STUMP » Articles » On COVID (and all-cause) mortality and Political Affiliation Studies (plus a geeking-out, know-your-data coda) » 6 October 2022, 09:57

Where Stu & MP spout off about everything.

On COVID (and all-cause) mortality and Political Affiliation Studies (plus a geeking-out, know-your-data coda)  


6 October 2022, 09:57

Yes, I’ve seen the latest ones, and many of the earlier ones.

I will have some specific comments on the most recent report, and my comments will be at the bottom of this post, because I just discovered something very specific about one of the two states (yes, they looked at only two states in their analysis, not the whole United States) when I was looking for something specifically, which will kick off my discussion below.

The most recent paper released: mortality experience in Ohio and Florida for a very specific period in 2021

Here is the coverage from Think Advisor:

Why Republicans Faced a Greater Risk of Death Than Democrats in Late 2021

Some Republicans’ reluctance to get COVID-19 vaccinations led to a big gap between the total excess death rate for Republicans and Democrats in Florida and Ohio in late 2021, according to three Yale University researchers.

The overall excess death rates for Republicans ages 25 and older and Democrats ages 25 and older in those states were about the same up until April 2021, when all adults in those states were eligible to get COVID-19 vaccines, the researchers say.

The overall excess death rate for Republicans then increased to be about 5 percentage points higher than the overall excess death rate for Democrats, until a giant new wave of COVID-19 cases began, in August 2021. That month, the overall excess death rate was about 20% for Democrats and about 35% for Republicans.

Here is the paper: Excess death rates for Republicans and Democrats during the COVID-19 pandemic


Political affiliation has emerged as a potential risk factor for COVID-19, amid evidence that Republican-leaning counties have had higher COVID-19 death rates than Democrat-leaning counties and evidence of a link between political party affiliation and vaccination views. This study constructs an individual-level dataset with political affiliation and excess death rates during the COVID-19 pandemic via a linkage of 2017 voter registration in Ohio and Florida to mortality data from 2018 to 2021. We estimate substantially higher excess death rates for registered Republicans when compared to registered Democrats, with almost all of the difference concentrated in the period after vaccines were widely available in our study states. Overall, the excess death rate for Republicans was 5.4 percentage points (pp), or 76%, higher than the excess death rate for Democrats. Post-vaccines, the excess death rate gap between Republicans and Democrats widened from 1.6 pp (22% of the Democrat excess death rate) to 10.4 pp (153% of the Democrat excess death rate). The gap in excess death rates between Republicans and Democrats is concentrated in counties with low vaccination rates and only materializes after vaccines became widely available.

Here is a nice scatterplot, so you can eyeball how good the fit and correlation looks to you:

Why are people doing this?

This is not the only such paper I’ve come across. There are academic papers, and then there are various analyses that appear in places such as The Economist, Financial Times, and the New York Times.

Here is just a sampling of some of those papers/articles:

June 2022, BMJ: Political environment and mortality rates in the United States, 2001-19: population based cross sectional analysis — note, this stops in 2019.

March 2022, Pew: The Changing Political Geography of COVID-19 Over the Last Two Years

Here’s an interesting graph from that one, but there are two issues with it:

I’ll come back to this, and another graph, from this piece. It’s somewhat okay, this Pew study.

December 2021, NPR: Pro-Trump counties now have far higher COVID death rates. Misinformation is to blame

The NPR story has some interesting scatterplots, too, but have the same problem the Pew story does.

Now, I’ve done a few ranking posts on STUMP, and I’m getting ready to do a new visualization/ranking again, because I’m working on a dashboard implementation (but it’s going to take me a while). I have an ulterior motive.

Why are they doing this?

There are many possible motives.

Now, yes, we know the Republicans are EEEEEEVIL and people want them dead. So yay their mortality rates are higher. That’s the purpose of these studies, right?

No, they cry — it’s to show that vaccination is good!

Then why focus on political affiliation? Why not just use vaccination rates as the independent variables?

Uh, we’re showing that Republicans are less likely to be vaccinated!

Well, then show that — you don’t need to get mortality rates involved. By the way, the NPR and Pew stories did have graphs that were directly about the vaccination rates by political affiliation. Here’s the Pew graph:

Not exactly party affiliation, but you get the drift. It’s not a bad proxy. The blob is not bad for a correlation – they didn’t give the correlation coefficient, but it should be pretty good for this sort of thing.

In any case, motives for the academics can be something as simple as they need publications for tenure. Showing something as mundane as “where mortality was bad before, it was bad during COVID” tends not to get the headlines, grants, clicks, and attention that “Hey! The Republicans are dying more! HA HA!”

The trend was there before

I want to note that I don’t think these correlations are entirely spurious, but they may be measuring something entirely different from partisanship, per se. Or, rather, the partisanship is the result of something in common that is driving the higher mortality, and no, it’s not necessarily just differences in vaccination rates.

Also, I’m sorry, simply giving me vaccination rates for all adults over age 18 is not very interesting or useful for a disease with such a strong “age co-morbidity” profile. COVID is far more deadly to old people than young people. It is far more important to get seniors vaccinated compared to young people, and plenty of vaccinated old people have died of COVID. No, it’s not because the vaccines are useless — it’s just that the mortality risk for COVID is high compared to a disease like the flu. Our treatment regimes are still lacking, and if you notice, our flu vaccines have also lacked for years, and many of us still go get them annually.

But here is what’s the problem with many of these analyses (but not necessarily the new one) — often, they’re not age-adjusted.

Many of these graphs can be explained by populations that skew older. Most COVID deaths were of old people. Certain areas of the U.S. have a considerably older population than other places — rural areas, the midwest, etc.

(The other problem with the Pew graph was that it was based on when deaths were reported, not when they occurred — thus the dip in deaths near holidays like Christmas or Thanksgiving. People were dying then… it didn’t get into the reporting system til people came back from vacation.)

So in many of these cases, all you’re capturing is that Republicans skew older than Democrats. Or fatter (I do not know if this one is the case), or more likely to have one of the many co-morbidities that makes it more likely that the delta/omicron wave of 2021 was deadlier for you than the Democrats who were less likely to have those co-morbidities.

It can just be chalked down to demographic differences between the political groups, and vaccination isn’t even in there.

In some cases, I could tell the mortality differences supposedly coming from Republicans were due to specific geographies (like the Southeastern states) having higher mortality pre-pandemic, just as a baseline (cf diabetes, obesity, etc – this is even taking age-adjusted aspects into consideration). In some cases, for certain geographics, COVID impacts were disproportionately coming from non-white populations that weren’t very likely to be Republicans, but the overall county or state was skewed Republican.

Now, for my motives — yes, I am working on some Power BI dashboards, but I am seeking areas where COVID and all-cause mortality effects were worse (or less bad) in 2020 and 2021 compared to other places, and there are geographic patterns, and racial/ethnic differences, as well as big relative differences by age groups. I’ve been developing these graphs and analyses since 2020, and taking you along for the journey.

No, I don’t do political affiliation as one of the dimensions, and you’re about to see why.

It’s not just because I don’t feel like dunking on Republicans (given I am a registered Republican deep in blue country in New York.)

The newest paper: focuses on only two states

I am not going to flay the new paper, for a few reasons.

First, I’d like to get my hands on the data. No, it’s not that they won’t give it to me — they’re going to be publishing the data on a site where people share their data sets. It’s that I’d like to actually look at their data sets and see the analysis directly for myself to critique that, rather than just try to eyeball some graphs.

Second, this is just the pre-print of the paper. They’ve just released the draft publicly, and it will go through peer review before official publication (and, unfortunately, the data and model code aren’t going to be released until after official publication.) I don’t want to waste my time flaying the pre-print.

But I will point out some features. They did try to sex- and age-adjust for expected mortality. I don’t know if they did it properly, but that they knew they had to do that is better than most of these papers, so yay.

But this is the part that made me unhappy: they were looking only at Ohio and Florida:

To calculate excess deaths, we use 577,659 deaths of individuals linked to their 2017 voting records in Ohio and Florida who died at age 25 or older between January 2018 and December 2021.

Third, our study is based on data from the only states where we could obtain voter registration information (Florida and Ohio); hence, our results may not generalize to other states.

Overall, our results suggest that political party affiliation only became a substantial risk factor in Ohio and Florida after vaccines were widely available. Lack of individual-level vaccination status limits our ability to draw broad conclusions, but the results suggest that the well-documented differences in vaccination attitudes and reported uptake between Republicans and Democrats [10, 7, 8, 13] have already had serious consequences for the severity and trajectory of the pandemic in the United States. If these differences in vaccination by political party affiliation persist, then the higher excess death rate among Republicans is likely to continue through the subsequent stages of the COVID-19 pandemic.

Frankly, I would not assume that Republicans had a different uptake of vaccines in Florida than Democrats did (I know lots of individuals in Florida, of a variety of political bent). They may have had different uptake in the cities versus more rural areas due to distribution method differences, but for the old folks, a lot went for it ASAP. And a lot of neurotics of the liberal bent avoided vaccines. People make this a political thing where some of the conspiracies are of a more medical nature.

But no, I have a dispute here that’s different — and it involves Ohio.

The weird way Ohio defines “registered party members”

I’m a registered Republican in New York, and I’m on a local party council in a very small town in Westchester County. Unsurprisingly, Republicans are outnumbered by Democrats in the town, officially… but we have a Republican Town Supervisor, and many Republicans on the town council. Locally, there is a relatively large number of unaffiliated voters, but that’s unusual in our state.

New York has extremely strong party protection regs on the books, a result of the history of strong machine politics in the state. To vote in a party’s primary, you need to have registered for that party officially (which you do when you register to vote) at least a year before that primary. They do not make it easy to switch parties rapidly.

So you don’t see the “tee hee, I’m going to vote in the other party’s primary to screw with it” shenanigans that various people played in 2016, for instance, thinking how cute it would be to foist Trump on the Republicans as a candidate. Har har.

Ohio does it differently.


Under Ohio law, a voter affiliates for voter registration purposes with a political party by voting in that party’s primary election. A voter is a member of a political party if:

(1) they voted in that party’s primary election within the preceding two calendar years, or
(2) they did not vote in any other party’s primary election within the preceding two calendar years.

Likewise, by law, a voter is identified as unaffiliated with a party if they did not vote in a party’s primary election within the preceding two calendar years. Party affiliation numbers as recorded by the county boards of elections fluctuate based upon a number of factors; most prevalently voter participation in primary elections.

According to the county boards of elections’ voter registration and voter history data from 2019-2021, Ohio’s voter affiliation breakdown is currently as follows:

Number of Registered Voters in Ohio: 7,982,501
Number of Registered Democrats: 947,027
Number of Registered Republicans: 836,080
Number of Registered Libertarians: 2,847
Number of Unaffiliated Registered Voters: 6,196,547

This is an important consideration — the vast majority of Ohio’s registered voters are considered unaffiliated with a party, because they’ve not voted in any party’s primary within 2 calendar years. That’s 78% unaffiliated.

In New York, according to the latest stats, only 23% of voters are unaffiliated with a party. 50% are registered Democrats, 22% Republicans.

What the heck? You’re going to analyze party affiliation based on a state where 78% of the registered voters aren’t even affiliated with a party, officially? Come on, man.

There were fewer “registered Republicans” than Democrats listed, even though the vote in 2020 went for Trump, because there was no contested Republican presidential primary. So fewer people show up for down-ticket primary races in that period. There is going to be data problems! Republicans will have gone undercounted!

I was going to do a map trying to contrast party affiliation with something like population density or urban/rural stuff, but now I’m not even going to bother. This is absurd if you want to prove anything about actual voting affiliation.

So yes, I want to wait for their data set, to check out what we’ve got here. I want to do some data quality checks, because the smell test has something very stinky going on. If you’re doing a Democrat vs. Republican comparison, but about 80% of the data is “None”… COME ON, MAN!

Imputed data/missing data is a big deal in data science, and if you’re doing analysis of variance (essentially) where the largest category is omitted, I just don’t know what to tell you.

Short version: So what?

So let me stop right there. It may be that everything is on the up-and-up, and I don’t have the time to dig through all their details now. I really do have some mortality things to write (watch this space — forthcoming!)

But my point is this — there are real differences in mortality trends geographically/demographically in the U.S. which preceded the pandemic, and many of which were exacerbated by the pandemic.

If you want to convince others that vaccines are effective in reducing mortality, sorry, this ain’t it — use vaccination rates, not partisan affiliation, for your analysis.

If you want to dunk on your political enemies with these studies, then go right ahead, but I don’t know what result you think you will get out of it. If that’s the sort of thing that makes you feel good, then memento mori.