Abstract: | Uses principles of latent trait theory to derive conditions under which mean differences on an observed variable are indicative of mean differences on an underlying construct, focusing on the logistic curve as a model for describing the relation between the observed variable and construct. In this situation, and with normally distributed scores on the construct, a t test comparing groups on the observed variable produces correct inferences at the construct level only if group variances on the construct are homogeneous or the difficulty level of the test equals the mean ability level of the examinees. (27 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved) |