Determinants of the Body-Mass Index in Canada and England: An Analysis Using Relative Distributions

Rationale:The body-mass index (BMI) is commonly used to identify individuals who are underweight, of ideal weight, overweight or obese. The rise in the proportion of individuals classified as obese or overweight has become a health and economic concern for developed countries. A problem with this classification is that it does not consider the whole of the BMI distribution, and changes in BMI which do not cross a category boundary will be undetected. Further, more (or less) BMI is not better (or worse). Ideally individuals would have a BMI of around 22, with very low and very high BMI considered a risk factor for poor health. A standard regression methodology explaining changes in the mean is therefore inadequate for explaining relevant changes in the distribution of the BMI and it is important to measure and explain movements across the whole distribution. Objectives: The objective of this paper is to examine changes in the distribution of the body-mass index (BMI) in England and Canada during the period 1994-2005. This paper highlights key changes in the distributional shape and location and also evaluates the impact of socioeconomic characteristics, considering how these variables impact on the distribution of the BMI. Methodology: We use relative distribution methods to investigate differences across the whole distribution. The relative distribution is the basis of a nonparametric methodology for comparing two distributions based on observing where an observation from a comparison distribution would rank in the cumulative density function of a reference distribution. The relative distribution is invariant to monotonic transformations and while statistics analogous to the Gini and Concentration Indices can be constructed based on the relative distribution, important characteristics of changes in the distribution of the BMI are more naturally handled using measures of relative entropy and polarization. The relative distribution can then be decomposed with respect to socioeconomic characteristics. Counterfactual distributions of BMI controlling for socioeconomic characteristics are compared to the actual distributions. From these we can evaluate the compositional effect on the relative distribution and what the relative distribution of England and Canada would have looked like if the socioeconomic characteristics in the two populations were the same. Results: Our results show that while BMI has increased in both countries, BMI in England has increased at a much faster rate than in Canada. Both groups show polarization over time towards both tails of the weight distribution, with the English polarizing towards the upper tail at a faster rate than Canadians. The decomposition results show the importance of socio-economic and lifestyle variables on the differences in the distribution of BMI between the two populations. These results are sometimes hidden using traditional regression based methods. Conclusion: This paper shows the importance of considering the whole distribution when investigating the BMI and making comparisons between two different populations. We demonstrate the advantages of relative distributions over traditional methods of comparing distributions. We also demonstrate the importance of considering whole distributions when investigating the impact of socioeconomic characteristics and the impact these have on the distribution of BMI.