. The models represent generalisations of generalised linear mixed models and can be estimated by maximising the marginal likelihood using numerical integration. The method is implemented in a Stata program called gllamm6. This work was done jointly with Anders Skrondal. All welcome!! Venue: Room 2.22, Third Floor The MacKay Building Keele University http://www.keele.ac.uk http://www.keele.ac.uk/depts/ma/seminars/medstats.html http://www.keele.ac.uk/university/campus/maps/ ____________________________ Janet Drewery Secretary Centre for Medical Statistics Keele University Staffordshire ST5 5BG England Email: [log in to unmask] Tel: (01782) 583269 Fax: (01782) 583269/584268 ______________________________