Machine learning psychometrics: Improved cognitive ability validity from supervised training on item level data
Andrew Cutler gave the talk since I was unable to attend.
Slides are here:
Live tweeted by Russell Warne:
Main result (preliminary):
Y axis is here coded as the fraction of the sumscore metric, which is in concordance for binary outcomes, and correlation for continuous outcomes. Thus, we see that ridge (l2 penalization) outperforms the others every time, and usually categorical item coding outperforms binary coding.
Video:
[forthcoming]
