Abstract
Potential sources of bias in a numerical aptitud test. In this work two problems derived from the application of numerical aptitud tests on a population which is made up of more than one academic grade (4 th and 6 th grade at primary level) are evaluated. On one hand, the different level of cognitive development and, on the other hand, the time period between administering and instruction of the tests. The interaction with the nature of the item may be the source of bias. In order to evaluate these hypotheses, all the methodology derived from the study of bias are utilisec (detection of the differential item functioning and the analysus of content). The results of the application of Lord's χ2 (3 parameter logistic model) and the Mantel-Haenszel statistic are compared. Furthermore, the DIMTEST is put to the test as a non-parametric techniqye for the evaluation of the essential unidimensionality.
