The influence of missing data in estimates of parameters and hypothetical testing has been recognized in the literature and studied in the context of uniform differential item functioning (DIF) detection. Such research has suggested a complicated relationship among types and amounts of missing data, methods for dealing missing values, and uniform DIF detection. However, no studies have addressed this issue in the context of nonuniform DIF detection, which was the focus of the present study. The authors used a very comprehensive design to examine if the previous findings in uniform DIF could be generalized to the context of nonuniform DIF detection with manipulations on types and percentage of missing data, impact, magnitude of DIF and methods for dealing with missing data, along with popular methods for nonuniform DIF detection.
Findings of the present study suggested that listwise deletion might be deal with missing data along with nonuniform DIF detection because it yielded very similar results as complete data did. Also, complicated relations among manipulated variables indicated that none of the methods yielded the optimal results across all conditions. Overall, the author took into consideration most of situations and this paper is very organized and well structured. But some concerns need to be addressed to clarify confusions:
(1) It is not clear why the type I error rate was higher than .05 in the complete data.
(2) The paper missed the information about software used for IRTLR and CSIB.
(3) When the type I error was inflated, tests of power might not be as meaningful.
(4) I am not quite sure the criterion of “comparable” in the present study. For example, a difference value around ".02" was recognized as "comparable" in some conditions but not for others.