Thread by @joel_c_miller

@mammuthus The effect he's looking at is real, but his analysis is not good.

I think my main criticism is that he's treating Sweden as a place with no interventions, even though behavior has changed a lot (and there are interventions in place).

@mammuthus I don't have time to really address it right now, but I do believe it's plausible that with interventions in place, in a country where 40% of households are single individual, herd immunity has been reached in Sweden.

@mammuthus I doubt that herd immunity would persist if people went back to normal.

And it's clear that in other places this herd immunity threshold is quite a bit higher.

@mammuthus For example, in both the UK and the US (which is what the Ferguson paper he's trying to criticise studied), it's clear that this threshold is higher.

I do not know what they would have predicted if they had calibrated their model to Sweden.

@mammuthus I would very much like to see how he has calculated R_0=2.4 for figures 4 and 5.

He goes into a lot of detailed discussion of what variation he assumes in the distribution, but shows no evidence of accounting for the impact of that variation on R_0.

@mammuthus I would also like to see evidence for the level of variation he assumes. Data on this exists for sexual networks, but respiratory networks appear to be more homogeneous. It would not surprise me if the conclusions are a consequence of overstating the level of heterogeneity.

@mammuthus To achieve herd immunity with R_0=2.4 and <20% infected requires quite extreme heterogeneity.

An alternative explanation is that as the disease spread, there was significant heterogeneity in behavioral response to avoiding exposure.

@mammuthus It is hard to get large heterogeneities for a disease that is spread simply by being in the same place.

If a large fraction are trying to avoid being in the same place as others, that would do it.

@mammuthus A couple other comments - This comment is just completely and totally wrong.

"The herd immunity threshold may be significantly higher in areas that ... are repeatedly reseeded from other areas."

@mammuthus A bigger crime is here: "the Swedish public health authority has surveyed the prevalence of infections... in Stockholm County, the earliest in Sweden hit by COVID-19. They... estimated that 17% of the population would have been infected by 11 April, rising to 25% by 1 May 2020."

@mammuthus To me this implies that they on 11 April they surveyed the population and on 1 May they surveyed the population.

This is not what happened.

@mammuthus Between 25 of Mar and 3 Apr, 707 people self-sampled for COVID-19, resulting in 18 positive tests. From this they inferred a positive fraction for the rest of the city on those dates, and from that inferred what fraction of cases are not reported.

@mammuthus Based on that, and the positive tests reported, they inferred a large fraction of unreported cases and tried to infer total proportion infected by Apr 11 and May 1.

I have issues with these methods, but different issues would push the error in different directions.

@mammuthus Regardless, if we take the measurement on face value, they certainly did not measure that 25% was infected on May 1. They have a pretty massive confidence interval. In one of the 3 scenarios they consider, it ranges from 11.5% to 50.7%.