ABSTRACT
Whenever multiple populations are sampled, there is potential for detecting differential item functioning, a lack of parameter invariance, or other violations of the homogeneity assumptions made in item response theory (IRT). Such findings may indicate that some items may not fit the IRT model well enough across some or all populations. If the populations are known and well defined, multiple-group IRT (Bock and Zimowski, 1997) models can be applied, and the item parameters can be estimated using equality constraints across groups while admitting “unique” item parameters where appropriate (Oliveri and von Davier, 2011).
