Hierarchical factor item response theory models for PIRLS: capturing clustering effects at multiple levels
This paper aims to look for a best model that takes into account of the item blocks which produce item clustering effects. It purposely compares a number of models on their fit to the PIRLS assessment data: a between –item two-dimensional model, and a bi-factor model with specific dimensions corresponding to item blocks and with two general dimensions. The results revealed that the bi-factor model with item blocks as specific dimensions provided the best fit than the uni-dimensional two-parameter logistic model and the two-dimensional two-parameter logistic model.
After the item blocks had been taken into account through the incorporation of specific dimensions corresponding to item blocks, it was evident that the model with a separate dimension for each of the two reading purposes provided a better fit than the corresponding model with only a single general dimension. This paper has successfully a message that the effect of item blocks can be attributed in part but not entirely to the fact that item blocks are clustered within reading purposes. It has implications for other secondary analysis work to be done on other large-sacle assessment where item blocks are used and individual differences on specific item blocks are of interest. One suggestion for further studies would be replication of analysis procedures into other datasets that also used the item blocks as item carrier to check if the same results can be observed.