Applied Analysis Note 7

Note 7: Draw an Item Characteristic Curve "by hand" using Stata

Below is some Stata code for a binary item ICC and a categorical item ICC. Both of these examples use runmplus.

The examples also use Bayesian plausible values of the latent trait scores as "observed" latent trait levels to plot expected and model-implied proportions. It's a visual check on the fit of the math used to generate the ICCs. But it is not exactly a check on model fit because the x-axis values of the plotted points are plausible values and are therefore model-derived, not observed. The y-axis values are observed. So let's call it an quasi-illustration of model fit.

The procedure demonstrated converts Mplus parameter estimates to the 2PL metric. Details on the conversions we use can be found here (pdf, scroll down to page 11).

The plots below use a custom Stata scheme (rnj) which is linked below. Another scheme (rnjwbg) is similar but uses a white background.