Abstract
Didactic developments and differential item functioning. This paper sets out the problems that arise in any empirical research on bias: on the one hand, tha fact that the different procedures used to detect the differential item functioning (DIF) do not match exactly, and on the other hand, a lack of correspondence between the concept of DIF and the bias. These problems were analysed by studying the differential item functioning in a numerical ability tests. We examined the source of bias originated by the cognitive development and the inter val between instruction and administration of the test. The results obtained by Mantel-Haenszel statistic and the Lord χ2 procedure (under 2 and 3 parameter logistic models) were compared. Results show tha different conclusions can be drawn according to the procedure used to detect DIF.