This is an unusual post. I was reluctant to make it, but it seems to be necessary. A few days ago, the paper https://arxiv.org/pdf/2003.12055.pdf appeared, with various suggestions for controlling the epidemic in India: a lockdown of 49 days or a lockdown of 21 days followed by 28 days etc. This paper seems to have gone viral, even appearing in neighbourhood WhatsApp groups.
I would like to point out that the conclusions in this paper are absurd and, moreover, it requires no expertise in modelling to understand their absurdity.
The question one should ask is: how do the authors arrive at their various precise figures for the length of the lockdown? The assumption is hidden away on Page 6: “This brings the number of infective below 10 where explicit contact tracing followed by quarantine may be successful in preventing a resurgence.” You read that right! Their “lockdown” is aimed at ensuring that only 10 infectious cases remain in the entire country!
Why 10? Why not 50? Indeed, why can’t a country do contact tracing when 500 people remain infectious? Their choice is completely arbitrary and depending on whatever threshold one sets for contact tracing, one can use their model to get any prediction for the length of the lockdown: starting from 0 days to an arbitrary long time.
Second, it should be clear that any model that suggests a lockdown can reduce the absolute number of infections to 10 in the whole country shouldn’t be taken seriously. At small numbers, these models are grainy and shouldn’t be used.
This, by itself, should be enough for everyone to stop taking the paper seriously at least insofar as serious epidemiological predictions are concerned.
But, let me mention a few additional minor points about the model itself. First, they make a simple mathematical error. They fail to correctly account for delays in the effect of the lockdown. So, in their model, the number of infections starts decaying as soon as the lockdown starts, which is absurd, even for a mathematical model. (Diseases have incubation periods, and so any lockdown takes time to act.) In particular, their predictions have already been falsified by more-recent data. Second, they use official data on the number of infections until the start of the lockdown to fit their parameters, but this has enormous error bars given our poor testing rate. So of what value are their numbers?
I am very sorry to have to debunk a paper on social media. But some time back, I had some questions about this paper, and another colleague communicated them to the authors. The authors have not responded, and I know that they have also not responded to other scientists who have tried to contact them. While refusing to engage the Indian scientific community, they have nevertheless propagated their findings to the news media. And, in various news-articles on this paper, I don’t find the authors making any attempt to point out the caveats in their analysis.
I feel this violates elementary scientific ethics. So, given the exceptional situation at hand — where this “study” may have real-life consequences — I felt it was important to make this post. It is a public post, and feel free to share it.
If the authors feel that I am misrepresenting their paper, I would still be happy to engage with them.