Changes in UK alcohol consumption
This project examines longitudinal alcohol data gathered from the Health and Lifestyle Surveys (HALS) of 1984/5 (HALS1) and 199112 (HALS2) and considers different models to suit different types of response variables. We investigate changes in alcohol consumption and habits between the first and second Health and Lifestyle Surveys and concentrate on examining methods for large longitudinal data sets with sizeable amounts of missing data. Using three distinct approaches to suit the response type, we considered population averaged marginal models for binary responses, in particular, the generalized estimating equations (GEE's) of Liang and Zeger. Since missing data is commonplace in the HALS data, the GEE's must be adapted to take account of this. Hence, the idea of weighted general generalized estimating equations (WGEE) arises. Secondly, we modelled the changes in drinking patterns over time for various groups of the population using multi-category response variables. Based on transitional models, multinomial logit models were used which allow a multi-category rather than just a binary response variable to be used. Finally, we considered general linear mixed models that examine changes over time in continuous variables, in our case, the numbers of units of alcohol consumed per week. In addition to this, we applied a goodness-of-fit test for longitudinal data, devised by Horton et al. (1999), to our WGEE models. We found this test to have several drawbacks, which we investigated further using simulations generated in the R programming language. Our results showed several interesting patterns. In particular, although men consume more alcohol than women, the rate at which women were increasing their consumption was steeper than for males between HALS 1 and HALS2. Furthermore, we saw that older females who drink heavily and have a potential addiction to alcohol are more likely to remain this way than younger people. This thesis examines not only changes in alcohol consumption between the HALS but also looks at possible ways of analysing large scale alcohol surveys.