A have evaluated DIF in a single group MIMC model and have evidence that 9 of the 19 items have significant DIF.
Traditionally, the ICCs are plotted for each group to highlight the DIF. However, when I try this with the PLOT3 command, summarizing the ICCs for each group with the define set function in the plot command, the ICCs are exactly the same for each group.
Is this because the test for DIF in the single group MIMC model tests only for the direct effect on the observed item (i.e., uniform DIF) and not the thershold or loading parameter?
It sounds like you want an item characteristic curve for an item for each of say 2 groups, where the groups are represented by a binary covariate. This can be done in the Mplus plot function. One of the last windows for the icc shows top right "Name a set of values" - here you give say "male" and then you set the value of the covariate gender at the male value; then do the same for female. Both the male and female icc will then be plotted in one graph as a function of the latent variable, showing the DIF.
You are correct and I have attempted to plot the ICCs this way, but the ICCs are identical in each group even in items with significant DIF. I was just trying to figure out why this might be and thought it might be related to the assumptions in a single group MIMC model? I do get different Plotted ICCs if I conduct a multiple group MIMC model where the equality constraints are lifted for threshold and slope parameters (but these aren't plotted on the same graph, but seperate graphs). I realize this is a different model, with different assumptions, but am just curious as to why I can't get DIF to appear in the Plots of the single group MIMC model?
This is the model set up I've used and is the final model. When I plot the ICC for u3 by gender (for example), I get the same exact ICC for each covariate set, but the output suggests DIF by evidence of significant direct effects of u ON x.
For example u3 is: u3 ON GENDER 0.301 0.087 3.455
MODEL: f1 BY u3*1 u5*1 u8* u11-x13*1 u15*1 u17*1 u20*1 u21*1; f2 BY u24-u28*1 u34-u36*1 u38*1; f1 ON Gender; f2 ON Gender; [f1@0f2@0]; f1@1; f2@1; u3 u5 u8 u11-x12 ON Gender; u24 u25 u27-u28 ON Gender;
I figured out the issue. I was misslabeling the covariates in the "name set value" screen (0 = male, 1 = female) when they were (1 and 2 in the data set). Sorry to bother you this! It's always something that simple.