Overview

Abstract (English)

We analyze the body mass index (BMI) distributions of Canada and the United States. Using the 1999/2000 and the 2012 cycles of the Canadian Community Health Survey, and data from 1999/2002 and 2009/2012 for the US National Health and Nutrition Examination Survey, we contrast differences in the relationships between socio-demographic variables and the BMI distributions both between and over time within each country for males and females. We compare the method proposed by Chernozhukov et al. (2013), which to our knowledge has not been applied in this literature previously, to that of Firpo, Fortin and Lemieux (2009). Both methods allow decompositions of the relationship of each variable separately, but each has particular pros and cons. The former performs better at the tails of the distribution, whereas the latter produces results that are independent on the order in which explanatory variables are employed. Consequently, this study is also an exercise to compare these two methods. Our results provide evidence that the relationships between socio-demographic variables and BMI differ at different points of the BMI distributions. Across the two countries, we find race, immigration status and household income to have statistically significant associations with BMI distributions at all quantiles, but with differing magnitudes. Within the same country, males and females differ in both countries in terms of how their BMIs are associated with educational attainment and household income at different points of the BMI distribution.

Abstract (French)

We analyze the body mass index (BMI) distributions of Canada and the United States. Using the 1999/2000 and the 2012 cycles of the Canadian Community Health Survey, and data from 1999/2002 and 2009/2012 for the US National Health and Nutrition Examination Survey, we contrast differences in the relationships between socio-demographic variables and the BMI distributions both between and over time within each country for males and females. We compare the method proposed by Chernozhukov et al. (2013), which to our knowledge has not been applied in this literature previously, to that of Firpo, Fortin and Lemieux (2009). Both methods allow decompositions of the relationship of each variable separately, but each has particular pros and cons. The former performs better at the tails of the distribution, whereas the latter produces results that are independent on the order in which explanatory variables are employed. Consequently, this study is also an exercise to compare these two methods. Our results provide evidence that the relationships between socio-demographic variables and BMI differ at different points of the BMI distributions. Across the two countries, we find race, immigration status and household income to have statistically significant associations with BMI distributions at all quantiles, but with differing magnitudes. Within the same country, males and females differ in both countries in terms of how their BMIs are associated with educational attainment and household income at different points of the BMI distribution.