Napier, Gary Alexander
Modelling obesity in Scotland.
MSc(R) thesis, University of Glasgow.
Full text available as:
Overweight and obesity prevalence has been on the increase in Scotland since the 1990s when monitoring began with the Scottish Health Surveys. In this study models are developed that describe the prevalence of obesity in Scotland by analysing data from the Scottish Health Surveys. In particular, we study how the Body Mass Index (BMI) and waist-to-hip ratio (WHR) vary with gender, age and socio-economic status, and investigate whether there is a shift in the entire BMI/WHR distributions or simply a stretching-out of the upper tail (roughly corresponding to the overweight/obese categories).
Logistic regression is employed for modelling obesity prevalence, and generalised additive models are employed to examine the relationship between BMI/WHR and gender, age, socio-economic status and survey year. The odds of being overweight or obese increase with age, which is also the case for log(BMI) and WHR, with differing gender patterns. The rate of increase in BMI and WHR is at its greatest for individuals aged between 16 and 30, and gradually slows down before decreasing for males over the age of 55, but remains increasing for females of the same age. No significant difference in obesity prevalence is observed for males in social classes iii manual and iv & v, but males in social class iii non-manual are 1.27 times more likely to be obese in comparison to males in social classes i & ii. For females, the odds of being obese increase with each consecutive social class.
Quantile regression is used to study how the entire conditional distribution of BMI and WHR vary with gender, age, socio-economic status and survey year. By specifying changes in the quantiles of the response (BMI/WHR), quantile regression highlights an uneven increase in the BMI and WHR over time; in each subsequent survey all quantiles shift to the right, but this increase is larger for the upper tail of the distribution. The effects of socio-economic status also vary across the quantiles of the BMI and WHR distributions, with males in each subsequent social class who lie at the lower end of the distribution having lower BMI values than males in social classes i & ii, but higher BMI values at the upper end of the distribution. Finally, subtle gender differences are observed in the relationship between BMI/WHR and age. In conclusion, quantile regression allows us to go beyond obesity prevalence and examine finer aspects of the BMI and WHR distributions.
Actions (login required)