I have a conceptual question regarding item anchor selection that seems to be so basic to the literature that authors of various online and print resources assume the reader understands. Namely, I am interested in the selection of appropriate item anchors in the case when all the items of the scale being investigated have significant DIF (as the lordif R package states when running my analyses) across groups. Obviously without appropriate anchors, I cannot run any models and the "purification" approach does not seem to work. I have attempted to pull in other items separate from the scale of interest and that should not meaningfully deviate across groups and receive the following error:
ERROR: The following items had negative slope parameters. 11ERROR: The following items had negative slope parameters (11).
Error in probgrm(theta, DISC[i], CB[i, ]) : slope is missing or negative In addition: Warning message: In lordif(mydata, GroupID, criterion = "Chisqr", alpha = 0.01, : bad anchor items; no anchor items will be used
Unfortunately, I have not been able to find a great deal of guidance on troubleshooting this predicament and this is almost certainly due to my very shallow understanding of IRT broadly and DIF specifically (considering my psychometric training was based in CCT).
I am hoping for one of the following from any individuals with a deeper understanding of the appropriate approach to exploring DIF:
- Examples of troubleshooting strategies utilized in similar situations
- Links to appropriate materials that will allow me to troubleshoot the issue myself
- Alternative R packages with more robust supporting literature (note: the scaling is polytomous).