1. The assumption of Fisher transformation is that the paired item responses come from bivariate normal distribution. Is it held when responses were generated from CDM?
2. The misspecification is much more complex in real data. The resulsts of simulation seems useless for practitioners.