Intraklassen

Intraklassen, in statistics commonly referred to as the intraclass correlation coefficient (ICC), is a measure of reliability that evaluates how strongly units within the same group resemble each other. It is used when data are clustered or when ratings are produced by different observers, to determine how much of the total variance arises from differences between groups versus variation within groups.

There are several forms of the ICC, depending on the statistical model and the question at hand (for example one-way random effects, two-way random effects, or two-way mixed effects). The ICC is typically estimated from analysis of variance components and reflects the proportion of total variance that is attributable to between-group differences. The coefficient generally ranges from 0 to 1, with higher values indicating greater similarity within groups. In some models, negative ICC estimates can occur, often indicating model misfit or greater within-group variability than between-group variability; such values are usually interpreted as zero reliability.

Interpretation guidelines vary by field, but common benchmarks classify ICC values as poor (<0.5), moderate (0.5–0.75), good (0.75–0.9), and excellent (>0.9) reliability, particularly for ratings and measurements intended to be interchangeable. Applications include psychometrics, education (consistency of test scores across raters), medicine (imaging or pathology ratings), and any research with clustered or hierarchical data.

Limitations of the ICC include sensitivity to the chosen model, the number of raters, and the range of scores in the sample. Reporting should include the model used and confidence intervals. See also reliability, interrater reliability, ANOVA, and variance components.