

DIF with categorical data in multilev... 

Message/Author 


I wish to test for differential item functioning (DIF) using MIMIC method for the following multilevel model. Details are: ESTIMATOR = WLSMV; TYPE = TWOLEVEL; Items are treated as categorical; Model = same 3 factor model at within and betweenlevels (each factor reflected by 4 items); Casual Indicator and possible DIF source is a continuous single indicator variable; As I understand it thresholds may only be estimated at the betweenlevel for this model and that with categorical data DIF refers to prediction of thresholds across levels of the DIF source which is my case is a continuous single indicator variable. How would uniform DIF manifest itself in such a model? Would we look for differences in item thresholds in two different models 1) the first model with each item being regressed onto the DIF source and 2) without the item being regressed onto the DIF source? And if the thresholds were different in each model how would they be compared? Am I cluthcing at straws here? 


With a continuous DIF source there would be a continuous variation in the threshold. You would run only the model with the item regressed on the DIF source variable. The threshold variation is  tau + b*x where x is the DIF source and b its withinlevel (assuming the DIF source is a withinlevel variable) regression slope. If x=0 represents the mean, the threshold is tau at that point. 

Back to top 

