Intraclass correlation coefficient (ICC) calculator
The intraclass correlation coefficient (ICC) measures how similar the scores of the same people are when measured two or more times (test-retest) or when several raters score them. Unlike Pearson’s correlation, it also detects when one measurement is systematically higher than the other. Paste the scores from Excel: one row per person and one column per measurement or rater.
The calculation runs in your browser. We don’t store what you type.
How it’s calculated
An analysis of variance is run with people (rows) and measurements or raters (columns), separating the variance between people from the error variance.
Absolute agreement, ICC(A,1): counts level differences between measurements as error. It is the one recommended for test-retest and inter-rater agreement (Koo & Li, 2016).
Consistency, ICC(C,1): only checks whether people keep their order, even if one measurement is higher. One-way, ICC(1,1): when each person was rated by different raters.
The "average" (k) versions apply if the final score you will use is the average of the measurements. The confidence interval follows McGraw and Wong (1996), like SPSS.
How to read the result
- According to Koo and Li (2016): below 0.50, poor; 0.50 to 0.75, moderate; 0.75 to 0.90, good; above 0.90, excellent.
- Look at the confidence interval too: with few people it can be very wide. Koo and Li recommend interpreting the lower bound.
- If the consistency ICC is much higher than the absolute agreement ICC, one measurement came out systematically higher or lower.
- Test-retest studies usually use 30 people or more, with 2 to 4 weeks between administrations.
Frequently asked questions
- Which ICC do I use for test-retest?
- Absolute agreement, single measures: ICC(A,1). In SPSS it is "Two-Way Mixed", type "Absolute Agreement", reading the "Single Measures" row.
- Why not use Pearson’s correlation?
- Because Pearson only checks whether the order holds. If everyone scores 5 points higher on the retest, Pearson gives 1 even though the scores don’t match. The absolute agreement ICC catches it.
- Does it match SPSS?
- Yes. It uses the same formulas (Shrout & Fleiss, 1979; McGraw & Wong, 1996). Values and intervals match SPSS, R (psych package) and Python (pingouin).
- Does it work for raters using a scale?
- Yes, if the scores are numeric (for example, 1 to 10). For unordered categories, use Cohen’s or Fleiss’ kappa.
Know what you want to ask?
Build the survey in SurvIQ, share it by link, WhatsApp or QR code, and watch results come in live. The Free plan is free forever.
Create a free survey