Design Effect and Intra-cluster Correlation

Computes the Kish design effect and corresponding intra-cluster correlation for a single cluster-sampled variable

deff(y, cluster)

Arguments

y

variable to analyze

cluster

a variable whose unique values indicate cluster membership. Any type of variable is allowed.

Value

a vector with named elements n (total number of non-missing observations), clusters (number of clusters after deleting missing data), rho(intra-cluster correlation), and deff (design effect).

Author

Frank Harrell
Department of Biostatistics
Vanderbilt University
fh@fharrell.com

See also

bootcov, robcov

Examples

set.seed(1)
blood.pressure <- rnorm(1000, 120, 15)
clinic <- sample(letters, 1000, replace=TRUE)
deff(blood.pressure, clinic)
#>             n      clusters           rho          deff 
#>  1.000000e+03  2.600000e+01 -7.080501e-03  7.289867e-01