Comparison of mirt and regDIF

[wbelzak]

Comparison of mirt and regDIF

[wbelzak]

Hi CW,

regDIF standardizes the predictor data by default (i.e., standardizing predictor data is recommended for regularization.) If you include stdz=F, you should get nearly identical results with mirt. I did this and got the following impact effects:

fit <- regDIF(item.data = res, pred.data = gender, stdz = F,
              ,item.type="2pl", num.tau = 5,control=list(tol=1e-7)
              ,anchor=1:itemnum,tau=0)
fit$impact

mean.cov1 1.0203
var.cov1 -0.1208

var.cov1 is the effect of the DIF predictor on the latent variance. So, var.cov1 = -0.1208 means that the female group has a variance of .88 (given the default variance of the reference group is 1, i.e., 1 - 0.1208 = .88).

I hope that helps.

Best,
Will