In an upcoming American Greatness article I will be discussing the contentious issue of which political party is responsible for the spread of the COVID-19 disease and which party is suffering more deaths from the disease. That article is based on a calculation which is presented in my github page here.
The purpose of this post is to provide somewhat more details on the technical calculation short of simply providing the code and data (which is the purpose of the github page). In the end, the github page is the final source on what is in these calculations. I am nevertheless happy to answer any technical questions posed to this post.
The first question which I ask is whether more Republicans or more Democrats are dying/have died from the disease.*
As a proxy (and perhaps a more interesting sample) for Republican versus Democrat I used instead the number of Trump voters (by county) in 2016 versus the number of Clinton voters (and accounting as well for the number of non-voters).
To calculate this I took the data on the daily U.S. deaths due to COVID-19 as compiled in the Johns Hopkins University Center for Systems Science and Engineering database (henceforth called the JHU data). This data is compiled by day, since January 22, 2020, and by county (3145 in the United States).
The election data by county is available in many places. I used the dataset found here.
Additionally, in a second half of the analysis described below I also included data on the party which holds the governorship in each state. That dataset is included in my github page for convenience but of course it can be found anywhere.
Finally, I needed a dataset that gave the population in 2016 of each county. This was needed in order to determine the number of children in the county and thus the number of eligible voters who did not vote in 2016. That dataset is also included in the github page. The original source is the U.S. Census data which is here.
Here I present three figures based on these calculations.
Figure 1
The number of deaths on a given day in a given county is multiplied by the fraction of Trump voters, the fraction of Clinton voters, and the fraction of voting age non-voters in the given county to obtain the number of Trump voter/Clinton voter/non-voter deaths on that day in that county. These fractions are computed as percentages among all eligible voters (not all citizens). Thus I specifically exclude citizens under the age of 18 (who number approximately 100 million in the United States).
To explain this: if children are included in the analysis the result is that the number of Trump voters and Clinton voters (indeed, the number of voters) become a significantly smaller portion of the total population. In 2016 roughly 55% of the eligible voters cast a ballot. But only 38% of the total citizens cast a ballot. Since in fact the number of deaths among children from COVID-19 is thankfully very small it is sensible to assume for the precision of this argument that is it zero and that all deaths occur within the voting public.
Once the number of deaths per county per day is computed they are simply summed over counties on each day, giving a number of deaths of Trump voters, Clinton voters and non-voters nationwide. The result is figure 1.
Figure 2
If we assume that the vote proportion in a given county is indicative of the overall support for the candidates – even among citizens who did not cast a ballot – then we can calculate the number of deaths of Clinton supporters as compared to Trump supporters. This is shown in figure 2 where all the deaths in a given county are attributed to one or the other candidate’s supporters. The absolute numbers are thus higher for each group, but the ratios of the “D” to “R” numbers does not change from figure 1.
Figure 3
As described in my forthcoming article, there is considerable political interest in assigning blame to one party or another for the 185,000+ COVID-19 deaths. The analysis presented here gives some evidence for whom is to blame, but it is only very sparse evidence and cannot be relied on to produce an absolute attribution of blame.
In other words, even though figures 1 & 2 show that more Clinton voters and probably more Clinton supporters have died from COVID-19 than voters and supporters of Trump, that does not prove that Democrats are at fault. It is obvious that under any political policies many people would have died. It is also obvious that the disease entered the United States through coastal areas mainly and that spreading occurred before preventative measures were in place to stop it.
So, do figures 1&2 provide any information on culpability? Of course they do. But they are only one small piece of the puzzle. A piece that could weigh very little in comparison to other factors.
That said, I attempted to extend the calculation above to more specifically target the effect of political control on the mortality rate. ** In this final calculation of this post I defined a political control variable for each county by simply whether the county had been won by Trump or Clinton. Further, i defined state political control simply by which party controlled the governorship of the state. These are obviously both very crude approximations requiring refinement (e.g. who controls the state legislatures?), so of course the results need to be taken with a grain of salt.
That said, if one party had both county and state political control, then all of the deaths in that county were attributed to that party. If control was split between county and state then half of the deaths were attributed to Republicans and half to Democrats.
Note that this is at least somewhat more informative than simply counting the number of deaths in terms of the governor’s party alone. For example, Massachusetts has a Republican governor (my friend Charlie Baker) but it is a highly liberal/Democratic state and they also had a substantial number of COVID-19 deaths.
With all of those caveats and explanations, the result of this calculation is shown in figure 3.
Here the data indicate that an even larger percentage of deaths in the first wave were in democratic controlled jurisdictions as compared to figure 2 which splits the data by Trump supporter versus Clinton supporter. However, the second wave in early August is more evenly distributed and actually has a slight bias toward Republican-controlled states and counties.
* the motivation for this investigation is an interview (March 2020, MSNBC) with and subsequent articles by Washington Post columnist Jennifer Rubin who claimed that due to denial of the seriousness of COVID-19 and concomitant dangerous rapid-opening policies many more Republicans would die from the disease than Democrats.
** Note that while statistics are available for mortality, number of cases, hospitalizations and various other disease metrics I have limited my study simply to the number of deaths. This is in part because it is less ambiguous than number of cases (how many cases depends, for example, on how many tests are done). It is not completely unambiguous, of course, since dying of COVID-19 and dying with COVID-19 have notoriously been conflated in places at times.