Skip to content

Huge partisan differences in who wants to get vaccinated

Jonathan Falk writes:

This piece by Noah Rothman argues (appropriately hedged “It’s just one poll, and the breakdown of subsamples to the narrowest possible margins forces us to be cautious when citing the findings” which is always something good to see) that vaccine hesitancy, which shows a pronounced Republican/Democratic split may not be a Republican/Democratic split at all, but may simply be a rural/urban split with party affiliation forming an proxy. But since polls so frequently focus on the party subtabs rather than the geographic subtabs and since the correlation of these two slices has risen, we run the risk, do we not, of misattribution to Trumpiness what is in fact a population density driver.

Just as we “learned” in 2016 that we couldn’t understand the Trump vote until we disaggregated by education, maybe we can’t understand vaccine reluctance until we disaggregate by geography.

There’s a polling journalism point here as well, which is that topline numbers are complicated enough to give the public; at best you can disaggregate by one dichotomous variable and then repeat for other slices of the data. Of course, if the correlation is high enough, you can’t ever quantify the true drivers, but at least acknowledging alternative explanations ought to be part of the journalists’ toolkit.

I wasn’t sure what to think about this, so I passed it over to polling expert David Weakliem, whom we last heard from when discussing attitudes toward coronavirus restrictions—it turned out that some prominent pundits were way off on that one.

Regarding the issue discussed above of partisanship or urban/rural splits, Weakliem said:

I couldn’t access the original article (it’s subscribers only) but I found an Axios/Ipsos survey from late Feb. which asked “How likely, if at all, are you to get the first generation COVID-19 vaccine, as soon as it’s available?” The percent who said “not at all likely” (of those who hadn’t already had it):

Republican 40%
Democratic 11%
Indep. 25%
Other 41%

Urban 21%
Suburban 27%
Rural 41%

Breaking it down by both variables:

Urb Suburb Rural
Republican 28% 39% 54%
Dem. 9% 12% 11%
Ind. 23% 24% 27%
Other 33% 40% 57%

So in that survey, partisanship is a much more important factor than urban/rural. Of course, urban/rural can be defined in different ways, but given the size of the partisan divide I don’t think the general conclusion would change (they also had a variable for in/not in a MSA and the results were very similar).

Partisanship is usually the biggest division on things related to covid, so my guess is that the survey Commentary cites is just an outlier. But I agree with the general point about the limitations of relying on tabulations in published reports. Surveys should routinely make the data available for online analysis—although journalists aren’t going to do elaborate analysis, they could at least get used to doing three-way crosstabs.

Falk then replied:

The poll Rothman was talking about is here. The question he talks about appears at page 23, and I do think, glancing at it without any more analysis, that it seems to square with the Axios/Ipsos poll, though I’m not entirely sure what you do with people who tell you they already got vaccinated.

I agree that the poll you cite is not just a rural/urban divide, easily seen by the fact that Democrats and Independents have identical proclivities across geography. But something is clearly going on among Republicans, isn’t it?. After all, the leap from 28% in urban Republicans (in shouting distance of independents) to 54% is pretty stark.

It’s striking that the urb/suburb/rural variable matters for Republicans but not Democrats or Independents. It also seems to matter a lot for Others, but I guess there aren’t so many Others out there? Wealiem writes, “The other (the label is ‘something else’) group is about 15% of the sample, which is bigger than I’ve seen in other surveys, but I haven’t looked at the codebook to get the exact question. But it does seem that the urban/rural divide matters more among Republicans than among Democrats or independents.”


  1. Rahul says:

    I found the question framing confusing: “How likely are you to get it?”

    Does that query intent or outcome? Ie did they end up probing hesitancy or pessimism?

    • Dzhaughn says:

      I read that ambiguity right away as well. But hey, I studied math once. In such a reading, the Republican population thinks that the roll out will prioritize cities, and the Democrats trust it will be more fair.

      General request: Any good surveys on research about how poll questions are interpreted by respondents?

      • Martha (Smith) says: has some discussion on question wording and interpretation, plus links to further discussion and references.

      • jim says:

        In his book “Supercrunchers” Ian Ayers has a devastating critique of survey questions.

        There’s no telling how the respondent’s brain operates on these questions. Some people see the questions as a purely intellectual question and (in this case) would weigh the factors that influence how likely they really are to get it. Others see the question as entirely a social standing question – because they see everything that way – and their response depends entirely on how the think people will perceive their response. There could be and probably are hundreds of other potential takes on the questions, and even whether or not the respondent is taking care to answer the question to the best of their knowledge or just with their first reaction matters.

        I don’t put any stock in surveys at all. Zero.

  2. Joshua says:

    > which shows a pronounced Republican/Democratic split may not be a Republican/Democratic split at all, but may simply be a rural/urban split with party affiliation forming an proxy

    Kind of an odd argument, imo, as we could effectively argue that Republican/Democrat identity orientation is a proxy for the rural/urban split.

    • Jonathan (another one) says:

      I agree, Joshua, that being a rural dweller is not in any way more or less fundamental to one’s existence than being a Republican. But the *discussion* seems to talk a lot about R/D splits and very little about rural/urban splits. Rothman’s point was that rural dwellers are simply more naturally insulated from transmission, living in less dense places. (Working environment may be the opposite, as meat packing workers demonstrated.) Thus, increased vaccine hesitancy among rural dwellers could be a rational response. (The implicit assumption here is that all party-related motives are irrational, or at least less rational.)

      Really, I don’t think these polls prove much one way or the other about this. It’s just that there’s such an intense focus on partisan splits (and to a lesser extent, racial splits, though those seem to have rapidly declined) that any explanations that aren’t purely ideological strike me as useful, and a breath of fresh air (if you will) so long as the data support them.

      • confused says:

        @Jonathan >>any explanations that aren’t purely ideological strike me as useful, and a breath of fresh air (if you will) so long as the data support them.

        Yes, definitely.

      • Joshua says:

        Jonathan –

        > Rothman’s point was that rural dwellers are simply more naturally insulated from transmission, living in less dense places. (Working environment may be the opposite, as meat packing workers demonstrated.) Thus, increased vaccine hesitancy among rural dwellers could be a rational response.

        I dunno, I’ve pretty much lost any faith in cross-locality comparisons, as there are so many hard to control confounding variables. Sure, there’s a common sense logic that the more sparse the population the lower the spread and thus there’s a common sense to a different analysis in the risk analysis comparison between taking the vax vs. not taking the vax. But then again, we see that North and South Dakota, Iowa, and Utah are among the 5 highest per capita in COVID cases, despite none of them being at the top in the per capita testing rate, and in fact North and South Dakota being pretty much bottom dwellers in that regard.

        Of course, you’d have to figure in factors such as % living in congregate settings like migrant worker or senior housing..or the ratio of overall population density to population distribution (IOW, a relatively small number of people in a really large state being heavily concentrated in just a few population centers)…

        I linked this clip earlier – but when I see something like this it’s hard for me to get around the overlap between partisan identity and views on vaccines.

        Of course, Tucker is not necessarily representative and likely something of an outlier and although climate blogs and Twitter are hardly representative, it’s hard to ignore the clear alignment with political orientation and views on vaccines that I’ve seen there as well.

        I will say, however, that there is a fairly common perception that orientation on vaccines more generally, and GMOs, show a strong signal of partisan orientation even though most carefully assembled data (that I have seen) show no such signal.

        • confused says:

          >>a fairly common perception that orientation on vaccines more generally, and GMOs, show a strong signal of partisan orientation even though most carefully assembled data (that I have seen) show no such signal.

          What I’d read (pre-COVID) on anti-vaccine sentiment was that it tended to be associated with *both* extremes of the political spectrum (extreme nature-oriented thinking or anti-corporate “big pharma” conspiracy stuff on the left, anti-government conspiracy stuff or fringe religious motivation on the right)

        • confused says:

          >it’s hard to ignore the clear alignment with political orientation and views on vaccines that I’ve seen there as well.

          I think one thing that runs through a lot of this — I’ve mostly seen it in connection to climate and some other environmental issues, but I think it applies to COVID too — is an idea some (mostly right-leaning) people have that the scientific community/academia has been in recent decades “corrupted” by government funding, and therefore can’t be seen as objective any longer in issues where government money/power is at stake.

          To some degree, I think this pre-existing idea just got applied to COVID. I do think there is a somewhat important distinction in terms of “anti-science” between this and, say, creationism in that the “one-sided skepticism” only really gets applied when government action is suggested.

          (For example, I’ve seen arguments that, yeah, climate change is probably real to some degree, but people are only looking at the negative effects of it, not the positive; more people die in cold snaps than heat waves, etc. etc.)

        • confused says:

          I think the key point is that *early on* there was very little trouble in rural areas – by the time it got there, attitudes were already “locked in”. If we’d really known what was going on in February, if the CDC had been able to “get out ahead” of the President and other politicized actors, maybe it could have been different.

          But as it actually happened, NYC spiked, measures were taken in the West Coast and the Northeast, there was a widespread media message that because of Spring Break Florida and Texas were going to be “the next NYC”… and then *that didn’t happen*.

          For the next two months (mid-March to mid-May) nearly all the deaths were in the Northeast and Upper Midwest (Chicago, Detroit, etc.) Many Southern states started to reopen (Georgia earliest, IIRC), South Dakota etc. never did much in the first place — and through May, it looked like they’d dodged a bullet.

          By the time things DID clearly get bad outside the Northeast/Upper Midwest, things were hugely politicized, so opinions didn’t change much.

          A nitpick, not really countering your overall point:

          >>But then again, we see that North and South Dakota, Iowa, and Utah are among the 5 highest per capita in COVID cases

          The picture’s significantly different if you look at deaths, though. The Dakotas are high-ish, but not among the very highest*, but Utah is among the lowest.

          Partly this is likely because Utah is by far the youngest US state (median age 31 vs. 38 US average), but I think also the *total* number of tests per capita is a rather poor guide to what % of total infections were detected, since New England had a ton of infections early on when testing was near zero.

          *NJ, CT, MA, AZ, and MS – a rather odd list. CDC lists NYC and rest-of-NY-state deaths separately, so NY doesn’t show up on the top 5.

      • Joshua says:

        And btw –

        > Thus, increased vaccine hesitancy among rural dwellers could be a rational response.

        I think that classifying behavior as rational or irrational gets really complicated. I don’t think that behavior that’s heavily influenced by biases such as ideologically-based motivated reasoning is “irrational.” There’s a kind of logic to such reasoning, even if it isn’t a logic that stands up to a more clear-eyed and less biased analysis.

        IMO, a problem develops in that people link biased reasoning with irrationality, as if anyone is afforded some kind of immunity from biased reasoning by virtue of being more rational.

        So as much as I disagree with the reasoning employed by someone like Tucker in that clip, I try to employ a kind of “cognitive empathy” to try to avoid a tendency to conclude that those who disagree with me are necessarily being irrational.

        • Jonathan (another one) says:

          I almost used air quotes because ‘rational’ is indeed quite complicated. And I fully agree with you that cognitive empathy is often required. (Andrew has talked about this a lot in his discussions of the two modes of economist discourse, as here: I was really using a shorthand here.

          In any case, *if* one finds a strong population density gradient to vaccine reluctance, there are competing explanations for the fact. I assume that whatever the confounders, we can agree on that. By the way, the Carnegie-Mellon data that Alex Reinhart refers to below asks about the reason for reluctance with a number of possible different answers, including distrust of the government, distrust in the vaccine itself, and lots of other stuff. These two are reported at the county level. I’m planning to take a look over the next couple of days.

  3. name withheld by request says:

    Among those people I know who vote Republican in every election no matter what, there are two characteristics they tend to share. First, they voted for Trump ONLY because he had (R) by his name on the ballot. They didn’t like him, would never do something because he told them to and they hope he disappears from the scene as soon as possible. But they always vote Republican no matter who is nominated. The second characteristic is, they all wanted to be vaccinated as soon as possible if not sooner.

    It all makes me suspect there are at least two distinct “identity orientations” which both check “Republican” on any survey but otherwise are very, very different.

    • Andrew says:


      There’s also the interaction of partisanship by age. Republicans are more likely to be old, and old people are, I assume, more likely to get vaccinated. So young Republicans must be really really not into the vaccine to make the numbers work out the way they do.

    • kj says:

      You might find this poll/analysis interesting. Using some methodology I don’t know, they classified Republicans into ‘tribes’. The results feel reasonable enough.

    • confused says:

      At least two, yeah. And I think there are significant number of strong Trump supporters who didn’t generally vote Republican in pre-2016 elections, because they didn’t vote at all. The really odd thing is not so much the populism (which has plenty of precedent in US history) but the combined “being President and anti-government at the same time” thing.

    • Meg says:

      And also need to consider that in a very liberal state the identity of republican and democrat are slightly different than what you would expect at a national level (excluding the cases where nominative democrats went all in on the Trump cult of personality.)

      So in my state, someone who identifies as Republican probably has more in common ideologically with Democrats in more rural and conservative states, than they would with Republican in those states.

      Depending on how that Urban-Rural gradient is defined, areas of my state that residents would consider Rural may be defined as suburban or urban. And For SURE the actual distance between many of our rural zipcodes and an Urban zipcode could be <10 miles. (hence why it depends on how you are defining those, because in my probably very obvious but unnamed state, there are almost no areas (maybe none) that are outside of a major metropolitan area.

    • Dzhaughn says:

      “…those people I know who vote Republican” is a good sampling method for only a small set of questions. And do you really know how they vote?

      • name withheld by request says:

        I’m not trying to sample anything. I’m not a pollster.

        But yes, I know how the acquaintances I’m talking about vote. Unless every statement about politics they have ever made is a lie. Including their contempt for Pres. Obama and Vice-Pres. Harris specifically (not to put too fine a point on it).

  4. Greg Snow says:

    One thing that I don’t see mentioned, but should be considered is the wording of the question and interpretation of it.

    The version of the question quoted in the blog includes “as soon as it’s available?” and talks about first generation. When the vaccines started rolling out I figured that I would probably get mine sometime in May or June. Not that I did not trust the first doses, but rather I figured there were other people who should have higher priority than me since I have easily tranisitioned to working from home and did not attend sporting events or many large gatherings even before the pandemic.

    If the question was about trusting the vaccine, I would have responded “Yes”, if it was about contributing to herd immunity I would have responded that I would eagerly receive the vaccine to help with herd immunity and end the pandemic, but since the question looks more like: “will you take the vaccine that would better be used by someone who cannot work from home and needs to return to work, or is a much more likely to spread the virus in social situations” I would have responded that I would wait and give the early doses to others who need them more.

    • confused says:

      I would imagine there will be a significant number of people (even 5-10% is a huge number over the whole US) who have some doubts, will not actively seek vaccination as soon as possible, but will accept it especially if it’s encouraged by e.g. their own doctor who they trust.

      • Keith E. says:

        Yeah, I’m fine with getting vaccinated, but I’m not too personally worried about Covid, and I won’t press for it until everybody that eagerly wants it has gotten it.

        Which in truth means I’ll get vaccinated when my wife makes me.

  5. Alex Reinhart says:

    For any readers interested in trying to study geographical variations in vaccine acceptance, we (the Delphi group at Carnegie Mellon, with Facebook’s support for recruitment) actually conduct a survey that reaches about 50,000 respondents daily in the US, and gathers COVID-related data down to the ZIP code level. County- and state-level aggregates are available for download:

    Unfortunately we don’t ask respondents political affiliation or a proxy, but one could link the data to any other county-level covariates you might have, including measures of population density, political leanings, etc.

    We get past the wording questions in the comments above by asking “If a vaccine to prevent COVID-19 were offered to you today, would you choose to get vaccinated?” That, I think, avoids questions of motivation (it’s too hard to get vaccinated) or if people want to wait, are not yet eligible, etc.

    • Jonathan (another one) says:

      Thanks, Alex! This is really interesting. It only took me about an hour and a half to figure out how to make the relevant dataset. (This is about 75 percent my fault and 25 percent the fault of the documentation…) There are going to be some real problems trying to do this at the county level, but I’m going to take a look anyway. (As someone without an institutional affiliation, I’m apparently ineligible for the zip-level data, unfortunately.)

      • Alex Reinhart says:

        Yeah, the county-level sample size is only adequate for us to report a subset of counties, despite the huge overall sample size. Our public API does aggregate together unreported counties in each state into one “megacounty”, though, so you at least can get univariate estimates for all the less-populous counties in a state grouped together. Some of our blog posts include example code, like this one:

        Always happy for suggestions on how to improve the documentation, since a lot of it is pretty new.

        • Jonathan (another one) says:

          Well, my first problem was installing your python package, because the pip version won’t install the geopandas on Windows, but whatevs. I love stackoverflow.

          On the data itself, I was going to drop all the aggregated counties, because I have no way to get their density, and restrict the analysis to counties with known densities and vaccine hesitancy data. I know *why* you don’t simply report for every county, but it’s kind of a pain for some purposes, like this one.

    • Dzhaughn says:

      Good wording of the question!

      • Dzhaughn says:

        That is, if you are measuring baseline attitude.

        Later, we can see if that baseline attitude affects the outcome of the vaccination program. I will bet $5 that it does not have an affect that can be seen epidemiologically.

  6. Navigator says:

    I believe the response to the poll question changes with each passing day, now that more than 100 mil. have been vaccinated.

    It was funny to watch Bloomberg’s site over time:

    Folks in ‘red’ states were the first ones to roll up the sleeves and are vaccinated to much higher %. I know that it is because the states are smaller and distribution is easier (probably less unnecessary bureaucracy too), but it was amusing to see the unexpected trend nonetheless.

  7. Michael Nelson says:

    I think people want to use responses to questions like this as a proxy for answering the question, “Do rural Republicans not want the vaccine because they’re irrationally paranoid about government and science? Or are there just more Republicans in rural areas, and rural areas have been hit less hard by COVID than cities, so the perceived threat is lower?” But if we want that information, we should stop reading tea leaves and add questions to surveys like, “How likely do you think it is that the vaccines are effective at preventing illness and death in a) cities, b) suburbs, c) rural areas?” and “How much do you trust that those who are developing and promoting the vaccines are doing so for the public good?” and “Do you thinking being a good or patriotic citizen means a person is a) more likely to get vaccinated, b) less likely to get vaccinated, c) no more or less likely to get vaccinated?”

    Until we have that data, we don’t have any way to make a causal inference. However, it’s still very important, descriptively, to see that rural Republicans aren’t as eager to get the vaccine. Rural, conservative culture is against following scientists’ and doctors’ consensus advice at the national level. Therefore, something is broken in that culture. This is fundamentally different from the debate over the lockdown. People in rural areas may be anti-lockdown because their community appears to face greater economic than medical threats, so that’s a rational, location-specific response. But vaccination fights both types of threats. Recognizing the irrationality here requires no inference, only deduction.

    • Jonathan (another one) says:

      I agree (though I’m as willing as anyone to speculate in the absence of those questions.) But the Carnegie Mellon Delphi group referred to above asks those questions, so I’m going to take a look.

  8. Joshua says:


    > This is sort of an interesting example of response bias in polls. A new Quinnipiac poll found that 38% of adults in New York say they’ve “gotten a COVID-19 vaccine”. But we know what the *actual* number is and it’s closer to 30% (at least one dose).

Leave a Reply

Where can you find the best CBD products? CBD gummies made with vegan ingredients and CBD oils that are lab tested and 100% organic? Click here.