1. ARMSE is a strange for summary of results, although it is convenient. I still interested in the RMSE of individual item.
2. When it exist correlation between theta, the MIRT needs to estimate additional parameter than UIRT. So the MIRT will get slight higher estimation error when correlation is zero. It is well-known and confirmed in statistical theory. The simulation studies were redundant.
3. The GA is a kind of time-consuming algorithm. Other optimization method such as trust-region-reflective algorithm is more efficient.
4. I’m curious if ASSEST can estimate the standard errors of parameters under MIRT? If it can’t, the MCMC is more powerful than it.
5. We specify an item as unidimensional or multidimensional usually based on theory. In reality, how do we know that happened and formula (16) to correct it?