Back to Glossary
Glossary
When people from different groups but with the same trait level differ in their response to an item — a sign of possible item bias.
Differential item functioning (DIF) occurs when respondents from different groups who have the same underlying level of the trait nonetheless differ in how they respond to a particular item. The item behaves differently across groups after accounting for the trait it is meant to measure — a red flag for item bias.
DIF is detected with methods such as the Mantel-Haenszel procedure, logistic regression, or IRT-based comparisons. Finding DIF does not automatically make an item unfair — the difference has to be substantial and its cause understood — but items with strong, unexplained DIF are often revised or removed. DIF analysis is a key part of establishing measurement invariance and validity across diverse populations.