A General Approach for Testing for Correlated Errors in Longitudinal Data

Abstract

The present study utilized structural equation methods (LISREL) to estimate models of the pre-posttest paradigm, The data set comprised a group of 6 - 8th grade students involved in a gifted and talented program.

Two types of analyses were conducted. The first analysis was applied to test the validity of Bloom's taxonomy underlying performance on the achievement measure used in the program, the Ross Test of Higher Cognitive Thinking Skills. For the most part, the results demonstrated the existence of the structure, such that analysis skills were preordered with respect to synthesis and evaluation skills.

The second LISREL analysis was applied to assess the model of "best fit" among a set of alternative models that varied in the correlations specified among the measurement errors. There was a significant improvement in model fit when measurement errors were allowed to correlate, as compared to the zero correlation specification on the errors in the null structure. Generally speaking, a nonzero covariation specification between errors associated with all similar measures across the two occasions resulted in the most efficient estimate of ability change. The study pointed to the efficacy of LISREL-type analyses in longitudinal data.