Read on Twitter
I find discussions of herd immunity thresholds frustrating. Herd immunity (which I'll define as a current reproductive rate < 1) depends on policy & behavior. It is NOT an absolute number. THREAD 1/17
You cannot infer herd immunity from a high level of people already infected from the trend in reported cases.
TL;DR version: It assumes policy/behavior does not change. Data for predictions is BAD. We don't understand behavioral responses well enough. And the Lucas critique. 2/
TL;DR version: It assumes policy/behavior does not change. Data for predictions is BAD. We don't understand behavioral responses well enough. And the Lucas critique. 2/
If you say herd immunity threshold is, eg, 75% given current data, you have to assume that policy & behavior will not change--or be prescient about future policy change while making your calculation. (I have not seen this prediction about future policy done/done well.) 3/
Using a serological survey to calculate herd immunity is fraught cuz it uses data on the outcome of different contact rates under different policy/behavior regimes over time. I'm not sure how you get to herd immunity at your predicted (or current) contact rate from that data. 4/
Using the trend in confirmed cases (perhaps w a compartmental model) for your estimate requires assuming the TREND in non-representative testing data is representative of TREND in actual cases. Why would non-representativeness in levels also apply to trends? 5/
This last point (confirmed cases poorly estimate Rt) is especially problematic when testing rates change over time! Even if testing policy (aside from rates) is held constant, the population selected into testing changes as rates rise. 6/
(BTW, if you're estimating herd immunity w/o updating parameter estimates, watch out. The prediction error is likely much higher. Eg, if you start w/ R0 estimates, you have to multiply by fraction susceptible to get Rt. But estimating S(t) is *hard*. Cite: future thread.) 7/
Another complication when thinking specifically about what vaccination will do to herd immunity is that the answer depends on the fraction of ppl vaccinated in a given time period. The reason is that there are spillovers of vaccination on the non-vaccinated. 8/
Human behavior adds another complication. After we reach herd immunity, I may think it's safe to interact with others, which will raise Rt and thus possibly put us below the threshold. 9/
What makes this even more complicated is that what matters is not actual reproductive rates but what ppl think rates are. And we don't all think the same thing!
(An old article on this w/ Bennett & Chiang:
https://www.sciencedirect.com/science/article/abs/pii/S0304387814001059?via%3Dihub
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7116916/)
(An old article on this w/ Bennett & Chiang:
https://www.sciencedirect.com/science/article/abs/pii/S0304387814001059?via%3Dihub
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7116916/)
And I haven't even raised the Lucas critique @mattkahn1966! When ppl in authority give a herd immunity estimate, they may change the herd immunity threshold cuz both local govts & ppl will change their behavior. Solvign for that fixed point is *hard*. 11/
The only place where I know we have achieved herd immunity is when 100% are previously infected, there is natural immunity from prior infection, which in turn may require no substantial mutation. 12/
But this does not mean we should not relax suppression til we get 100% immunity via infection or vaccines. Suppression has a cost. As an economist, we need to weigh the marginal benefit of suppression against the cost. 13/
When doing that calculation, there is no magical value associated with herd immunity. It is possible that the marginal benefit of suppression or vaccination is greater than the cost even if the reproductive rate is below 1! 14/
By contrast the GBD folks are making the argument that the marginal benefit of suppression < the marginal cost even when Rt > 1. I'm not taking sides. I am saying it's an empirical question & epidemiologists/economists have a framework to think about the problem. 15/
Bottom line: herd immunity should not be our only guide. Be wary of any herd immunity estimates. Assume wide confidence intervals.
It doesn't mean we shouldn't try, esp is relevant for policy. But keep a large salt shaker handy.
Whew...nice to get that off my chest. 16/
It doesn't mean we shouldn't try, esp is relevant for policy. But keep a large salt shaker handy.
Whew...nice to get that off my chest. 16/
@nebuer42 @paulnovosad @thesamasher @satejsoman @DanielJHemel @profmohanan @anuacharya @michaelmina_lab @toxvaerd1 @tylercowen @sarahcobey @maciekboni @jonathangruber1 @ashishgoel @arpitrage @NeelanjanSircar @ethanbdm @cblatts @trvrb @kakape @srajagopalan @BallouxFrancois
Why do I think all this? I've been working w/ a few amazing teams trying to forecast COVID in India, Indonesia using epi models, doing serological surveys in India, developing vaccine allocation plans in a few countries. The issues above have come up throughout. 17/17
