Which ICC model should I use?

ICC(1) suits one-way designs where each subject is rated by a different set of randomly selected raters. ICC(2) suits two-way random designs where the same raters rate all subjects and raters are a random sample from a larger pool. ICC(3) suits two-way mixed designs where raters are the specific raters of interest (not a random sample).

What is the difference between Single and Average?

Single (ICC k,1) reflects the reliability of a single rater's measurement. Average (ICC k,k) uses the Spearman-Brown formula to reflect the reliability when the final score is the mean of all k raters. Average ICC is always higher and applies when you actually average raters.

What ICC value is acceptable?

Common thresholds: ICC 0.90 = excellent. These are guidelines — clinical measurement tools typically require ICC > 0.90 for interchangeable use.

How should I format my data?

Rows represent subjects (patients, items, trials); columns represent raters (judges, sessions, methods). Separate values with tabs — you can copy directly from Excel or Google Sheets. All cells must be filled; missing data is not supported.

What statistical test underlies the ICC?

ICC is derived from a mean-squares decomposition identical to a one-way or two-way ANOVA. The F-statistic tests whether subjects differ significantly from one another; a significant F (p < 0.05) indicates there is enough variance across subjects for ICC to be meaningful.

Yes. A negative ICC occurs when raters agree less than chance, for example when raters systematically disagree or when there is very little true variability across subjects relative to error. Treat a negative ICC as effectively zero.

Intraclass Correlation Calculator

Paste a subjects-by-raters data matrix to compute ICC(1), ICC(2), and ICC(3) with confidence intervals, F-tests, and plain-English interpretation.

Data Matrix

Rows = subjects, columns = raters. Paste from Excel or Google Sheets (tab-separated).

ICC Model

Measurement

Model Selection Guide

ICC(1) Each subject rated by different random raters

ICC(2) Same raters rate all subjects; raters are a random sample

ICC(3) Same raters rate all subjects; raters are the raters of interest

Reliability Thresholds (Koo & Li, 2016)

> 0.90 Excellent

0.75–0.90 Good

0.50–0.75 Moderate

< 0.50 Poor

Paste your data and click Calculate

ICC(2,k)

Subjects (n)

Raters (k)

F (subjects)

p-value

0.00.500.750.901.0

ANOVA Summary

Source	SS	df	MS

All ICC Estimates

Form	ICC	95% CI Lower	95% CI Upper

Copied!

Summary

The intraclass correlation coefficient (ICC) measures how consistently multiple raters or repeated measurements capture the same underlying value. Unlike Pearson r, ICC accounts for both correlation and agreement, making it the standard reliability metric in clinical and behavioral research. This calculator supports all six common ICC forms — ICC(1,1), ICC(1,k), ICC(2,1), ICC(2,k), ICC(3,1), ICC(3,k) — derived from one-way and two-way ANOVA mean squares with 95% confidence intervals.

How it works

Paste your data as a tab-separated matrix — rows are subjects, columns are raters.
Choose the ICC model: one-way random (ICC 1), two-way random (ICC 2), or two-way mixed (ICC 3).
Select whether you are averaging across raters (Average) or using a single rater (Single).
Click Calculate to run the one-way or two-way ANOVA decomposition.
Read the ICC value, 95% confidence interval, F-statistic, and reliability category.

Use cases

Assess inter-rater agreement between clinicians scoring a diagnostic instrument.
Evaluate test-retest reliability of a physiological measurement across sessions.
Validate a new rating scale before using it in research.
Report reliability statistics for a grant or journal submission.
Compare rater consistency before and after training.
Check whether automated and human scores can be used interchangeably.

Frequently Asked Questions

Last updated: 2026-06-11 · Reviewed by Nham Vu