compare the nonparametric item response functions against the 2PL item response functions

[Conal Monaghan]

Hi All,

      I am trying to work out how to compare graded response model plots to their corresponding nonparametric curves. This is because I wish to compare the nonparametric item response functions against the 2PL item response functions to assess their fit. I am choosing this approach because of troubles with Yen's Q3 for a larger dataset (N=4000) and thought a comparison of the two curves would be useful. I have linked the demonstration IRT model using the Science data below, which would be a great example dataset for this. Could we use the "KernSmoothIRT" package to produce the nonparametric curves then import the data into mirt() and plot them against the 2PL GRM ones? Alternatively, could the itemGAM() function do this?

library("mirt")

attach(Science)

Model1 <- mirt(Science, model = 1, itemtype = "graded")

cbind((coef(Model1, simplify=TRUE)$item), (itemfit(Model1)[,2:5]))

- Thanks in advance,

         Conal

[Phil Chalmers]

You'll need to extract out the probability tracelines directly to obtain the correct information. So in mirt, extract.item() and probtrace() will do the trick, and you just build whatever plot from there. Alternatively, save a plot object (obj <- itemplot(...), or whatever) and extract the relevant data from that object directly. I don't think you can do the same strategy with KernSmoothIRT because they use base R plots, so you'll have to dig a bit to see how they generate the traceline functions. And yes, you can extract the information from itemGAM() in the same way.

Phil

 

- Thanks in advance,

         Conal

--
You received this message because you are subscribed to the Google Groups "mirt-package" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mirt-package...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.