Well, I hope that lockdown wasn't for naught, as the price of such a swingeing measure is very high indeed!
And we already know that the assumptions used in the first model were overestimates, that we were then over-prepared and that great numbers of people were left to die in order to free up space for potential deaths (potential deaths according to the 'reasonable worst case', no less).
The Imperial paper is here:
https://www.nature.com/articles/s41586-020-2405-7
Criticisms of the Imperial paper, on first pass:
- It looks at 11 countries only - why those countries in particular?
- "We simulate a hypothetical counterfactual scenario where reproduction number remains at starting levels to estimate the deaths that would have occurred without interventions. " - why should the reproduction number stay constant? Doesn't Farr's Law and experience suggest that this isn't the case? This decision accentuates the effect of any measures, in accordance with:
- "For each country, we model the number of infections, the number of deaths, and Rt (Figure 1). Rt is modelled as a piecewise constant function that changes only when an intervention occurs. " - doesn't this assume what it is trying to ascertain?
- "Our estimates imply that the populations in Europe are not close to herd immunity (~70% if R0 is 3.814). " - There is no evidence that the entire population is susceptible. This 70% stat further increases the impact of modelled interventions
- "Our model relies on fixed estimates of some epidemiological parameters such as the infection fatality rate " - they don't specify their IFR and (inferring from their counterfactual assertions) choose one that is at the higher end, giving a highly implausible counterfactual death toll in the UK of 500,000(!); the IFR decreases with increasing cases as the early cases include a higher proportion of the highly susceptible, as can be seen in the progress of the disease in the UK.
- Their counterfactual Swedish death toll is 10x greater than the actual; is it really plausible that their half-assed measures were that effective?
Modelling generally sucks, tho, as they inevitably privilege some parameters over others and slight adjustments of the variables can result in totally whacked-out figures.