Interpretation of Spurious Differential Item Functioning

This article introduces a counterfactual discussion on the difficulty in transforming statistical evidence of differential item functioning (DIF) into the substantive interpretation of item fairnessor bias. Presented causal graphs that describe how the data were generated and the accompanying graphical explanation provide an intuitive explanation to understand which DIF should be considered genuine or spurious. In particular, the article presents three scenarios where spurious DIF is structurally embedded and will be detected by various statistical methods. The first scenario considers a case where an unknown common cause produces a spurious association while the other two scenarios consider cases where a collider variable plays a role in inducing an unexpected type of spurious DIF. Findings about colliders, especially due to sample selection,can be useful for resolving some confusion in DIF applications when analyzing a selected sample such as military data.