I'm conducting an IRT analysis on 60 MCQs to estimate the item parameters (b, a, & c) as well as Standard Error of the Estimate (SEE). However, some items showed large positive threshold values of b-parameter (i.e. 547.98, 600.38, 423.47), and also, large negative threshold values of c-parameter (i.e. -201.59, -155.78). Consequently, the SEE values of these items also displayed similar values.
When you examined the univariate and bivariate statistics for these items, did anything stand out?
Daniel Wright, the 60 MCQs were obtained from a Mathematics SAT, whereas the purpose of the study is to establish the psychometric properties (a, b, & c) of the tests items, so as to identify items in the instrument to be retained for future use.
However, it was observed that 3 out of the 60 MCQs produced extreme positive b-parameter indices. Furthermore, the a-parameter indices of these items are below acceptable range (0.37, 0.02, 0.37) as enshrined by Baker. In addition, the c-parameter indices showed extreme negative values (-201.59, -14.41, -155.78).
How do i examine the univariate/bivariate statistics of these items? Thank you.
Before you did your 3PL model in IRT, you must have done univariate and bivariate procedures to examine the items (as well as simpler IRT models).
Honestly, I didn't know that i was supposed to run a univariate statistics on the data before computing my threshold values.
However, I'm presently studying the procedure to do so via excel. I really your input Sir.
If you don't see anything odd (like almost all accurate for an item, or near infinity log-odds ratios for bivariate), look up Heywood cases. But in general, don't start with a complex model without looking at the data. Or make friends with your data as I say here: Article Making friends with your data: Improving how statistics are ...
. Anything that you find in your IRT models you should be able to show why that arises from the log-odds ratio table.
I found this publication extremely helpful. Thank you and may God bless you more wisdom 🙏
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling