A Gentle Look at the Role of Partisanship in the Spread of the Novel Coronavirus in the United States

The COVID-19 pandemic has plunged the world into a level of chaos and uncertainty unprecedented in this century. This has resulted in analyses and investigations into the possible trajectory that the disease will follow and how this might affect the world socially and economically in the years to come. Epidemiologists and Data Scientists all across the world have been attempting to attack this enormously complicated problem from multiple angles, as they should, and the literature on the various causes and the possible course of the coronavirus has been burgeoning at an astronomical rate.

In this context, 3 of my fellow students at the UC Davis MS in Business Analytics program and I decided to collaborate and participate in the 2020 COVID-19 Analytics Challenge, where we investigated the associative relationships between political leanings and the spread of the virus. We went on to win the first prize for our work. Additional to being a great validation for our conviction that this was a perspective that people genuinely cared about, we were motivated to publish our findings and the thought process that went into our analysis in this blog and a series of others to follow.

While these facts are not in dispute based on what we hear from the media, we decided to take a closer look at the statistical methodology used by the media outlets to come to these conclusions, besides anecdotal evidence. On closer inspection, a large proportion of these seemed to be associative analyses with very few controls. A high negative correlation between the search volume for hand sanitizers or the coronavirus and the percentage of people who voted for Mr. Trump in 2016 seems to make a powerful case at first glance, but without the required controls for potential confounders, the statement has no solid statistical ground to stand on.

The narrative of a Republican incredulity towards the severity of the coronavirus leading to public awareness being low in Republican-leaning states might be countered by a narrative that goes something like this: Republican-leaning states are more sparsely populated, and thus less likely to have many initial cases of the virus, and thus public awareness stayed low due to it not being as visible a threat. Narratives are easily created retrospectively, but notoriously difficult to verify. Our goal is to try and do as much justice to this analysis while keeping an even and unbiased approach to figure out whether the fact that a county leans right or left, has any bearing on the spread of the virus, and if so, then what?

We start first by answering the simplest of questions: after controlling for population density, is there any significant difference in the cases per million between Republican-leaning counties and Democrat-leaning counties?

The question of categorizing a county as Democratic or Republican itself throws up a few challenges worth discussing. A few methods come to mind: a county could be categorized as a Republican state if a plurality of its population voted for Trump in the 2018 midterm elections. Alternatively, a state that had a Republican mayor at the helm could also be construed as a “Republican-leaning” county, as county mayors are largely responsible for setting the baseline of response for the county’s citizens.

We chose to go ahead with the percentage of votes for Trump in the midterm as an indicator of the political leaning of the county. If this was greater than 50% would flag a county as republican, otherwise a Democratic. There is an implicit assumption here that the majority of the county not being Republican implies that it is Democratic, whereas it is quite possible the county may actually back an independent.

We have the midterm voting data for 1,909 counties. Plotting the cases per capita for these counties provided us with a heavily right-skewed distribution, which prompted us to investigate the log(cases per capita) to get a more well-balanced distribution. Next, we removed outliers counties with 0 cases, as these were highly likely to be counties either with meager testing capabilities, poor case reporting or so remotely located that the virus did not spread to these counties. Similarly, we removed counties with extremely high cases per capita, as these were much more likely to be exposed to travelers from Europe and Asia when the virus started spreading(such as New York City), and thus had certain extenuating circumstances that made stopping the spread of cases a much more complicated task. Alternatively, they might be very sparsely populated counties with very few cases (such as Baker County, Georgia with 22 cases but cases per million of 889). After removing such outliers, we had 1,832 counties remaining.

With the log(cases per capita) as our dependent variable, we carried out a linear regression with the Republican flag and log(population density) as our independent variables, the latter acting as a control for our regression. The result showed that a Republican-leaning county actually had ~21% fewer cases per capita, population density held fixed.

Even before we start diving deeper into confounders and mediation effects, it is interesting to note that the apprehension that Republican-leaning counties were not falling into line was seemingly unfounded on the surface. But then this piqued our curiosity: what then were the reasons that the Republican counties seemed better at coping with the coronavirus? In the future blogs in this series, we will take a closer look at what other factors actually did play a part, and maybe even try to determine if this is a data generation issue arising from poor reporting practices in certain counties. Stay tuned for more!

Related posts:

The talent in the NBA is as evenly spread as I can remember (aside from the whole Warriors thing once they get healthy). As Golden State has struggled lately, we’ve asked a particular question for…

But given that every business is different, there is no template solution to navigate this uncertain climate. Last week Fuckup Nights Toronto partnered with Xero and hosted a panel of entrepreneurs…

10 Simple and Easy Weight Loss Tips for Women. We all want to know how to make the best use of our time and effort for faster weight loss results.