Modeling often requires making assumptions and design choices explicit. Experimenters could benefit from such transparent thinking abt the design and knowledge of the range of experiments they are foregoing. They could also benefit from understanding how data produced is conditional on the design.

Yep, it comes down to whether the observations are properly weighted (which it seems to be arguing they weren't – will have to dig into original papers)

The 2020 papers also had a lot of dodgy (but testable) claims that were later quietly disowned...

There were also a whole bunch of direct-to-media and op-ed claims that followed (e.g. blogs.bmj.com/bmj/2020/09/...) – I rebutted to journalists and online at the time, but the obvious issue is that you simply can't infer this from fitting to a single case or death timeseries.

(Non-identifiability is the equivalent of looking at an equation 'A x B = 20' and claiming you know exactly what value A and B are.)

Scraped data from the plot ... gist.github.com/bbolker/d6a6... Scraping is imperfect because the plot is a fairly crappy JPG (hard to distinguish overlapping points). Dashed line is 156 particles per liter, nominal normal/superspreader cutoff (points fall on either side due to scraping error)

Kind of wild that Gladwell based a large chunk of his book and a marketing extract on that figure.

Thanks, I didn't realise it had been updated!

Meanwhile my youngest child has perfectly timed his wakes to be about 10 mins before my sunrise...