3 ms·

Wait, what about this: "A common challenge in studies of COVID-19 is that differences between people who have vs. have not been ill could relate to premorbid d

Wait, what about this: "A common challenge in studies of COVID-19 is that differences between people who have vs. have not been ill could relate to premorbid differences. To address this issue, a linear model was trained on the broader independent GBIT dataset (N = 269,264) to predict general cognitive performance based on age (to the third order), sex, handedness, ethnicity, first language, country of residence, occupational status and earnings. "

dash2 5y ago
That's ridiculously bad. In effect their logic is: "We controlled for various things in the original regression, but hey, there still could be unmeasured confounders that make people who got COVID score less on IQ tests, even before they got COVID! (Like coming from a poorer background, other health vulnerabilities etc. etc. etc.) Unfortunately, we don't have a measure of IQ from before they got COVID. So, we estimated one using a set of variables that we do have measures for!" Erm... if those extra variables predict IQ, then why not just add them as controls? And of course, if you do so, then you'll still hit the problem that there are plenty of potential unmeasured confounders out there. This is just a silly way to pretend you've controlled for something, when in fact you can't. The quality of statistics in medicine is so bad. Disgracefully bad. In particular, the Lancet seems to be a serial offender.