As the pandemic began reaching epic proportions at the beginning of April, I made a wager that the toll from the coronavirus will follow the all-too-familiar pattern of fortune on the planet:
The one thing we can be sure about is that the brunt of the pandemic will be borne by the poor nations of the world. During the 1918 pandemic, half the fatalities were in India. This time around, Africa will likely fare even worse because of a major divergence within developing nations. I wager that the best predictors of the distribution of the death toll across nations will be state capacity and health status. The present distributions reflects the intensity of global networks — the tripolar core of the world economy is the most tightly integrated and therefore the earliest to be exposed. But this is merely the beginning. This thing is going to go everywhere; in multiple cycles. The ultimate death toll will obey Matthew’s Law — those who are the worst off today will pay the steepest price. Why is our world so unfair??
Although there is still plenty of time for me to be proven right, it has been a while since the novel coronavirus has reached the four corners of the earth and Matthew’s Law is still nowhere to be seen. In fact, quite the opposite. Not only is the class gradient upside-down within the United States, it is increasingly obvious that so is the international class gradient. The per capita death toll from the novel coronavirus is positively correlated with all our best measures of everyday living standards. Life expectancy:
Sociodemographic index:
The puzzle of the coronavirus death toll is that it is positively correlated with every measure of living standards available to us: per capita income (Spearman’s rank correlation: r=0.73, P<0.0001), sociodemographic index (r=0.72, P<0.0001), mean adult height (r=0.66, P<0.0001), life expectancy (r=0.68, P<0.0001), protein intake per capita (r=0.70, P<0.0001), and calorie intake per capita (r=0.76, P<0.0001). Not surprisingly, it is also highly correlated with absolute latitude (r=0.69, P<0.0001). To truly grasp the puzzle, we must note that this is precisely the opposite of what we ought to have expected beforehand. The international cross-section of the coronavirus toll is completely upside-down. The pattern is the opposite of what we should expect!
As I wrote two months ago, we should’ve expected state capacity to be a strong predictor of the death toll from the novel coronavirus. As it turns out, state capacity is strongly but positively correlated with the coronavirus death toll (r=0.71, P<0.0001). In fact, state capacity is positively correlated with the death toll even after controlling for the incidence rate (r=0.39, P=0.0035). And it’s not just state capacity. Controlling for incidence rate, the death toll is still positively correlated with the sociodemographic index (r=0.37, P=0.0092), mean adult height (r=0.39, P=0.0056), protein intake per capita (r=0.34, P=0.0143), calorie intake per capita (r=0.40, P=0.0037) and absolute latitude (r=0.46, P=0.0008), although not significantly so with per capita income (r=0.21, P=0.1366) and life expectancy (r=0.24, P=0.0897). But controlling for incidence rate only makes sense if we want to understand the adequacy of state-society responses. Otherwise, we should not control for it. Both the incidence rate and the death rate from the novel coronavirus are, in fact, explananda that deserve our attention. Table 1 displays Spearman’s correlation coefficients from the international cross-section.
Table 1. The Puzzle of the International Cross-Section of the Coronavirus Toll. |
||||
Deaths per capita | Cases per capita | |||
r | P | r | P | |
GDP per capita | 0.726 | 0.000 | 0.726 | 0.000 |
SDI | 0.669 | 0.000 | 0.722 | 0.000 |
Height | 0.581 | 0.000 | 0.657 | 0.000 |
Life Exp. | 0.655 | 0.000 | 0.675 | 0.000 |
Protein | 0.647 | 0.000 | 0.698 | 0.000 |
Calorie | 0.698 | 0.000 | 0.756 | 0.000 |
Abslat | 0.590 | 0.000 | 0.687 | 0.000 |
Source: Washington Post, GDHx, author’s computations. N=72 nations. Estimates in bold are significant at the 5 percent level. Spearman’s correlation coefficients. |
Regular reader would know that the Policy Tensor is usually quite skeptical of biological reductionism. But in this case we are forced to confront the idea that there may be a genetic reason for the pattern we observe in the international cross section. This is probably why specialists have begun asking themselves the same question. Could it be that genetic factors are in play? That what explains the differential impact of this killer is differential genetic susceptibility to the novel coronavirus?
Let’s test the hypothesis. We compute the pairwise absolute differences in the coronavirus incidence rate and death rate. And then we test whether phylogenetic distance predicts absolute differences in incidence rate or death rate. We do this in three ways. First, we conduct Mantel tests without controls. Second, we compute the slope of phylogenetic distance allowing for random effects by country. Third, we compute the slope of phylogenetic distance (as computed here) controlling for geodesic distance and allowing for random effects by country.
We find that we cannot reject the null of zero correlation between phylogenetic distance and deaths per capita (h=-0.17, P=0.7911) and incidence rate (h=-0.03, P=0.6523). However, with random effects by country, we find very significant gradients for both, with and without controlling for geodesic distance. Table 2 reports our estimates.
Table 2. Gradient Estimates in mixed-effects models for absolute differences in the coronavirus toll. | ||||
Response: Absolute differences in incidence rate. | ||||
Slope | std error | Fstat | P | |
PhyloDist | 0.306 | 0.019 | 256.3 | 0.0000 |
Slope | std error | Fstat | P | |
GeoDist | 0.047 | 0.013 | 12.2 | 0.0005 |
PhyloDist | 0.279 | 0.021 | 183.0 | 0.0000 |
Response: Absolute differences in deaths per capita | ||||
Slope | std error | Fstat | P | |
PhyloDist | 0.365 | 0.028 | 172.4 | 0.0000 |
Slope | std error | Fstat | P | |
GeoDist | 0.105 | 0.019 | 29.2 | 0.0000 |
PhyloDist | 0.304 | 0.030 | 103.7 | 0.0000 |
Source: Washington Post, Spolaore and Wacziarg (2017), author’s computations. Predictors have been recentered and rescaled to have mean 0 and variance 1. Maximum likelihood estimates of dyadic regressions allowing for country random effects. N=5,184 country pairs. GeoDist is computed pairwise for country dyads via the Haversine formula. Estimates in bold are significant at the 5 percent level. |
These results suggest the possibility that differential genetic susceptibility to the novel coronavirus may be responsible for the upside-down world of the international cross-section of the coronavirus toll. A straightforward testable prediction of this hypothesis would be that immigrants from the “right” places sport much better odds ratios than others. That would be a more stringent test than the one we have carried out because we can then condition on practically any feature of importance. For instance, if the odds ratios differed for immigrants in New York and obeyed the pattern detected here, we would have to put much greater credence on the genetic hypothesis of differential susceptibility to the novel coronavirus.
GeoDist is computed from Wuhan?
The estimates in the second table are from dyadic regressions. GeoDist, like PhyloDist, is computed pairwise.