Skew reduces test-retest reliability.
William P. Dunlap, Rusan Chen, Tammy F. Greer · Journal of Applied Psychology · 1994
This article demonstrates that skew will reduce the test-retest reliability of otherwise normal data. Furthermore, the greater the skew the greater the decrease in reliability. In the first section, equations are derived for the test-retest reliability of lognormal data, a skewed distribution that can be retumed to normality by the log transformation. The second analytic treatment derives equations for reliability with power transformations (x a ) that will produce skew in originally normal data. Finally, analysis of test-retest correlations of actual skewed measures confirmed that if skew was minimized by transformation toward symmetry, the reliabilities increased; the increase in reliability was related to the extent of skew. Researchers are cautioned to consider transforming data that are skewed to obtain greater symmetry and, thus, improve the test-retest reliability of the resulting measure