About 1,100 Puerto Rican Deaths from Maria — NOT 2,795 or 4,645

The estimated number of above-average “excess deaths” in Puerto Rico attributed to Hurricane Maria (Sept 20, 2017) is a difficult figure to estimate objectively.  Puerto Rico’s official figure of 64 deaths by December 9, 2017 (which the President remembered) counted only those deaths directly attributed to the storm and confirmed by medical examiners.  Most of the direct deaths from Katrina were from drowning – which is much easier to attribute to the storm than many other causes of death. Studies of Puerto Rican deaths from Maria aspire to account for a wide range of indirect effects that are presumed (not proven) to be consequences of the storm such as suicides and heart attacks, infectious diseases, and damage to electricity and therefore to dialysis and respirator equipment.

Among at least eight major studies of direct and indirect effects on mortality attributed to Maria, two outliers stand out as being 3-5 times larger than the others, which all cluster around 1000. The first big number was from Harvard. On September 13, Time said, “Harvard’s report, which was based on systematic household surveys throughout Puerto Rico, reached an estimate of 4,645 storm-related deaths between September and December 2017, many as a result of ‘delayed or interrupted health care.’”  Nonsense. The Harvard study extrapolated from only 15 deaths reported in a survey of 3299 households to estimate that “between 793 and 8498 people died … up to the end of 2017.” By adding 793 and 8498 and dividing the result by 2, Time and others came up with a totally meaningless “average” which were widely reported with predictable sensationalism: “The hurricane that struck Puerto Rico in September was responsible for more deaths than the Sept. 11 attacks and Hurricane Katrina combined,” exclaimed The Daily Beast.” In reality, these “estimates of death from people who were interviewed” are little better than an opinion poll, and finding 15 deaths out of a sample of 3299 can’t plausibly be multiplied into 4645 for the whole island.

The latest sensational estimate of 2,975 excess deaths over six months is from an August 28 report from the Milken Institute School of Public Health at George Washington University  (GWU) commissioned by the Government of Puerto Rico. The study mentions two “scenarios” (census and displacement) yet only publicized the one with the biggest number: “Total excess mortality post-hurricane using the migration displacement scenario is estimated to be 2,975 (2,658-3,290) for the total study period of September 2017 through February 2018.” 

The 2,975 estimate only applies to the “displacement scenario.”  That is, the study “estimates cumulative excess net migration from Puerto Rico in the months from September 2017 through February 2018 and subtracts this from the census population estimates in these months.”  The population fell by about 8%, mainly due to migration rather than death, so the fact that there were more deaths than average after the hurricane means the death rate (deaths per thousand) rose more than the unadjusted statistics would suggest because the population is smaller.  But this issue is the number of deaths, not the death rate, and displacement (migration) did not make that number any higher than half a dozen other studies found (about 1000) much less three times higher.  

Trying to explain the high “displacement scenario” estimate, Eliza Barclay at VOX writes, “The ideal way to calculate the death toll from a hurricane, disaster researchers say, generally, is to count all the deaths in the time since the event, and then compare that number to the average number of deaths in the same time period from previous years. Subtract the average number from the current number and that’s the death toll.”  Unfortunately, the GWU “displacement scenario” estimate does not do that.  What it does instead is to compare what actually happened with hypothetical simulations of what might have happened without the storm.  Those projections come from “a series of generalized linear models (GLMs)… accounting for trends in population … in terms of age, sex, seasonality and residence by municipal level of socioeconomic development.”  And the estimates “also considered Puerto Rico’s consistently high emigration during the prior decade and dramatic population displacement after the hurricane.” Such complexity adds uncertainty.

The August 28 GWU report claimed to be “the first to use actual death certificates and other mortality data in order to estimate a more precise mortality count due to Hurricane Maria.” On the contrary, an earlier Aug 2 study in the Journal of the American Medical Association, by professors from Penn State and the University of Texas, had already used death certificate data to (as Ms. Barclay recommended) “count all the deaths in the time since the event and then compare that number to the average number of deaths in the same period from previous years.” Yet that ideal method found the number of excess deaths was 1,139 from September through December of last year.  As the Table from that paper shows, “excess deaths” means the number above the 2010-2016 average.  Since 90% of these atypical deaths happened in September and October, it appears quite plausible to attribute most of them to Hurricane Maria.  That is consistent with five previous credible estimates of  Puerto Rican deaths due to Maria, which, as a Washington Post fact checker noted in June were “all … roughly around 1,000 deaths.” 

Puerto Rico deaths by month

The rationale for the study’s novel choice of a six-month time frame was to find out if things are getting better.  But the more months pass after the disaster, the more arbitrary it appears to attribute deaths to the disaster, since an estimated 77% of those who died were seniors.

In marked contrast to the JAMA paper (where 90% of the deaths happened near the time of the hurricane) only 42.7% of the GWU study’s simulated 2,975 deaths occurred in September and October of 2017.  Another 27.8% occurred in November and December, and 29.5% occurred in January and February of 2018.  That timing seems counterintuitive and implausible, suggesting the September storm has lately been becoming more fatal rather than less.  

To attribute deaths over a six-month period to the hurricane per se is inherently difficult and subjective. What the Milken report calls “a failing health system “ and “multiple cascading failures in critical infrastructure” (telecom and power) may largely reflect negligence by Commonwealth or city governments, notably the island’s mismanaged government-owned electric utility, Prepa

To attribute the estimated 6-month deaths to FEMA, as some have, is even less believable. By August 8, FEMA reported it had awarded “more than $3 billion in Public Assistance funds … to the government of Puerto Rico and municipalities” for Hurricane María-related costs. “This is a massive job and it has taken a massive effort by everybody: the Government of Puerto and the municipalities, federal agencies, voluntary and faith-based organizations and the private sector,” said Federal Coordinating Officer Michael Byrne. 

The questionable 2,975 GWU estimate of hurricane-related deaths, like the unbelievable 4,645 Harvard estimate before it, is being widely misused as a criticism of emergency relief efforts by FEMA and numerous private charities, rather than to either the sheer magnitude of destruction to an isolated island, or to any shortcomings of local Puerto Rican efforts. 

In short, an actual 4-month count closer to 1,100 for above-average Puerto Rican deaths in the wake of Maria appears much more transparent and statistically relevant than the 6-month statistical simulation of 2,975 now being used.