Thread by @Michael_Spagat

1. Here's a new thread on #HurricaneMaria death toll numbers.

#PuertoRicoDeathToll @cblatts @ARGohdes @tanishafazal @AppDemography @KLdivergence @skoczela

2. There seems to be some confusion about what the death estimates are.

First, there is no list of, e.g., 3,000 named individuals killed by the hurricane.

3. Creating such a list would be valuable work but there could never be a fully compelling list of people unambiguously killed by the hurricane with 3,000 names on it.

4. This is because there will be a large grey area of people who died due to multiple causes of which the hurricane is, plausibly, one.

5. Of course, there will be cases where, for example, the hurricane uproots a tree that strikes and kills someone. Such cases can be unambiguously blamed on the hurricane.

6. But other cases will be more complicated. For example, a man has a heart attack, maybe the ambulance comes more slowly than normal and maybe the best qualified paramedic isn't there perhaps due to the hurricane and these lapses in service may play a role in the man dying.

7. So case by case classification is very hard work and it is not possible to unambiguously classify every single case no matter how hard you try.

8. Thus, we turn to statistical estimation which amounts to looking for spikes in post-hurricane death rates that are outside of recent historical patterns.

9. In other words, we compare actual post-hurricane death rates with counterfactual death rates that would have prevailed, one argues, if the hurricane had never happened.

10. Such estimates will depend on your "what if" theory.

11. The early estimates that were done assumed, in various ways, that the death rates that would have prevailed without the hurricane would have been just like death rates in the recent past.

12. There are variations in these estimates that don't have to detain us here. But, basically, you just do something like assume that the October 2017 death rate (right after Maria) would have equaled the average October death rate over the last 5 years without the hurricane.

13. When you do something like this then you find a bit more than 1,000 "excess deaths" during October-December of 2017.

14. Then a team at George Washington University added a clever idea which amounts to changing the theory of what would have happened without the hurricane.

15. Specifically, they accounted for the fact that many people left the island after the hurricane. This means that the post hurricane death rate for remainers was substantially higher than it appears to be at first glance because there was a large drop in the island population

16. Accounting for this population change adds something like another 1,000 plus deaths to the estimate.

17. In addition, the GWU team included January and February something like another 1,000.

18. Note - the numbers in the last two tweets were off the top of my head without rechecking the GWU report. Also, annoyingly, the GWU have not responded to two emails I sent then asking for more information so I don't understand what they did as well as I want to.

19. To summarize, the 3,000 number is not a list of particular individuals documented to have been killed by the hurricane. Rather, it is an estimate based on a counterfactual scenario for what would have happened without the hurricane, taking into account migration from PR.

20. There was also the earlier estimate of nearly 5,000 put out by a Harvard team that seems to have been, correctly, eclipsed by the GWU estimate.

21. The Harvard estimate is based on a not-fit-for purpose survey that led to a preposterously wide uncertainty interval that was largely ignored in media reporting. Essentially, this estimate should have been ignored and now it seems that it is getting ignored.

22. It is, perhaps, surprising that Trump didn't attack the Harvard study which was really vulnerable.

23. The GWU study seems to be quite solid although it would be a lot better if they published more details and responded better to questions.

End

Addendum

24. A number of people wrote in making the point that the people leaving the island were probably much healthier than those who stayed. I hadn't thought of this but it sounds highly plausible.

25. This factor would lower the 3,000 estimate to the extent it is true. This is because the migration adjustment amounts to augmenting the official vital statistics death tallies for the island with an estimate of off-island deaths for migrants.

26. Imagine an extreme case where all migrants were super healthy and none of them died after leaving the island. If you knew this fact then you wouldn't want to make any adjustment for migration.

27. @RAVerBruggen even consulted the GMU report and thinks that they must have assumed those leaving the island were less healthy than those who stayed. I haven't had a chance to check this myself.

28. I think that the GMU people did something rather complex and the full details could lay this question to rest. I am going to write to them again asking for clarification.

29. Also, I can report that I did something simple a few weeks back with my collaborator Stijn van Weezel and we found that we could get numbers like the GWU's ones assuming that the migrants from the island were equally healthy with the ones who stayed.