The studies proposed a non-parameter approach to classify the attribute patterns. The key of this non-parameter procedure is minimizing the distance between observed response and the ideal response, which are the response pattern corresponded to the attributes an examinee possesses and the attributes required for an item observed on Q matrix. The two distance measurement is adopted for simulation in this studies. They are weighted Hamming and penalized weighted Hamming distance. The magnitude of weight in the penalized weighted Hamming distance is decided by the type of item. The result showed that non-parameter performed as well as DINA, NIDA, and DINO in majority conditions. The guessing and slipping (noise), however, larger will lead the agreement of non-parameter to decline. The robustness of this approach also was investigated and the agreement held well in the small misspecification condition could be found.
1) All the result showed that the accuracy rate of classification for non-parameter method is lower than for parameter methods. The noise also affected the accuracy rate hugely. It seems the advantage of non parameter is only the time consuming.
2) An advantage of this process mentioned in the introduction is small sample size. However, the sample size didn't be manipulated in the simulation study.
3) In this approach, the Q matrix is still strongly relied. Whether we can apply this procedure to calculate the Q matrix via a small group of examinee whose attribute are known by interview?