Determinants of distributional changes in body mass index over time in the United States: Decomposition method

Background and Objectives: The increasing trend in obesity prevalence over years is one of the most serious health and social issues in the United States. The main objective of this study is to identify determinants of increases in Body Mass Index (BMI) among adults in the United States between 1984 and 2009. We investigate the contributions of economic contextual factors and socioeconomic status to the increase in BMI.  

Methods: We examine not only the increase in the means of BMI, but also the changes in other distributional statistics of BMI such as median, 10th and 90th percentiles, and variance. Recentered Influence Function (RIF) regression is employed to decompose the changes of the BMI distribution between 1984 and 2009 into two components: the contribution from changes in covariates (referred to as an “explained” component or a composition effect) and the contribution from changes in coefficients (an “unexplained” component or a structural effect). The RIF decomposition method easily conducts the detailed decomposition, which subdivide the composition and structural effects into each covariates. The data are from the Behavioral Risk Factor Surveillance System (BRFSS) linked with state level variables such as food prices, number of restaurants per 10,000 capita, and smoking restriction laws.

Results: Our preliminary results showed that only small portions of the changes in the distribution of BMI are explained by the changes in the covariates. Between 1984 and 2009, the mean and median of BMI have increased from 24.3 to 27.8 and from 23.7 to 26.6, respectively. The aggregated changes in the covariates account for approximately 12.4% and 33.3% of these increases of the mean and the median, respectively. The distribution of BMI not only shifted upward, but also widened. The 90th percentile (from 30.8 to 37.3) of BMI showed about 2.3 times faster growth rate than the 10th percentile (from 19.7 to 21.5), implying that the distribution became widespread. Furthermore, the variance of BMI increased by 22.6%. Similar to the mean and the median, only 20.5% and 10.5% of the increases in the 10th and 90th percentiles are attributed to the changes in the covariates. As for the variance, more than 100% of the increase is the structural effect, which is because the composition effect works in an opposition direction to the total increase. The detailed decomposition shows that the declines in the food prices explained larger portions of the increases in the BMI distribution, particular for the 90th percentile, than other covariates. However, the structural effects of the food prices are larger than the composition effect.

Conclusion: Our preliminary findings are that the changes of the BMI distribution over years remain unexplained to a large extent by the changes in socioeconomic status and the economic contextual factors. It may suggest that it is important to investigate changes of people’s behaviors over years.