Thread by @benwansell

Since the authors of this study kindly make their data available, we can see that their multivariate finding largely depends on using absolute deaths and flights not per capita. To wit, on the left their results replicated, on the right the per capita version 1/n

https://twitter.com/rowlsmanthorpe/status/1357331196200574978

Now if we then log flights per capita we can recreate their 'finding' - see below. But I'm somewhat nervous about what that means about the role of outliers. Basically we have too few observations and too much instability of results for me to be comfortable with this. 2/n

The moral of this tale is single-shot cross-country regressions as policy guidance ain't the way to go. This is where we really need something with a stronger claim to causality. You can play with data yourself and see how I created these at github.com/benwansell/COV… 3/n

This btw is what the residualized plot (i.e. the plot of the relationship controlling for other variables) of log deaths per cap vs flights per cap looks like. I... wouldn't make policy based on this.

Finally, if you want to double check all this - their article is available at bmjopen.bmj.com/content/bmjope…
Again, it's great I was able to convert their data from pdf into R and then reproduce their Table 2. So kudos to them for that. But IMHO journalists should consume with caution.