A learning-based account of fluid intelligence from the perspective of the position effect

Abstract

The current study addresses the question of whether performance on fluid intelligence tests involves learning processes by employing an approach emanating from research into the position effect of psychometric scales. This approach enables the modeling of learning processes that may occur while completing the items of an intelligence test. We analyzed the data of Raven's Advanced Progressive Matrices (Raven's Matrices) collected from a sample of 220 participants. Fixed-links models were applied for decomposing the variances and covariances of Raven's Matrices into a position component associated with the position effect, and a constant component independent of the item positions. These two components were linked to associative and complex learning. Results indicated that the two types of learning accounted for 66% of the latent variance of Raven's Matrices. Complex learning displayed an especially strong association with the position component. It is concluded that learning processes are an important ingredient of fluid intelligence.