This paper discusses the statistical technique of fitting structural models, which is illustrated with data from a survey on drinking and driving. The only important research findings discussed are those which help to illustrate the technique. A 1991 survey of about 600 subjects used a 21-page questionnaire, which was completed by a trained interviewer. Subjects were selected from driving licence records on the basis of age and sex, with a quota for each interviewer. Only current drivers, who sometimes drunk alcohol away from their home, were interviewed. The questions about drink and driving were particularly important, because they are examples of items that form the dependent drink drive variable when combined. The underlying model for the questionnaire design was the Theory of Planned Behaviour. The questions were all based on a seven-point scale. The statistical analysis was based on a LISREL model which, in its most general form, consists of a structural equation model and a measurement model. The resulting correlation matrices were analysed. The 'final' analysis presented has two latent output variables, referring to the different behaviours concerned with: (1) any drinking and driving; and (2) drinking and driving while over the limit. The paper ends by raising some questions about the interpretation of such analyses.
Samenvatting