Differential Item Functioning

[Page 440]Differential item functioning (DIF) is a phenomenon in which a test or survey item performs differently based on the respondent who is using it, and this difference in performance is unrelated to the latent concept the item is intended to measure. For example, a test item might be more difficult for a male student than a female student even if both students have the same ability (and have gotten the same total score on the test). An example of DIF in a survey context would be, when using a 5-point Likert-type scale, one respondent’s “4” might be another respondent’s “5,” because of question or scale wording, though both respondents perceive the same latent value of the concept the scale is measuring.

The implications of DIF are vast. In the classic educational testing context, DIF can result in erroneous disparities between groups of students. For example, if a question intended to measure reading comprehension uses an anecdote about city life, rural students may answer incorrectly because of their unfamiliarity with the anecdote’s context, regardless of their latent reading comprehension abilities. In a political science context, two countries could receive similar scores on a latent regime trait—even if they diverge drastically along this trait—because respondents coding one country apply lower standards than those for the other. DIF is, thus, a fundamental concern in social science research that involves multiple respondents and one or more items. This entry details how DIF can vary in different survey contexts, offers an example in which survey design can affect DIF, and describes how to avoid DIF a priori, assess it post facto, and account for it in survey designs.