search

philchalmers / mirt

Multidimensional item response theory
https://philchalmers.github.io/mirt/
201 stars 75 forks source link

Strange results: coefficients of an item do not correspond to its itemplot... #173

Closed alinaoshina closed 4 years ago

alinaoshina commented 4 years ago

Hi, I fitted a 3PL model to a dichotomous 27-item spatial reasoning test data. the M2 function shows nice fit, even though model fit terminated after 500 iterations... I noticed an item with strange parameters: a1 = 5.74216; d = -2.247903; g = 0.3263463. But when I created an itemplot of this item, it showed a very steep curve indeed, but the steepest point somewhere around theta = 0.2, not -2.25 at all... I checked several times that I do not mix that item up with any other item... this question is also not supposed to be an easy question.... item trace lines this is item SRT_Q19... What am I doing wrong? the graph is not consistent with the coefs.... Thanks and sorry for the bother...

philchalmers commented 4 years ago

d is an intercept parameter, not a difficulty parameter. It represents the logit location when $\theta = 0$ rather than the point at which $P(y | \theta)$ inflects. If you want the classical parameterizations use coef(mod, IRTpars = TRUE). HTH.

alinaoshina commented 4 years ago

Thank you so much, Phil. That helps a lot. All the best, Alina

On Thu, Jan 16, 2020 at 1:12 AM Phil Chalmers notifications@github.com wrote:

Closed #173 https://github.com/philchalmers/mirt/issues/173.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/philchalmers/mirt/issues/173?email_source=notifications&email_token=AJINUATDLU4SYQ4CY7HQ2ZLQ56J4FA5CNFSM4KHK5OSKYY3PNVWWK3TUL52HS4DFWZEXG43VMVCXMZLOORHG65DJMZUWGYLUNFXW5KTDN5WW2ZLOORPWSZGOV763UGI#event-2952641049, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJINUAXCGT5QZZ53RRQ5MB3Q56J4FANCNFSM4KHK5OSA .