The aim of the paper is to investigate the effect of the exclusion of essential attribute or the inclusion of superfluous attribute on attribute estimation. The CDM focus on the estimation of person’s latent attributes rather than the item parameters. The RSM combines the latent trait in their model for auxiliary information. The idea of RSM is to transform person’s knowledge state and ideal response pattern into centroids on multidimensional space. With the error considered, the range of the centroid was constructed. Thus, it is easy and intuitive to identify the person’s attributes.
Simulation was conducted to investigate the effect of inclusion or exclusion. The main results show that the exclusion of essential attribute tends to decrease the accuracy of classification, but oppositely for inclusion of superfluous attribute. When an attribute was required for each item, it makes nearly no effect on attribute estimation whether the attribute was excluded or included. It can be noted that the influence on accuracy is slight when the attribute required among items is less or very often. The RMSE and bias are small when essential attribute was excluded.
Qs:
1. Random specification of exclusion or inclusion of attribute is more practical.
2. How to decide which CDM to be chosen to study? Why chose RSM here?
3. Any important reason to study the effect of exclusion or inclusion?
4. It seems it should include as more attributes into Q matrix as much, even the unknown attributes were included?