2
$\begingroup$

I'm trying to use IRT to select the best items for a one-factor scale that has 20 items. I was wondering what I need to be looking at if I wanted to select 5 or so items from these items (for a short version of the scale). I would really appreciate any suggestions/references!

$\endgroup$
1
$\begingroup$

Higher slope values are better - this are discrimination parameters, the higher the slope, the greater the ability of the item to distinguish ability.

A range of difficulty parameters - or difficulty parameters in the area where you want the measure to have the most precision.

$\endgroup$
2
  • $\begingroup$ thanks! I'm wondering whether there are some criteria; Also, can the slopes be too high? some of my items have slopes going up to 6.5 and I'm not sure if it's always the higher the better... $\endgroup$ – llbia May 18 '20 at 13:34
  • $\begingroup$ Ah, yes, slopes can be too high. This is often caused by two items that are highly related (essentially the same item repeated twice) and is caused by a local dependency. $\endgroup$ – Jeremy Miles May 18 '20 at 15:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.