1/2 For example, data I am currently working with show that the latent "anxiety" construct is positively associated with alcohol use, yet negatively associated with some callous unemotionality items. This is in line with clinical theory, and neatly captured by multidim. IRT. https://twitter.com/Nate__Haines/status/1330646086004973568

2/2 Multidim. IRT offers one alpha (discriminability) parameter for each item and latent trait, such that multiple traits can influence a single item response. This is different than, say, assuming that correlation across measures arises purely from correlated latent traits.

(Also I just got these results and still need to carefully think through, etc., so don't take them too seriously lol I have to share progress with Twitter now that I do not have officemates for support )