Day 372: But Sweden...

Last June, Nature published a much-cited paper modeling the efficacy of lockdowns vs. other non-medical interventions, "Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe" (where "intervention" means the government did it), and concluding that lockdowns, acting alone, saved three million lives. PlagueBlog will not be publishing the authors' names because the math is too embarassing, but Philippe Lemoine had no such compunctions when he tore the paper a new one back in a December blog post: "Lockdowns, science and voodoo magic".

Based on what I have seen by playing a bit more with simulations, it seems that the model tends to ascribe most of the reduction in R_t to whatever intervention took place after the peak of infections and was closest to that peak or, if no intervention took place after the peak, to the last intervention. This makes sense when you think about it. Indeed, if you assume that one intervention is responsible for the overwhelming majority of the reduction in R_t, the peak of the infections curve you infer is going to be located at the time this intervention took place. So both the infections curve and the deaths curve you infer are going to be shifted to the left or to the right depending on whether you assume that R_t collapsed sooner or later. Moreover, since both the infection curve and the distribution of the delay between infection and death are right-skewed, the death curve, on which the model is fit, is also going to be right-skewed. Now, when a curve is right-skewed, the fit is going to be worse if the peak of infections, hence also the peak of deaths, inferred by the model is to the left of the actual peak than if it’s to the right. So the model is better off ascribing most of the reduction in R_t to the intervention that happened closest to the peak of infections but after it, unless that intervention happened really too late after the peak.

It seems that it’s those purely mathematical considerations and the fact that lockdowns were implemented last and before or not too late after the peak of infections in every country, not the fact that lockdowns really are the only intervention that have a meaningful effect on transmission, which explain why the model reached that conclusion.

Well, not quite every country. Sweden never had a lockdown, so the model instead ascribes all non-medical efficacy to their last intervention, the cancellation of public events—a measure that had no effect elsewhere because it was followed by the supremely effective lockdowns, but was truly miraculous in Sweden. The paper doesn't show this explicitly because all European data has been pooled together, so Lemoine had to run his own unpooled calculations to see the Swedish miracle. Much of his post is devoted to the alleged malfeseance of the original authors in omitting this damning per-country calculation:

I actually think that not mentioning this fact about the country-specific effect in Sweden comes very close to scientific malpractice. It makes their main conclusion, which as [I] just noted can be seen to be implausible without any complicated modeling, very hard to maintain and I have a hard time believing they weren’t aware of that and that it’s not why they carefully avoided the topic in describing their results. In any case, what is clear is that, once you realize that the model was only able to find that no intervention except lockdowns had a meaningful effect on transmission by estimating a huge country-specific effect for Sweden, it becomes impossible to take that conclusion seriously.

The blog post was not the end of the story. A few weeks later, Nature finally published a Swedish response to the paper that it had been sitting on for six months and that largely agrees with Lemoine. (He discussed it and other responses in a second blog post.) It also inspired Lemoine to come up with his own theory about Swedish deaths, that the disease had progressed further there than elsewhere in Scandinavia. We at PlagueBlog still prefer the "dry tinder" theory of Sweden (that, rather than having an outsized number of deaths due to the lack of lockdowns, Sweden had more deaths because, after two mild flu seasons, they had an outsized population of fragile old people hanging on by the proverbial thread), but both may very well be true.

If you enjoyed that takedown, Lemoine also takes on the hydroxychlorquine question. In any event, PlagueBlog recommends that you check whether Sweden immediately disproves your theory before you publish it in a respectable scientific journal.