variancePartition (version 1.2.5)

ESS: Effective sample size

Description

Compute effective sample size based on correlation structure in linear mixed model

Usage

ESS(fit, method = "full")
"ESS"(fit, method = "full")

Arguments

fit

model fit from lmer()

method

"full" uses the full correlation structure of the model. The "approximate" method makes the simplifying assumption that the study has a mean of m samples in each of k groups, and computes m based on the study design. When the study design is evenly balanced (i.e. the assumption is met), this gives the same results as the "full" method.

Value

effective sample size for each random effect in the model

Details

Effective sample size calculations are based on: Liu, G., and Liang, K. Y. (1997). Sample size calculations for studies with correlated observations. Biometrics, 53(3), 937-47.

"full" method: if V_x = var(Y;x) is the variance-covariance matrix of Y, the response, based on the covariate x, then the effective sample size corresponding to this covariate is \Sigma_{i,j} (V_x^{-1})_{i,j}. In R notation, this is: sum(solve(V_x)).

"approximate" method: Letting m be the mean number of samples per group, k be the number of groups, and rho be the intraclass correlation, the effective sample size is m*k / (1+rho*(m-1))

Note that these values are equal when there are exactly m samples in each group. If m is only an average then this an approximation.

Examples

library(lme4)
data(varPartData)

# Linear mixed model
fit <- lmer( geneExpr[1,] ~ (1|Individual) + (1|Tissue) + Age, info)

# Effective sample size
ESS( fit )