In a small simulation study I am doing, I am estimating the accuracy of the SE between two models by plotting the distribution of replication SEs and comparing that replication mean SE to an observed SE. I am comparing the SE between two models for eight different language tests, and so I am looking at a total of sixteen different comparisons all in all. However, in almost all cases (about 15/16), the observed SE is less than the mean replication SE (of 100 replications). This strikes me as being a bit odd. Then it struck me that this might be due to the "EDFILE=" line that was in the control file that is being replicated. I marked several misfitting responses as missing in that original control file and resampled with replacement from that original file--with the EDFILE line left in.
My question is: Should replications be performed on the original/initial control file with or without the same responses marked as misfitting? Should the EDFILE= line be left in the control file during replications?
Thanks a million. So much of my testing education has been through your .PDF tutorials, Winsteps' Help menu, and this forum; very much appreciated.