Which statistic is used to measure test-retest reliability?

Which statistic is used to measure test-retest reliability?

• A. Cronbach's Alpha
• B. ICC
• C. Kappa
• D. Pearson Correlation

Answer: (B)

Measuring how stable a measurement is across time. The intraclass correlation coefficient (ICC) is used for this because it assesses agreement of repeated measurements on the same subjects, not just how well the scores track in a linear way. ICC captures both consistency among subjects and how close the actual scores are to each other on different occasions, which is what test-retest reliability aims to quantify. If the second test simply tracked the first with a consistent bias (everyone’s score goes up by a fixed amount), Pearson correlation could still look strong, but ICC would reflect the lack of exact agreement depending on the chosen ICC model. Cronbach’s alpha focuses on internal consistency within a single test’s items, not stability over time; Kappa is for agreement on categorical outcomes; and while Pearson correlation can describe association between two sets of scores, it doesn’t fully capture the agreement needed for reliability across repeated measures.

Measuring how stable a measurement is across time. The intraclass correlation coefficient (ICC) is used for this because it assesses agreement of repeated measurements on the same subjects, not just how well the scores track in a linear way. ICC captures both consistency among subjects and how close the actual scores are to each other on different occasions, which is what test-retest reliability aims to quantify. If the second test simply tracked the first with a consistent bias (everyone’s score goes up by a fixed amount), Pearson correlation could still look strong, but ICC would reflect the lack of exact agreement depending on the chosen ICC model. Cronbach’s alpha focuses on internal consistency within a single test’s items, not stability over time; Kappa is for agreement on categorical outcomes; and while Pearson correlation can describe association between two sets of scores, it doesn’t fully capture the agreement needed for reliability across repeated measures.