1. The case of RR, which means a correct answer is never erased, was excluded in the modeling. In other words, the presence of RR becomes missing in recoded data matrices. In our experience, the proportion of RR is unignorable, and the large proportion of missing data eventually influences the precision of parameter estimation in later analyses.
2. Although the authors believe that considering guessing is nonsense when reviewing initial responses is allowed, the assumption is still debatable. I think the main reason of ignoring randomly guessing is merely the simplicity of 2PLM compared to 3PLM.
3. The connection between the diagnostic tool for irregular response and cheating behavior is rather weak. Conceptually, erasing answers cannot directly link to cheating!