I am having difficulty finding information on the assumptions for the intraclass correlation. Can someone please tell me what they are?


There is no context to your request... here, I give an attempt to answer it in the context of regression models. More specifically, I shall refer to the usual linear mixed model.

Let $y_{ij}$ be the observation for subject $j \in \{ 1, \ldots{}, n_i\}$ from group $i \in \{1, \ldots{}, g\}$. The model I shall consider takes the form

$y_{ij} = \underbrace{\mu_{ij}}_{\textrm{fixed part}} + \underbrace{u_{i}}_{\textrm{random part}} + e_{ij}$,

under the assumption that $u_{i}$ is a realisation of a $N(0, \sigma^2_u)$ random variable, $e_{ij}$ is a realisation of a $N(0, \sigma^2)$ random variable, and under independence between these two random variables.

Observe that all subjects from the $i$th group share the same value for the random effect. Therefore, the random effect accounts for the association between observations from the same group. Mathematically, this can be seen by computing $\textrm{Corr}(Y_{ij}, Y_{i'j'})$ (see, e.g., here). When $ i = i'$, $\textrm{Corr}(Y_{ij}, Y_{i'j'}) = \textrm{Corr}(Y_{ij}, Y_{ij'}) > 0$ is known as the intraclass correlation.


Adding to @Ocram's answer, the assumption for intraclass correlation coefficient is that the variance of observation of a subject before a fixed effect is same as after the fixed effect.

For example, in an experiment measuring the (fixed) effect 'j' of a medicine on the blood pressure (BP= response 'y') of different subjects, the BP of a subject before and after the medicinal treatment are assumed to have equal variance.

Let '$y_{i1}$' be the observation for ith subject i∈{1,…,ni} with fixed effect value j=1 representing before medicinal treatment and '$y_{i2}$' the observation for ith subject with fixed effect value j=2 after medicinal treatment, then

Variance $\sigma^2$ = $Var(Y_{i1})$ = V$ar(Y_{i2}$)

This assumption is not applicable for measuring Pearson correlation coefficient.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.