Read on Twitter
New @Env_Sci paper out with @Fcorowe in @RGS_IBG Area on understanding trends in BMI. So what is it all about 
https://rgs-ibg.onlinelibrary.wiley.com/doi/10.1111/area.12675
https://rgs-ibg.onlinelibrary.wiley.com/doi/10.1111/area.12675
A while back, I had the pleasure of working with @DrFahadRazak and Subu at @HarvardChanSPH to investigate trends in BMI. We showed that while increasing trends in median BMI had begun to level off, the upper end of the distribution continued to increase https://jech.bmj.com/content/70/8/832.short?g=w_jech_ahead_tab
These trends were consistent by age group, sex and education level (as proxy of socioeconomic status). What was driving these trends was clearly affecting a diverse range of population groups.
This had important implications for population health, however the reasons behind it were puzzling. What was driving these trends? There wasn't any evidence out there. I posted a question into our @geodatascience slack channel and @Fcorowe knocked on my door - they had an idea!
@Fcorowe proposed we use a combination of quantile regression and decomposition analysis to try and identify whether these trends can be explained by contextual or compositional drivers? So we set out a plan to apply these novel methods.
Context here would the drivers of BMI changed in strength, direction or importance over time at each point of the BMI distribution. Composition would be changes in the make-up of people across the distribution.
We found evidence of the widening of the BMI distribution being driven by changes at the top of the distribution - consistent with the evidence in our JECH paper.
Teasing out the reasons behind these trends, we find that the factors associated with BMI are not consistent across the distribution of BMI. For example, physical activity, race and SES had far stronger influences on BMI at the 90th centile vs 10th or 50th centiles.
Decomposing these associations over time revealed contextual changes in the nature of relationships appeared to be stronger determinants of the trends, than compared to compositional explanations.
Race and SES had become stronger factors over time in explaining BMI at the 90th centile, with changes in population composition counter-balancing these contextual changes.
Our results suggest that we need to start trying to unpick the reasons behind the heterogeneous nature of correlates of BMI. Utilising this knowledge to design interventions to target the different parts of BMI distribution may be effective for policy makers.
Full code for the analysis can be found here https://github.com/fcorowe/bmi
