1. The semi-parametric model is appealing. I appreciate this approach. At least it allows the data to speak to some extent in contrast to the model-based method before.
2. Is it possible used for dimensionality assessment? Imagine a simple case. Assume two sets of items that are related to distinct contents (or dimension). They may be classified by experts or some statistical method. However, it may be somehow a few items (1 or 2) were classified into incorrect content. Using the same logic proposed in the study to detect the misclassified items would be useful. Yet the definition of item contribution (not shown in the paper) shall be developed by ourselves.
3. In Table 5, why are there estimated cutpoint for LR test?
4. In page 13, it says the Rasch tree needs no Bonferroni adjustment. Yet it shows the opposite results in Table 8, indicating that the Rasch tree perforemed better with Bonferroni method. Thus, use it or not?
5. Only one DIF item was used. It seems the study aimed at testing the differential test functioning (DTF) instead of DIF based on the framework of Rasch tree. The H0 is on covariate, which in turn, on the whole test-- DTF.