Sum of deaths differs over race/ethn groups vs. over income quintiles

[Dan Shawhan]

Newbie question: In a set of US inmap results we have, the total population is exactly the same whether you sum the populations of all the race/ethn groups or the populations of the 5 income quintiles: 360,041,391. But if you sum the deaths, the sums are not the same. They are about 29% higher if you sum the deaths of all the race/ethn groups than the deaths of the quintiles. Any idea about the likely reason? I browsed all the thread titles of the past users group threads, didn't see one that seemed likely to address this.

Thanks!

Dan Shawhan

 

Daniel L. Shawhan

Resources for the Future & Cornell U.

(202) 328-5027

rff.org/shawhan

 

[Chris Tessum]

Hi Dan,

If I remember correctly, the racial ethnic groups are overlapping. It's been a while since I looked at that particular data but it might be something like Hispanic (which is an ethnicity) is one of the groups, but all of the races can also be either hispanic or not, so there is some double counting.

Best regards,

Chris

[Dan Shawhan]

Thanks very much, Chris! Teammates and I also figured out another explanation: For the race/ethn groups, we used group-specific hazard ratios. For the quintiles, we used a society-wide hazard ratio. The society-wide one is smaller than the weighted average of the group-specific ones (and perhaps even more so for the pollution change pattern we are analyzing). We used hazard ratios from Di et al. 2017: https://www.nejm.org/doi/suppl/10.1056/NEJMoa1702747/suppl_file/nejmoa1702747_appendix.pdf.

Dan Shawhan