8

It's been claimed in the media that

Back in September, health care analyst Charles Gaba also told the New York Times that counties where Trump received at least 70% of the vote in the 2020 presidential election, COVID-19 had killed about 47 out of every 100,000 people since the end of June. But in counties where Trump won less than 32% of the vote, the number was a fraction of that — about 10 out of 100,000.

A more detailed version of the claim appears on his website:

enter image description here

That ratio seems to refer to the higher vs lowest 10th population "bracket"; the latter voted <26% (not <32%) for Trump, so he was probably slightly misquoted or miscommunicated that to the press. (The highest bracket voted >70.6% for Trump, so that fits. The brackets here are taken by equal division of the population rather than vote share deciles, so the division is somewhat confusing/unusual.) The Covid-19 death rates (per 100K pop.) in these two extreme brackets were 7.29 vs 34.89, which gives a 4.786 multiplier.

At the end of that page there's update in which he (further) claims that

the [death] ratio between the highest and lowest bracket has increased further yet, from 4.79x higher to 5.46x higher in just 8 days.

enter image description here

Based on some discussion in the comments below, it might be the case that what was reported to the wider press was actually this 2nd graph, which has a 46.74 ~ 47 death rate for the last bracket while the 2nd (rather than the 1st) bracket ends at 31.6% ~ 32% vote share for Trump. The death rate average of these first two brackets is (12.19 + 8.55) / 2 = 10.37 ~ 10 from the first press report. This while the graph itself highlights the 5.46 ratio of death rates in last to of 1st bracket.

Is what Charles Gaba claims true, were Covid-19 (per capita) death rates at approx. 5 times greater in the >70%-vote-for-Trump counties vs the 26%-32% vote-for-Trump counties, between end of June and mid-Sep 2021?

Also worth nothing perhaps that an earlier (peer-reviewed) analysis by Gao and Radford covering the time period of April 6 to May 25 of 2021 found that:

controlling for several potential confounders, counties with higher levels of Trump support do not necessarily experience greater mortality rates due to COVID-19.

So what Gaba claims is a pretty substantial change in that regard.

This G&R paper is actually a bit confusing [to me] as they include stuff like stay-at-home policy/orders and restaurant closures as independent variables/confounders, basically assuming that whether these orders are issued is unrelated to the political orientation of the county inhabitants. Apparently G&R considered only whether compliance with such orders is a function of political orientation of the inhabitants.

There is also a much simpler paper/model on 2020 by Neelon & al. but at US state rather than county level. It found a reversal of Covid-19 death odds from Democrat-led to Republican-led states in July of that year, i.e. after July the odds were greater in states with a Republican governor. That analysis stops in mid Dec 2020 though. In relation to that simpler model, Gaba's claim is less surprising, but the [death] odds ratio determined in that paper is fairly small i.e. 1.18 (95% CI: 1.02-1.31) for July-Dec 2020. Gaba seems to claim a much more sizeable effect (later on). For the sake of trying to come up with a similar two-bracket measure from Gaba's last graph, if I average the last 5 and the first 5 bins, the death [odds] ratio for these is 32.92 : 15.45 = 2.12. If using instead the last 4 vs first 6 bins (which puts the voting cutoff at 49.1% -- the closest to 50% in that bracket data) the result is a fairly similar 35.005 : 17.065 = 2.05 ratio. These ~2 ratios are less shocking than the ~5 ratio of the extreme brackets, but still quite a bit higher than the 1.18 from Neelon's paper.

11
  • 1
    Your first link suggests this is about relative vaccination rates. Your second shows (in a different graph) that the vaccination rate is similar in the bottom three categories. (0-30%) It seems likely that the data are accurate, but even so there are other issues than vaccinations, such as different COVID rates at different times in different areas, for cyclical reasons (e.g. the reddest counties will be rural, the bluest urban). So likely accurate but doesn't tell is if they have cherry picked the dates
    – thelawnet
    Nov 25, 2021 at 11:31
  • 3
    The 47/100,000 does not seem quite consistent with the 34.89 in the chart, though they may involve different dates. The broad message seems correct, though it will have been slightly distorted by other factors. For example a majority of people under 30 voted for Biden, and young people face much lower death rates than older people because of their age, so counties with disproportionately large numbers of students and other young people would typically see lower death rates than counties which lacked such young people even ignoring politics and vaccines.
    – Henry
    Nov 25, 2021 at 20:18
  • 2
    @Fizz June is a reasonable cutoff for when everyone who wanted to be vaccinated was. Nov 26, 2021 at 5:33
  • 1
    @thelawnet "cyclical reasons (e.g. the reddest counties will be rural, the bluest urban)" Reddest counties being rural is cyclical? Nov 26, 2021 at 6:49
  • 1
    @accumulation, COVID goes in cycles, which won't be consistent across the country. E.g. we expect a pandemic to first spread in larger cities such as New York, and reach rural areas later. The existence of multiple waves of covid means that for example during the chosen period we would be in a different portion of the curve on average for rural places compared with urban.
    – thelawnet
    Nov 26, 2021 at 7:07

1 Answer 1

11

Charles Gaba here. If you have any questions about my methodology, feel free to inquire.

The county-level death rate data comes from the Johns Hopkins University Center for Systems Science and Engineering Github, except for Nevada & Utah (which are via the NY Times Github) and Florida, which is via CMS's COVID-19 Community Profile Report:

https://github.com/CSSEGISandData/COVID-19 https://github.com/nytimes/covid-19-data https://beta.healthdata.gov/Health/COVID-19-Community-Profile-Report/gqxm-d9w9

1
  • 2
    Just gotta say, this is pretty cool
    – dgo
    Nov 27, 2021 at 3:07

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .