ICC Calculator

Compute the ICC(2,1) intraclass correlation coefficient from your ANOVA mean squares to judge how reliable a repeated fitness measurement really is.

Quick Facts

Method
ICC(2,1) = (MSR − MSE) / (MSR + (k−1)·MSE + (k/n)(MSC − MSE))
Two-way random-effects, single-rater, absolute-agreement (Shrout & Fleiss, 1979).

Your Results

Calculated
ICC(2,1)
-
Single-measurement reliability
Reliability grade
-
Koo & Li (2016) benchmark
ICC(2,k)
-
Average of k measurements

Ready

Enter your ANOVA mean squares and click Calculate.

Frequently Asked Questions

What does the ICC value mean?
The ICC ranges from 0 to 1 and gives the proportion of total variance attributable to true differences between subjects rather than to measurement error. Using the Koo & Li (2016) benchmark, values below 0.50 indicate poor reliability, 0.50–0.75 moderate, 0.75–0.90 good, and above 0.90 excellent reliability.
Which ICC form does this calculator compute?
It computes ICC(2,1), the two-way random-effects, single-rater, absolute-agreement coefficient from Shrout & Fleiss (1979). It also reports the average-measures ICC(2,k). Enter the ANOVA mean squares for subjects (MSR), measurements/raters (MSC) and residual error (MSE), plus the number of subjects (n) and measurements (k).
What is the difference between ICC(2,1) and ICC(3,1)?
Both are two-way models, but ICC(2,1) treats the raters/sessions as a random sample and measures absolute agreement, so systematic differences between sessions (the MSC term) lower the score. ICC(3,1) is a mixed model measuring consistency: it ignores that systematic bias and answers only whether scores rise and fall together. Use ICC(2,1) for test-retest reliability where the actual value must be reproducible.
Why can an ICC be negative?
When MSE exceeds MSR the formula can return a small negative number, meaning the estimated between-subjects variance is essentially zero and the measurement carries no reliable signal. Negative estimates are conventionally reported as 0, which is what this calculator does.

What the intraclass correlation coefficient measures

The intraclass correlation coefficient (ICC) tells you how much of the variation in a repeated measurement is real, subject-to-subject variation versus how much is measurement noise. In sport and exercise science it is the standard way to report the test-retest reliability of a fitness assessment (does the same athlete get the same score on two different days?) and the inter-rater reliability of a rated test (do two coaches score the same jump the same way?). The ICC ranges from 0 (all the spread in your data is measurement error) to 1 (measurements are perfectly repeatable).

The formula this calculator uses

This tool computes ICC(2,1) — the two-way random-effects, single-rater, absolute-agreement coefficient from Shrout & Fleiss (1979), equivalent to ICC(A,1) in McGraw & Wong (1996). You run a two-way ANOVA on your ratings matrix (subjects in rows, repeated measurements or raters in columns), read off the three mean squares, and plug them in:

ICC(2,1) = (MSR − MSE) / [ MSR + (k−1)·MSE + (k/n)·(MSC − MSE) ]

  • MSR — mean square for rows (between-subjects). Large when athletes genuinely differ from one another.
  • MSC — mean square for columns (between-measurements/raters). Large when one testing session or one rater is systematically higher or lower than another; this systematic bias is what pulls an absolute-agreement ICC down.
  • MSE — residual mean square (error). Random, unexplained noise.
  • n — number of subjects (rows); k — number of measurements or raters (columns).

To report the reliability of the average of all k measurements rather than a single one, use the average-measures form ICC(2,k) = (MSR − MSE) / [ MSR + (MSC − MSE)/n ], which this calculator also returns. Averaging always raises reliability, so ICC(2,k) is never below ICC(2,1).

Interpreting the value

The most widely cited benchmarks come from Koo & Li (2016):

  • Below 0.50 — poor reliability; the measurement is too noisy to track individual change.
  • 0.50 to 0.75 — moderate reliability.
  • 0.75 to 0.90 — good reliability; acceptable for most applied testing.
  • Above 0.90 — excellent reliability; suitable for clinical decisions and detecting small individual changes.

Report the ICC together with its 95% confidence interval whenever possible: a point estimate of 0.85 from only 8 subjects can have a lower bound below 0.50, which changes how much you should trust it.