Intraclass Correlation Coefficient (ICC)

ICC measures the consistency or agreement of continuous measurements made by multiple raters or across repeated measurement occasions. The standard reliability statistic for inter-rater and test-retest reliability.


Step 1 — Provide your data

Matrix of subjects x raters (numeric). Each row = one subject, each column = one rater/occasion.

Supports Excel Import, Sample Data Generator, and Manual Entry.


Step 2 — Choose ICC model and type

Model:

Model When to use
One-way random Different raters per subject
Two-way random Same raters, want to generalise
Two-way mixed Same fixed raters, only these matter

Type:

Type When to use
Single Using one rater's score in practice
Average Using average of all raters

Step 3 — Results

  • ICC value with specific form label (e.g. ICC(2,1))
  • 95% confidence interval
  • F-statistic with df and p-value
  • Reliability interpretation
  • ANOVA table (SS, df, MS)
  • Rater means and SDs

A UniversalChatBot is available for discussion.


Statistical methods used

Six ICC forms (Shrout & Fleiss, 1979):

  • ICC(1,1), ICC(1,k) — one-way
  • ICC(2,1), ICC(2,k) — two-way random (absolute agreement)
  • ICC(3,1), ICC(3,k) — two-way mixed (consistency)

Built on two-way ANOVA decomposition (MS_between, MS_within, MS_error, MS_raters).

Interpretation (Koo & Li, 2016):

ICC Label
>= 0.90 Excellent
0.75 - 0.90 Good
0.50 - 0.75 Moderate
0.00 - 0.50 Poor
< 0.00 No agreement

H0: ICC = 0. H1: ICC > 0. alpha = 0.05.