Test–retest bias, reliability, and regression equations for neuropsychological measures repeated over a 12–16-week period

• Published in: Journal of the International Neuropsychological Society 2001-07, Vol.7 (5), p.597-605
• Main Authors: SALINSKY, MARTIN C., STORZBACH, DANIEL, DODRILL, CARL B., BINDER, LAURENCE M.
• Format: Article
• Language: English
• Subjects: Adult; Bias; Confidence intervals; Data Interpretation, Statistical; Education; Epilepsy; Female; Health sciences; Humans; Male; Middle Aged; Neurology; Neuropsychological Tests - statistics & numerical data; Neuropsychology; Psychometrics; Reproducibility of Results; Stability; Test–retest reliability
• ISSN: 1355-6177; 1469-7661
• DOI: 10.1017/S1355617701755075

The interpretation of neurobehavioral change over time requires knowledge of the test–retest characteristics of the measures. Without this information it is not possible to distinguish a true change (i.e., one reflecting the occurrence or resolution of an intervening process) from that occurring on the basis of chance or systematic bias. We tested a group of 72 healthy young to middle aged adults twice over a 12-to-16-week interval in order to observe the change in scores over time when there was no known intervention. The test battery consisted of seven commonly used cognitive measures and the Profile of Mood States (POMS). Test–retest regression equations were calculated for each measure using initial performance, age, education, and a measure of general intellectual function (Wonderlic Personnel Test) as regressors. Test–retest correlations ranged from .39 (POMS Fatigue) to .89 (Digit Symbol). Cognitive measures generally yielded higher correlations than did the POMS. Univariate regressions based only on initial performance adequately predicted retest performance for the majority of measures. Age and education had a relatively minor influence. Practice effects and regression to the mean were common. These test–retest regression equations can be used to predict retest scores when there has been no known intervention. They can also be used to generate statistical statements regarding the significance of change in an individual's performance over a 12-to-16-week interval. (JINS, 2001, 7, 597–605.)