It now seems appropriate to reassess some of the statements made in the wake of the tragic disaster that struck Puerto Rico in September 2017. First, the deaths of Americans (contrary to Mr. Trump’s apparent claim, they were not Americans) were much higher than some commentators claimed. Second, arguments that economic policies undertaken in 2016 and 2017 (i.e. austerity measures associated with PROMESA) caused more deaths than Hurricane Maria are incorrect. Finally, the economic challenges that existed before the hurricane hit – including insufficient tax revenue – remain, even as the economy has rebounded.
The impact on excess mortality
First of all, let us recall that certain commentators affirmed, following the hurricane, that no more than 200-400 deaths have occurred. Even after being educated on the significance of the data, some continued to provide unrealistic estimates.
DeSandberg, et al. (July 2019) in Epidemiology:
Compare now with shoot-from-the-hip reviews on Econbrowser, like this one from Steven Kopits on 05/31/2018:
Excess deaths in PR up to the end of the year, those recorded by the Bureau of Statistics, were just 654. Most of them occurred in the last ten days of September and all the month of October. Although power outages there were exacerbated by state ownership of PR utilities, much of the excess fatalities would likely have occurred regardless, given the terrain and the strength of the hurricane. So maybe 300-400 of the excess deaths would have occurred regardless of what action anyone could have taken to fix the power supply. The remainder can be attributed primarily to public ownership of the electric utility.
I note that excess mortality halved in December. Thus, the data suggests that the hurricane hastened the death of sick and dying people, rather than killing them outright. I would expect the excess deaths over a one-year horizon (until October 1, 2018, for example) to total maybe 200-400. Still a notable number, but certainly not 4,600. [emphasis added-MDC]
Updated analysis (6/4) by Mr. Kopits concludes, even with updated data:
Thus, the year-end additional death toll of 1,400 can be treated as a firm figure in practice.
I think “firm” is an adjective to avoid in these situations. Here is a chart showing selected estimates, from this Article from April 2019which includes some of the previous estimates.
Figure 1: Cumulative excess deaths from September 2017, for simple temporal dummies OLS model (blue), population-fitted OLS model (green), and population-fitted quantile regression model (red), Milken Institute point estimate (black square ) and 95% confidence interval (grey +), Santos-Lozada, Howard letter (chartreuse triangle), Cruz-Cano and Mead (pink squares), Kopits (teal triangle). Not shown: Estimate of Kopits 300-400 for October 2018. Source: Author’s calculations, Milken Institute (2018), Santos-Lozada and Howard (2018), Cruz-Cano and Mead (2019)and Kopits (2018).
Austerity has killed more than the hurricane?
What about the arguments that the austerity measures associated with PROMESA caused more excess mortality than the hurricane [extensive argument here] . In order to assess this argument, first consider mortality data through February 2018.
Figure 2: Mortality by month (blue). Gray indicates the period in the sample; orange shading indicates Hurricane Maria and the post-hurricane period; dotted line to PROMESA legislation. Source: Santos-Lozada and Howard, 2017June publication of civil status data.
Second, now consider constructing the counterfactual not incorporating austerity measures before and after the implementation of PROMESA (legislation passed July 2016, control effective October 2016). I achieve this by estimating two equations: (1) a simple average over the period 2010-2015, and (2) a log-log OLS regression specification incorporating population estimates (as well as a dummy variable for October 2014) . 2016 seems an appropriate breaking point for austerity given Brad Setser’s Discussion on Puerto Rican Finances. These specifications are discussed in this post. I show in Figure 3 the implied excess mortality figures.
Figure 3: Excess mortality by month calculated using 2010-15 averages (blue) and adjusted population using 2010-15 sample (red), population is IMF cubic interpolation World Economic Outlook database data. Gray indicates the period in the sample; orange shading indicates Hurricane Maria and the post-hurricane period; dotted line to PROMESA legislation. Source: Santos-Lozada and Howard, 2017June publication of civil status data, IMF WEO April 2018 databaseand the author’s calculations.
Note that in neither case are most of the pre-Maria deviations statistically significant at 10% msl. In other words, one could generally not reject the null hypothesis of no austerity-induced excess mortality, before Maria.
Third, it is instructive to consider the excess mortality from 2016M01-2017M08, and how it compares to the excess mortality from 2017M09-2018M02. Assuming no population change between 2016 and 2017, then the cumulative death estimate (“avg ’10-’16”) is shown as a red line, indicating an impact minimal austerity.
Figure 4: Cumulative excess mortality by month using population adjustment specification (blue) using IMF cubic interpolation World Economic Outlook database data and using the 2010-15 average (red). Orange shading indicates Hurricane Maria and the post-hurricane period; dotted line to PROMESA legislation. Source: Santos-Lozada and Howard, 2017June publication of civil status data, IMF WEO April 2018 databaseand the author’s calculations.
However, the most realistic assessment relies on adjusting the counterfactual for the population. This leads to the blue line, labeled “log-log”), apparently verifying the proposition that the excess deaths began before the hurricane made landfall. However, interestingly, neither approach directly contradicts the fact that most of the excess mortality since 2016M01 is due to the impact of Hurricane Maria.
I conclude that with the help of statistical analysis, the inference that the excessive number of deaths due to pre- and post-Maria austerity exceeds that of the consequences of Hurricane Maria is extremely fragile.
To support this view, I also find that, using the data on power grid outages reported in Shermeyer (2018)the excess mortality calculated using a population adjustment corresponds very closely to outage data, derived either from PERPA (the utility) or the Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (NPP VIIRS ).
Figure 5: Excess mortality by month using population adjustment specification (black) using IMF cubic interpolation World Economic Outlook database data and blackouts as a proportion of total, PREPA (pink) and VIIRS as shown in Shermeyer (2018). Orange shading indicates Hurricane Maria and the post-hurricane period; dotted line to PROMESA legislation. Source: Santos-Lozada and Howard, 2017June publication of civil status data, IMF WEO April 2018 databasepersonal communication from Jacob Shermeyerand the author’s calculations.
Regressed over the period 2007M04-2018M02, the slope coefficient on the outage is 826 using Puerto Rico Electric Power Authority (PREPA) data and 950 using VIIRS data, both statistically significant using errors robust HAC, with adjusted R2 = 0.83 and 0.72, respectively. This means 639-690 excess mortality attributable to power outages (and correlates) in October, for example. (The excess mortality could be due directly to power outages, or to communication failures and water service failures correlated with power outages.)
How bad was the federal response to Maria in Puerto Rico?
Skipping the paper towel optics, multiple reports (e.g., BIG from DHS) have a document exactly how poorly executed FEMA. Additionally, academic analyzes have documented the differential response that occurred over the same period – i.e. much larger responses to hurricanes making landfall in Texas and Florida.
After the catastrophic failure of the disaster response, the Puerto Rican economy recovered to pre-hurricane levels, only to be struck down by the Covid pandemic. Figure 6 below shows the GDP and economic activity index.
Figure 6: GDP in mn Ch2012$ (blue bar, left logarithmic scale) and economic activity index (EDB-EAI) sa (blue, right logarithmic scale). The NBER has defined peak-to-trough recession dates as shaded. Source: BE A, Economic Development Bank of Puerto Ricoand NBER.
The broadest measure of activity – GDP – is only reported through 2020, while the economic activity index – based on employment, electricity production, gasoline consumption and cement sales – which extends to August 2022 only measures part of the economy. As noted, while there is a rebound in (measured) economic activity, it has slowed in recent months. This pattern appears in civilian employment (but not in the establishment series).
Figure 7: Non-farm wage employment in Puerto Rico (blue), private non-farm wage employment (tan), civilian employment (green), all in thousands, sa. The NBER has defined peak-to-trough recession dates as shaded. Source: BLS and NBER.
Federal assistance associated with the pandemic is ending. This fiscal slowdown comes on top of the main long-term challenges of an uncompetitive economy (partly due to shipping and other transportation regulations) and a large public debt mired in restructuring (see SCR). See also 2017 Economic Analysis by Gregory Makoff and Brad Setser.
Some of these challenges are intractable, while others could be solved relatively easily. For example, a waiver of the Jones Act for Puerto Rico would, by some estimates, lead to lower prices of $1 billionwhich is substantial for an economy measured at around $103 billion in 2020.