Nate Silver vs AnEpidemiolgst

This beef started with this tweet

which is just something else for multiple reasons.  Tone policing a neologism is just stupid, especially when it’s basically accurate.  Doing so without providing a preferred term is even worse.  But, you know, I’m probably not writing a post just because somebody acted like an asshole on twitter.  I’m doing it for far more important reasons, namely:


And in this particular case, it’s not Nate.  She also doubles down with

which is obviously wrong, even for a fuzzy definition of meaningfully, if you stop and think about it.  R0 is a population average.  Some people act like hermits and have little chance of spreading the disease much if they somehow catch it.  Others have far, far more interactions than average and are at risk of being superspreaders if they get an asymptomatic infection (or are symptomatic assholes).  These average out to R0.

Now, when 20% of the population is immune (assuming they develop immunity after infection, blah blah), who is it going to be?  By definition, it’s people who already got infected.  Who got infected?  Obviously, for something like COVID, it’s weighted so that >>20% of potential superspreaders were already infected and <<20% of hermits were infected.  That means that far more than the naive 20% of the interactions infected people have now are going to be with somebody who’s already immune (the exact number depending on the shape and variance of the interaction distribution), and so Rt is going to be much less than (1 – 0.2) * R0 at 20% immune, or in ELI5 language, 20% immune implies a lot more than a 20% decrease in transmission rate for a disease like COVID.

This is completely obvious, but somehow junk like this is being put out by Johns Hopkins of all places.  Right-wing deliberate disinformation is bad enough, but professionals responding with obvious nonsense really doesn’t help the cause of truth.  Please tell me the state of knowledge/education in this field isn’t truly that primitive.    Or ship me a Nobel Prize in medicine, I’m good either way.

8 thoughts on “Nate Silver vs AnEpidemiolgst”

    1. The first preprint suggests herd immunity at 10-20% TOTAL of the population infected. The lower end of that is likely too low IMO, at least for pre-distancing behavior, but that’s a gigantic decrease. The second one implies a “reduction in R0 used to calculate the threshold” (that is mangling the usage of the terms and I know it) from 2.5 (60% threshold) to 1.75 (43%). My guess is the truth is somewhere between the two models,

      The confound is behavior. Under lockdown, the right tail is going to be even more overrepresented than under distancing (making this effect larger/threshold lower), and more represented under distancing than under naive behavior.


      1. I was just quoting Lipsitch’s takeaway, not venturing an analysis of the two papers. He offers some additional thoughts here on why the herd immunity threshhold may be fairly high despite this factor: For one thing, R0 may be far higher than 2.

        I don’t know enough to have a strong opinion on this. But intuitively, the high infection rate we’ve seen in places like the USS Roosevelt seems inconsistent with the idea that there is huge variation in susceptibility to infection. A COV of 2 or 3 implies a huge population of pretty resistant people and a relatively small number of vulnerable people, and that doesn’t seem consistent with what we’re observing. It also seems odd to see the same people arguing both that a high proportion of population is already infected *and* that there is huge variation in susceptibility — very unlikely that both are true.


      2. I agree that there’s unlikely to be a big susceptibility variation, but there *is* a huge exposure variation (how many other people’s air you breathe, and how much of it), and compared to the basic mixing model (all contact pairs are equally likely), that has a similar effect (as long as social people stay more social on average, which they do)

        I wasn’t trying to say that reaching 20% is beyond a threshold for full herd immunity, although at this point it’s quite plausible to me that reaching 20% under current distancing behavior is more than enough to get Rt<1. NY seems to have clearly turned the corner, and other places with similar or stricter distancing didn't, and the biggest difference to me is that they're still in the low single digits in infection rate.


  1. The problem with your analysis is that you’re divorcing it from the political and cultural context within which it is occurring, which is problematic considering that the space Nate operates in these days is primarily punditry. The way that the phrase “herd immunity” was used early on to advance the “take the punch” strategy of foregoing mass social distancing was somewhere on the scale between dishonest and misinformed, leading to a bunch of people who don’t understand the consequences saying “actually, it’s *good* for us if we let more people die”.

    That is what @AnEpidimiologist is pushing back on, and, for good or for ill, epidemiologists have been enlisted in the war on COVID disinformation the same way a prominent SABR nerd was enlisted in the war on crappy electoral analysis. It’s not her job to find a different, less misleading catch phrase for Nate to use in his armchair analysis.


    1. I think there’s a fundamental disconnect. She and you saw Nate’s stuff as being boneheaded and him as something like a useful idiot for #TeamGetEverybodySick and acted accordingly. What she missed at the time, and what both of you may still be missing, is that Nate was basically correct and that such terms are important for describing *what has already happened*. I don’t think he knew why, or he would have articulated the mechanism better (and probably commissioned 50 different 538 articles about it by now), but his observation that 20% infected in NYC seemed likely to be a very big deal *is almost certainly correct*.

      The main point of my post is that the linked Johns Hopkins piece IS MISINFORMATION. Her tweet that 20% doesn’t matter (much beyond 20%, obviously) IS MISINFORMATION. Maybe I’m naive, but I want the scientist-experts giving out good info, and that’s nowhere close to good info.

      Look at the preprints that are linked to in the tweet from Guy’s comment. Sit down this evening and code a basic mixing model that allows for differential exposure rates, seed an outbreak, and look at how differently it plays out- and how early Rt can go below 1- when you have a decent exposure differential vs. when you don’t (especially under conditions where distancing behavior is already reducing transmission rates). This is an important effect.


      1. I think there are several fundamental disconnects, many probably resulting from Twitter as a medium being a terrible way to express nuance and include the many caveats that are needed to properly have this conversation.

        The first and most obvious (to me, anyway) disconnect is that even if we assume arguendo that Silver’s theory is correct, that doesn’t mean that he’s not being a useful idiot for #TeamGetEverybodySick, because the words we use matter. Like it or not, the concept of herd immunity was commandeered by #TGES as an *alternative* to social distancing, and even if the underlying idea is sound, slapping the word “partial” on it plays into the “actually, letting the virus infect and kill people is good!” mindset. If this is proven to be a real thing, it absolutely needs a different label. I don’t know if that’s where @AnEpidimiologist’s hositilty toward the term is coming from, but it would be quite understandable if someone who studies diseases for a living in this stressful time for the field’s practitioners were to be a bit glib or even a bit misleading in a tweet that could reverberate into “see, SCIENCE says don’t worry about it. We got this!” moment that she then has to walk back.

        Now, I get that we don’t want epidemiologists massaging answers or outright lying if the science is definitive in a particular direction, but I’m sure they’re adjusting their message to the medium when they speak publicly on social media, and to pretend otherwise is silly. They’re scientists, and right now, they’re public intellectuals that have the ability to change society’s response to the disease with their words.

        Meanwhile, I don’t think that two non-peer-reviewed studies, one of which had one of its authors loudly downplaying its applicability to this conversation and reminding everyone that social distancing is still what’s doing much of the work [1] , prove the existence of an effect, nor would me cooking up some Python to run a simulation. If it were that easy, we wouldn’t need the epidemiologists, right?

        But here’s my main complaint about the bits of this conversation I’ve seen (with the caveat that I’m not on Twitter, and only ran into this because I follow your blog): Nate’s tweets about this outline a scenario where some non-negligible number of people with antibodies PLUS social distancing leads to a lower R0. However, the concept of herd immunity is based on a population that *isn’t* social distancing, because social distancing on the magnitude we’re doing now is atypical. What they’re saying with herd immunity is that if enough people have antibodies, we’ll hit a sub-1 R0 and eventually the virus will fade away, even absent social distancing. Of course the social distancing could have an impact on the R0, and maybe there are key findings in these papers that will be confirmed via peer review and replication that can help us come toward a better understanding of what exactly the parameters are, but until that happens, this is all conjecture with a bit of experimental validation, not the kind of thing one can authoritatively say “is an important effect” based on two papers and one’s own intuition.



      2. I agree that the language could be different, but.. any form of expressing that idea can potentially be latched onto by #TGES, and I’m not willing to acquiesce to a world where legitimate ideas are inexpressible because bad-faith actors can use it to whip up morons. They’re already plenty good at doing that with zero basis in reality.

        As far as the science itself, you have far too much faith. That’s why I suggested coding something yourself to see. You don’t need a highly sophisticated model to figure out that punting on 4th and 1 from your opponent’s 40 late and trailing is terrible. Any halfway reasonable numbers all tell you the same thing, and it’s no different here. Plus, you know that the “expert model” of 1-1/R0 doesn’t consider the effect at all and assumes a random distribution of immunity, so you know exactly why they’re wrong. It really is that easy here. Which is quite sad, but that’s another story.

        Of course social distancing is huge. The rest of the country would have gotten nuked out of the gate otherwise, and they didn’t under distancing, even at very low infection percentages. And that shows up in a basic simulation. I put in parameters I thought were conservative and I was getting herd immunity thresholds for various R0s going down from 72%->46%, 50%->29%, 33%->18%, etc. That’s in the ballpark of the second paper. And large differential exposure isn’t some theoretical thing, you can observe it, you can read about it, etc. Basically, this can’t work any other way.

        Distancing caused a much larger subset of people (households) to transmit at R<<1. Only a subset can still transmit the disease above replacement, and that subset is much smaller than 100% now because of distancing. So once that subset is immune enough, the disease can't spread above replacement (if behavior remains constant).

        The differential exposure effect isn't saving the world when most people still transmit above replacement (although you can still see it do something), and it's not, IMO at least, something we should accept and fall back on. If we do idiotic things like Seoul reopening nightclubs, we're going to get blasted again accordingly.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.