Scrutable

https://www.theguardian.com/science/202 ... ates-study

There's a link to the preprint.

Is there actually any evidence for the idea that people get more colds in winter because of staying in with the windows closed?

From a quick skim of the paper:
- With an N of 300,000 everything is going to be statistically significant, so most of the effect sizes (b coefficients are not the best way to measure this, but they will do in the absence of better measures such as semipartial correlations) are not very big, especially for day length (Table 2), which one might expect to be of more interest for potential intervention than time of day.
- Table 5 seems to have a typo - the heading of the left-hand column says "Day length', but that was controlled for in Model 3, so I think it's "Time of day".
- I don't see any evidence that they fitted their model on part of the data and used it to predict the rest. There are quite a few covariates in there, which makes me wonder how many models they tried before settling on one they liked (and reporting confidence intervals as if they had only tried the ones that they settled on).