On task complexity and “simplex” correlation matrices

This article is concerned with task complexity and patterns of correlation mamas. The position advocated here is that an important initial step in the study of task complexity is the choice of an appropriate explicit model. It is argued that in general an empirical “simplex” correlation matrix between tasks involving different levels of complexity cannot be a safe basis for infemng an increase in task complexity. Given the multitude of factors affecting sample correlation coefficients, a model-building approach to the study of task complexity is recommended in place of preoccupation with simple comparisons of the magnitude of empirical correlations between administered tests. This structural modelling approach is applied to data from Stankov (1992). For the purposes of model testing and parameter estimation, the popular structural modelling programs LISREL and COSAN are employed.