Did close to 3,000 Puerto Ricans die in Hurricane Maria?

Question

CNN reports:

Earlier this month, [Puerto Rico]'s governor formally raised the death toll from Hurricane Maria to an estimated 2,975 from 64 following a study conducted by researchers at The George Washington University.

It also reports that President Trump just tweeted this:

3000 people did not die in the two hurricanes that hit Puerto Rico. When I left the Island, AFTER the storm had hit, they had anywhere from 6 to 18 deaths. As time went by it did not go up by much. Then, a long time later, they started to report really large numbers, like 3000. This was done by the Democrats in order to make me look as bad as possible when I was successfully raising Billions of Dollars to help rebuild Puerto Rico. If a person died for any reason, like old age, just add them onto the list. Bad politics. I love Puerto Rico!

Is President Trump correct that the official death toll figure is grossly exaggerated?

Answer 1

No, President Trump is not correct that the official death toll has been exaggerated. While he's right that 3,000 people did not die in Hurricane Maria, there is evidence that nearly 3,000 people likely did die as a result of the storm and this is the statistic that is currently being reported/discussed.

The death toll of 2,975 is from this report by the School of Public Health at George Washington University. They are not claiming that nearly 3,000 people died during the hurricane, but that there were nearly 3,000 more deaths than expected in the months following the hurricane.

In other words, if past mortality rates predicted that 100 people would die from age-related issues in January and 125 people actually died, then the 25 people that were added to the list of deaths were likely caused by the hurricane.

As summarized in the report's methodology section:

We implemented the project as three studies, each with specific yet complementary methodologies. Our excess mortality study analyzed past mortality patterns (mortality registration and population census data from 2010 to 2017) in order to predict the expected mortality if Hurricane María had not occurred (predicted mortality) and compare this figure to the actual deaths that occurred (observed mortality).The difference between those two numbers is the estimate of excess mortality due to the hurricane.

...

Total excess mortality post-hurricane using the migration displacement scenario is estimated to be 2,975 (95% CI: 2,658-3,290) for the total study period of September 2017 through February 2018.

This number was estimated with a high amount of confidence by comparing expected population changes had Maria not occurred with actual population changes:

To perform this analysis, we obtained vital registration mortality data including deaths by age, sex and municipality of residence from the Puerto Rico Puerto Rico Vital Statistics Registry (PRVSR) for the period July 1, 2010 to February 28, 2018. We derived baseline estimates of population size in each month from annual census estimates of population size by age, sex and municipality of residence. Cumulative monthly population displacement after the storm in each month was estimated using Bureau of Transportation Statistics (BTS) data on monthly net domestic migration provided by the Puerto Rico Institute of Statistics and a survey of airline travelers provided by the Puerto Rico Planning Board (Planning Board 2018).

Using the ~8 years of past population statistics, the researchers were able to estimate the expected number of deaths in the few months after the hurricane with the actual number of deaths:

---

As a side note, the purpose of the report wasn't necessarily to provide a fully accurate death count. Rather, the purpose was to show why this exact situation, where death estimates are so varied and potentially inaccurate, is dangerous and why it shouldn't even happen in the first place. It was meant to act as a policy guide for improvements to Puerto Rican mortality assessments and communication during natural disasters, so that a disaster of this magnitude doesn't happen again:

RECOMMENDATIONS ON MORTALITY SURVEILLANCE FOR NATURAL DISASTERS

I. Strategic Objectives

To have a reliable and resilient institutional mortality surveillance process that provides trustworthy and accurate evidence during natural disasters to: Establish the magnitude of the impact of the disaster, identify areas and groups of highest risk, monitor the performance of public health protection and prevention, and inform policy-making and program implementation.

Answer 2

Other answers have discussed the statistical methodology of studies used to estimate the death tolls, but let me approach this from a different perspective: Do we have reliable on-the-ground reporting of significant deaths (in the hundreds or even thousands) during and immediately after the hurricane?

The answer: YES

Miami Herald: Hurricane Maria’s death toll in Puerto Rico is higher than official count, experts say (September 2017, 1 week after the storm):

The dead are at the hospital morgues, which are at capacity and in remote places where the government has yet to go. In many cases, families are unaware of the deaths....

“We’re finding dead people, people who have been buried, [people] have made common graves,” Rodríguez-Mercado told CPI. “We’ve been told people have buried their family members because they’re in places that have yet to be reached.”

BuzzFeed News: 911 People Died After The Hurricane — Of "Natural Causes" (October 2017, 1 month after the storm):

The Puerto Rican government told BuzzFeed News on Friday that it has allowed 911 bodies to be cremated since Hurricane Maria made landfall, and that not one of them were physically examined by a government medical examiner to determine if it should be included in the official death toll....

Without guidance, different funeral home and crematorium directors told BuzzFeed News they had vastly different ideas of what they considered hurricane-related deaths. Some said they counted heart attacks and people who died for lack of oxygen because there was no power, while others said they counted those as "natural deaths."

Disaster death toll experts told BuzzFeed News that these cases should definitely be counted, and that they were in the aftermath of Hurricane Katrina.

CNN: Survey of 112 Puerto Rican funeral homes (November 2017):

To check the accuracy of the Puerto Rican government's figures, we called nearly every funeral home in Puerto Rico.... Some funeral homes did not answer our calls, and several declined to provide data. CNN was able to collect responses from 112 of the island's funeral homes. That's about half the total number in Puerto Rico, according to Eduardo Cardona, director of the Puerto Rico Association of Funeral Home Directors.

Those funeral homes identified 499 deaths in the month after the storm -- September 20 to October 19 -- which they say were related to Hurricane Maria and its aftermath. That's nine times the official death toll. And, again, it represents only about half of funeral homes.

CNN: It's been almost six months since Hurricane Maria, and Puerto Ricans are still dying (January 2018):

Some of them died during the storm. A mudslide in Utuado, Puerto Rico, killed two "bedridden" sisters. Another person drowned in Toa Baja. But the aftermath of Hurricane Maria appears to have been most deadly....

In Corozal, farther into the mountains, Victor Manuel Belen Santiago wept as he told me that his mother, Zoraida Santiago Torres, 58, had saved his life by helping him kick drug addiction.

Their home was destroyed by the storm, and Belen Santiago rebuilt it by hand, puzzling scraps of the roof and walls together like a reassembled house of cards. But he couldn't restore the power his mother needed to run an oxygen machine. She died on February 13, he said, after getting fluid in her lungs that could not be cleared. Her death certificate lists organ failure and a bacterial infection among the causes of death, along with chronic liver disease.

NBC: Puerto Ricans knew the official Hurricane Maria death toll was fake. We saw too many dead to believe it. (May 2018):

When funeral directors started telling people that they were burying way more bodies than usual, or when our family members told us about their neighbors dying in still-darkened rooms, or being buried outside their homes, we knew that the official death toll was much higher than the 64 people the government had eventually admitted to.

When we heard the stories of people having no refrigeration for their insulin, that dialysis machines weren’t operational or that hospitals were still in the dark but had people on life support, we knew that it wasn't some small counting error.

So Puerto Ricans are not suddenly shocked by the Harvard study published this week estimating that a total of 4,645 excess deaths occurred between September 20 through December 31, 2017, because the proof was already there months ago.

But almost nobody else wanted to look for it.

Washington Post: I saw what Maria did to Puerto Rico’s hospitals. The death toll is no surprise (June 2018):

We found an ongoing human disaster during the months of September, October and November, when we made daily visits to more than 60 senior homes and independent-living facilities. At one high-rise, within walking distance of a hospital, residents were trapped on the upper floors because the backup generator had failed and they couldn’t walk down the stairs. Some folks I spoke with didn’t remember when they had last eaten. There was no potable water because the pump to get it up to the apartments also depended on electricity. Dialysis patients hadn’t been treated in days; diabetics couldn’t refrigerate their insulin....

The hospital couldn’t reach ambulance services for other residents who needed urgent care, because they had no means of communication. A nearby police car was similarly incommunicado. Four people were eventually evacuated from the building in municipal ambulances — stretchers and wheelchairs were carried down cramped stairwells — but I overheard the EMTs wondering where they would take the dialysis patients; they didn’t know if any centers would be open.

With conditions that bad at one of so many buildings, of course more than 64 people died. What is surprising is that the government has so far been unwilling to admit it.

While we can't extrapolate specific numbers from these stories, it is absolutely inconceivable that only a handful of people died due to Hurricane Maria. In my opinion, 3,000 deaths is absolutely plausible, if not understated.

Some further references:

• CNN: How Puerto Rico's death toll climbed from 64 to 2,975 in Hurricane Maria
• 538: Media Reports About The Death Toll In Puerto Rico Are Needlessly Confusing
• Snopes: Did All of the Patients in a Puerto Rican Intensive Care Unit Die?. This claim is rated as "unproven", but the article contains some other useful quotes and links.

Answer 3

There are a number of different institutions which try to quantify death tolls and other damages after natural disasters. The definitions and methodology differ between estimates, and numbers vary depending on which method you choose.

For example, if some people are injured badly during a hurricane and die months later in hospital from their injuries, should this be included in the toll?

A special article in the New England Journal of Medicine study, Mortality in Puerto Rico after Hurricane Maria was published in July. They used a survey methodology to make an estimate of the death toll. They did not count the deaths recorded by medical services, but surveyed over 3,000 people in Puerto Rico about damage and deaths. From the answers they calculated that the death toll was 4645 with a 95% confidence interval of 793 - 8498.

From the survey data, we estimated a mortality rate of 14.3 deaths (95% confidence interval [CI], 9.8 to 18.9) per 1000 persons from September 20 through December 31, 2017. This rate yielded a total of 4645 excess deaths during this period (95% CI, 793 to 8498), equivalent to a 62% increase in the mortality rate as compared with the same period in 2016. However, this number is likely to be an underestimate because of survivor bias. The mortality rate remained high through the end of December 2017, and one third of the deaths were attributed to delayed or interrupted health care. Hurricane-related migration was substantial.

According to these numbers, a death toll of 3,000 people is likely not an exaggeration.

To answer your question: Donald Trump's statement is most likely not accurate.

Answer 4

Just corroborating what was pointed out by @redleo85 and @Giter above by including a particularly poignant plot using the data from the NEJM paper to illustrate what "excess mortality" looks like graphically (credit to @yanivbrandvain on Twitter for posting this):

The open-source repo for the NEJM paper is here:

https://github.com/c2-d2/pr_mort_official

We can quibble for ages about choices of bandwidth, naive technique (the fit at the endpoints shows this was done a tad hastily), confounding/mitigating factors, etc., but the basic idea is to integrate blue curve - average(red curve + green curve) & the massive spike is clear. Please consult the original papers for full details/justification.

Answer 5

What does Trump claim?

• He is confronted with research using mortality data and a household survey showing that several thousand people died in the months after the hurricanes hit; absent other causes these excess deaths are attributed to the hurricanes. On-the-ground observations corroborate the cost of lives.
• His response on twitter (part 1, part 2) is
  • "3000 people did not die in the two hurricanes that hit Puerto Rico"
  • "This was done by the Democrats"
  • "in order to make me look as bad as possible".

In context this must be understood as a refutation of the research, not as a linguistic fight about fine print ("in" vs. "due to"). We must also take into account that we are talking about Presidential twitter messages here, not well thought-out White House press releases. As if to highlight the absurdity, the first message has the maximum length of 280 characters leaving no space for a more exact wording.

Conclusion: What we see is an understandable defense against damaging claims by doubting their validity and accusing the researchers of bias. We are not having a semantic discussion.

Bias

• The study of mortality data by the George Washington University was commissioned by the government of Puerto Rico. The governor, Ricardo Rosselló, is member of the New Progressive Party which is not directly affiliated with the U.S. Democratic Party. In national U.S. politics they side on a case-by case basis with Democrats and Republicans, according to a dissertation mentioned on wikipedia. While the Puerto Rican government has an obvious interest in deflecting blame and inflating numbers, the allegation of Democratic Party bias appears wrong.
• The household survey published in the New England Journal of Medicine lists medical institutions as supporters; the disclosure forms do not list conflicts of interest or financial contributions by third parties. With no discernible connection to the Democratic Party Trump's allegation appears wrong for this study as well.

Conclusion: The allegation of Democratic Party bias appears wrong for the actual research. So the question is whether the research is factually correct or not; and if yes, how it is possible to arrive at these vastly different numbers.

Are the claims of thousands of deaths scientifically defensible?

Trump does not make any factual or methodological claims regarding the science of the studies beyond quoting the number of deaths reported immediately after the hurricane. I have not seen the studies criticised by third parties either.

Conclusion: The studies appear conducted properly according to scientific standards and the results should be valid.

How can we explain the discrepancy?

Interestingly, today a disaster researcher, Scott G. Knowles, discusses exactly this question in the New York Times. He quotes the George Washington University study that estimates 2,975 fatalities.

His explanation of the discrepancy with Trump's assertion of no more than a few dozen victims focuses on the developments in disaster research over the last decades. Until about the 1960s disasters were understood as events with defined limits in time and space, and consequently only damage in this time and space was considered. But as the result of ongoing research,

by the 1990s there was an emerging consensus that the historical background and long-term aftereffects of disasters were just as important, if not more so, than the event itself, because a disaster is really an interconnected chain of occurrences.

This is actually the same paradigm shift that happened in other social sciences like history towards what one could call a "systemic" view. Historic events are not isolated and can be properly understood only in context.

Knowles continues

We will never understand why some people are vulnerable and others aren’t if the focus is only on a disaster as an event. Nor will we comprehend why some communities recover quickly while others languish, nor grasp the debilitating consequences of post-traumatic stress disorder. So, too, we will be unable to comprehend the burden of care for future generations, or past victims. Event thinking is amnesiac; worse, it inhibits planning for the disasters to come.

But if one’s interest is in playing down the impact of disaster, then event thinking is the ticket. You may avoid learning anything significant about the causes and consequences of what happened, but you can get back to the business of building, even if it is in dangerous places.

Knowles then outlines how the systemic understanding of disasters, "based in scientific reality", facilitates better preparation and better handling of the aftermath.

He provides a direct answer to President Trump's point of view:

The clear counterpoint is to acknowledge that a disaster is a process that reveals our values as a society. A disaster’s beginning and end might be obscure, but that doesn’t excuse us from figuring out the causes and true costs.

Illustration: Compare with a shipwreck

I would like to compare Trump's claim to that of an imagined designer of the RMS Titanic who defends himself thusly: "This bogus claim that 1500 people died in the sinking of the Titanic is grossly exaggerated in order to make me look bad. When I saw the ship go down there were fewer than 1000 people on board. I don't know how they arrive at these inflated numbers." The 1500 number was, of course, calculated from the difference in the ship's passenger manifest vs. the number of rescued people. The logical conclusion was that the ones not left behind in the ship drowned in the Arctic Sea, technically after the actual event.

This comparison may serve as an illustration why it is relevant to consider a disaster "an interconnected chain of occurrences" which started before the actual event (manufacturing flaws, lack of emergency exercises) and continued afterwards: people drowned in the cold water and potentially died of pneumonia hours and weeks later. With this more systemic view we are able to draw valuable conclusions which are lost when we focus just on the actual sinking: People embedded in groups (families, friends) have a higher survival chance; people with health insurance have a better chance to survive illness from exposure later on; adequate safety procedures improve the overall outcome of collisions with icebergs, etc.

It's obvious that nobody would make the claim that the 500 drowned people did not die in the sinking of the Titanic. We are probably making this attribution without much thought because the connection in time, space and cause is very tight. In a hurricane the connections are not so obvious: The aftermath is playing out over thousands of square miles, months afterwards, and in ways whose relationship to the storm is not as immediately obvious. But the structural parallels are striking. Focusing on the sinking or the hurricane as an isolated event deprives the decision makers of the knowledge base for adequate preparation and response.

Bottom line:

From the perspective as laid out by Knowles, a narrow and obsolete traditional understanding of "disaster" excludes the true cost in lives.

Answer 6

For other hurricanes, the National Oceanic and Atmospheric Administration (NOAA) uses the definition "deaths as a direct result of the event" - e.g. the deaths caused by drowning, etc. occurring immediately during the event.

The official government death toll includes only deaths in which documentation of “hurricane-related” as the cause of death appears on the individual’s death certificate and does not account for indirect deaths, including from infectious disease outbreaks or lack of services (such as electricity, water, and medical care)

(Santos-Lozada & Howard, 2018)

The 3,000 figure is Puerto Rican death rate increase from any cause for 6 months following, according to Ascertainment of the Excess Mortality from Hurricane Maria in Puerto Rico.

Total excess mortality post-hurricane using the migration displacement scenario is estimated to be 2,975 (95% CI: 2,658-3,290) for the total study period of September 2017 through February 2018

While 3,000 may be an accurate figure in its own category, it isn’t the standard NOAA uses in calculating deaths from hurricanes.

Given this definition, the most accepted figures for deaths by Maria as defined by NOAA are less than 100. This isn't to downplay the loss, just making sure we are comparing the same things.

Blake, E. S., Landsea, C. W., & Gibney, E. J. (n.d.). NOAA Technical Memorandum NWS NHC-6, 49.

Santos-Lozada, A. R., & Howard, J. T. (2018). Use of Death Counts From Vital Statistics to Calculate Excess Deaths in Puerto Rico Following Hurricane Maria. JAMA. https://doi.org/10.1001/jama.2018.10929

Answer 7

It seems the answer to this question has turned toward an argument about whether it is a new standard to include indirect deaths when discussing deaths caused by a hurricane. Some comments have said that reports are including indirect deaths for Hurricane Maria in order to make Trump look bad, suggesting death reports are a partisan issue where pro-Trump people have one fact and anti-Trump people have another fact. These comments are wrong. This is not a new standard. The inclusion of indirect deaths in fatality reports from hurricanes occurred for previous hurricanes. It is not a partisan decision to include them. As you can see below, death reports for prior major hurricanes include indirect deaths.

Perhaps this shouldn't be included as an answer to this question, but the details seem to be easy to lose in comments. I'm open to suggestions for how to appropriately contribute this information to an answer re: the claim "Is President Trump correct that the official death toll figure is grossly exaggerated?" Perhaps it's better as an answer to a different question: was the reporting standard changed for Hurricane Maria (in order to make Trump look bad)?

Hurricane Katrina's death toll:

According to the National Hurricane Center, 1,836 fatalities can be attributed to the storm... many of the deaths are indirect

Hurricane Harvey's death toll

Harvey caused at least 107 confirmed deaths: 1 in Guyana, and 106 in the United States.

From the side bar:

68 direct, 39 indirect

Hurricane Sandy's death toll

At least 233 people were killed across the United States, the Caribbean, and Canada, as a result of the storm

From the table:

106 direct 87 indirect

Excess mortality

It might be useful here to explain the excess mortality statistic the official death toll is based off. This is the statistic (2,975 (95% CI: 2,658-3,290)) the claim refers to as a gross exaggeration.

An excess mortality statistic compares an expected number of deaths (including a measure of the uncertainty in that expected number, based on the variability) with an observed number of deaths. This involves examining a set of data to predict the number of deaths in a counterfactual scenario -- one where an event didn't occur.

In the GW analysis, the prediction model is based on data on the mortality rate and its variability over a period of 7 years. In addition to crude mortality rates, it analyzes rates for subgroups (age, sex, and SES). It is important to recognize that the prediction model includes all of the deaths over those 7 years, including 7 seasons of tropical storms that affected Puerto Rico. During this period, people died as a result of the residual effect of Bonnie and Earl, as well as all the entire impact of Otto, Emily, Irene, Bertha, Cristobal, and other unnamed storms. In addition to including previous storm related deaths, the prediction model includes the mortality impact of Puerto Rico's many public health challenges, and a particularly bad year for the Caribbean in general.

The excess mortality, then, is the number of deaths above and beyond what is predicted. It is specifically a measure of how much worse things were after Hurricane Maria than the previous 7 years would predict. When politicians and pundits state that the excess mortality includes deaths from all causes as a way to call the measure into question, or say that Puerto Rico is in the middle of a financial crisis, or has infrastructure problems, so those people would have died anyway, they are entirely missing the point of an excess mortality analysis. In a Caribbean island that has had its share of hurricanes and tropical storms in the period used for the prediction model, the model expects storms during hurricane season and deaths due to those storms. You subtract actual deaths (all causes) from predicted deaths (all causes). Cancer deaths are in both numbers. Storm deaths are in both numbers. The statistic is the difference.

Answer 8

An author of the report finding "2,975" has provided some additional information and commentary in the Washington Post.

In September 2017, when Puerto Rico recorded a total of 2,906 deaths,

(So with the hurricane being present in Puerto Rico from ~19-21 September, in the entire month including before the hurricane 2,906 people died of all causes.)

..we found there was an excess of 574 deaths above what would have been expected in a year without the storm. The death toll continued to mount every day, with an excess of 697 deaths in October, 347 in November, 479 in December, 558 in January, and 320 in February, for a total of 2,975.

Throughout that time, researchers produced other estimates. The government of Puerto Rico came up with a figure of 68 excess deaths through October 2017; the New York Times, 1,052 through October 2017; a team at Harvard University, 4,645 through December 2017; and researchers at Penn State University, 1,139 through December. Of these, only the estimate by the government of Puerto Rico involved the examination of individual deaths to determine if the hurricane had caused them. The Harvard study, based on a household survey, was later found (by Milken Institute SPH researchers) to have overestimated the number of deaths because they did not adjust the household death reports for household size. When analyzed correctly, their study produces similar numbers.

Separately, the government of Puerto Rico compared the deaths in September-December of 2017 to the same period in 2016 and found 1397 excess deaths. It also compared deaths to the average of years 2013-2016 and found 1427 excess deaths.

However, the George Washington Univ. report finding "2975" comes up with a higher number for two reasons. Instead of comparing to the actual number of deaths in previous years, correction is made for the population of Puerto Rico decreasing from migration after the hurricane. Secondly, the period of study is six months instead of four months. Unfortunately, depending upon the age and health of the people leaving Puerto Rico, it would be unjustified to assume random exodus of the island, but the report does not detail the methodology used to correct for migration.

Instead of providing the actual methods used to arrive at 2975 the report says:

For more details on the methodology, data and programs used in the excess mortality calculations, these will be made available online at: http:// prstudy.publichealth.gwu.edu/

But nothing has appeared so far.

In conclusion, it is straightforward that excess deaths due to the combination of hurricanes Irma and Maria, both of which struck in September of 2017, is at least in the 1397 range. But more than doubling this due to correction for migration is not well supported without further explanation of methodology.