Wednesday, May 6, 2020

5/5/20: Sweden v Denmark: Covid19 experiences and outruns

For those interested, there's an ongoing debate about the benefits and costs of two different approaches to dealing with the Covid19 pandemic that are being contrasted in the case of Sweden (low level of restrictions) and Denmark (high level of restrictions). The two countries offer a decent 'natural experiment' data, due to their physical, cultural, historical and socio-economic proximities.

Peter Turchin dissects the evidence on the outcomes here: in a very readable and, yet, empirically rigorous analysis.

The chart above is the key, although not the only source of the insights. Lines represent a fitted model, while points represent actual data.

What is notable in the above (some of it is in Peter's post, some is not) are the following features of the data:

  1. Death rates models in Denmark trail below those in Sweden, albeit the two converge into late April and reverse in early May. We do not know why, though Peter identifies one specific potential cause: slower and lower rate of testing in Sweden. Another potential cause can be the duration of treatment differences between the two countries. A third potential one, differences in vintage/strand of the virus. Etc...
  2. Actual death rates uptick in Denmark around May 1 seem to be relative outliers to the Denmark data (we do not know why, nor do we know if these are going to become a 'new normal' or a 'new trend'). These outliers are certainly responsible for the trend lines reversals.
  3. Actual death rates in Sweden are massively more volatile than those in Denmark. This volatility is most evident in April. This should imply serious differences in the accuracy/precision of both models, with Swedish model potentially down-weighing these upward outliers (this depends on the model used, of course).
The rest of conclusions are down to you, folks.

No comments: