Re: Odd Information Curves

[Phil Chalmers]

Hi Conal,

There's nothing really wrong with these plots, it just means that the item intercepts are somewhat separated from each other, which gives this multi-camel-hump effect. So it's no cause for concern, and just represents a fact of life. That said, if the sampling variability of these intercept terms is too high due to data spareness, then yes collapsing would be a viable option. Cheers. 

Phil

On Tue, Jul 25, 2017 at 12:37 AM, Conal Monaghan <conal.m...@gmail.com> wrote:

Hi,

      I have been running GRMs on 7 point Likert data, there are 3 items worded for the trait of interest, and 3 items worded in the opposite direction and reverse scored (denoted by "R"). When looking at the item level plots, the shapes are not smooth, instead taking on a more stepped hill look (see attachment "Item Information Curves"). I was wondering if anyone has come across this before and hypotheses about causes and corrections (if an issue). Below is the code and respective output attached to the post. The response distributions of several of the items are positively skewed. Potential ways forward are to collapse extreme categories (i.e. response options 1 and 7) or to keep model how it is, if this represents the data appropriately. I cannot create a minimally reproducable dataset given I am unsure of the issue here. 

plot(IRT.Model, type = 'infotrace', facet=FALSE, theta_lim = c(-4, 4), npts = 1000)

 > Results in "Item Information Curves"

plot(IRT.Model, type = 'infoSE', facet=FALSE, theta_lim = c(-4, 4))

 > Results in "Scale Information Curve"

plot(IRT.Model, type = 'trace', theta_lim = c(-3, 4))

> Results in "Thresholds"

cbind(coef(IRT.Model, simplify=TRUE,IRTpars=TRUE)$item ,(itemfit(IRT.Model))[,2:4])

>

                  a          b1             b2              b3               b4             b5             b6         S_X2      df.S_X2       p.S_X2

7   1.555222 -2.065223 -0.6449500 0.03970031 0.7035501 1.893802 3.678467 74.03558      69         0.31735609

8   2.282830 -1.176583  0.1519065 0.87046416 1.3217755 2.449030 3.457850 76.80347      61         0.08344321

9   1.596773 -1.775203 -0.2441585 0.46933072 0.9031817 2.335987 4.447076 69.80151      70         0.48420458

10R 1.702901 -1.207227  0.4942209 1.34162018 2.2339415 3.262145 4.534759 75.97628      60      0.07991436

11R 2.105054 -1.703085 -0.3368991 0.39002851 0.8202274 1.640103 2.906099 97.94722      64      0.00405411

12R 2.016935 -1.697292 -0.2046045 0.54051518 1.1768065 2.213889 3.763472 64.37722      61      0.35926525

Q3 <- residuals(IRT.Model , type = "Q3", digits = 3, method = "ML") 

cor.plot(mat.sort(Q3),TRUE,zlim=c(-.7,.7), 

         main="Q3 Correlations (sorted by LD size)",  n = 10)

# Tests for local dependence. I have coloured the graph to make troublesome ones visible. Dark Blue and red should have attention paid. 

> Results as "LD".

Kind Regards,

       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+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.