Tools / free calculator
Intraclass correlation calculator
Paste your grouped data and get the ICC, the between and within variance it comes from, and what it means for how much independent information you really hold. Works with unequal group sizes, which is where hand calculations usually go wrong.
What matters is the ICC together with how many observations sit in each group. A small ICC with large groups can still cut your effective sample size to a fraction of its row count.
Turn this into an effective sample size →In short
What is the intraclass correlation?
The share of total variation that sits between groups rather than within them. An ICC of 0 means group membership tells you nothing about an observation. An ICC of 1 means every observation in a group is identical, so the group carries only one piece of information. It is calculated here from a one-way random effects model, using the correct average group size when groups differ in size. Read it alongside your group sizes: the two together decide how much independent information you hold.
Common questions
Answers before you ask
Which ICC does this calculate?
ICC(1,1) from a one-way random effects model, which is the form used for clustering and design effects. It treats groups as a random sample from a population of groups. The reliability forms used for inter-rater agreement, ICC(2,1) and ICC(3,1), answer a different question and need a two-way model.
How does it handle groups of different sizes?
With the standard correction. Instead of the plain average group size it uses m0, computed from the total and the sum of squared group sizes, which gives the right answer when sizes vary. Using the plain average instead understates the ICC and is the most common hand-calculation error.
Can the ICC be negative?
Yes, when the within-group variance exceeds the between-group variance. It means observations in a group are less alike than two picked at random, which is usually sampling noise and sometimes a sign that the grouping variable is wrong. A negative value is normally reported as zero for design purposes.
What counts as a high ICC?
It depends entirely on the setting. Repeated measures from one person often run 0.4 to 0.7. Patients within a GP practice usually sit between 0.01 and 0.1. Pupils within a school commonly fall between 0.1 and 0.2. There is no universal threshold, and a small ICC is not automatically safe when groups are large.
Is my data stored?
No. Everything runs in your browser. Nothing you paste is transmitted or stored, and there is no account.
Keep reading
Related
Clustered data and the unit of analysis
What the ICC does to your sample size, and how to choose the right unit.
Effective sample size calculator
Turn an ICC and your group sizes into the number your data is really worth.
Mixed effects models
Where the ICC comes from, and how to model clustering properly.