Compute the design effect for correlated binary data

Share:

Description

This package computes the intraclass correlation, the design effect, and standard errors for the intercept for a binary-valued model with a group-level random effect, when the number of groups and number of observations per group are specified.

Usage

1
DEFFbinary(logitMean, sigma, NperGroup = NULL, Ngroups = NULL, Nsim = 1000)

Arguments

logitMean

The mean response, on the logit scale

sigma

The standard deviation of the group-level random effect

NperGroup

Number of observations per group, if unspecified only the ICC is computed

Ngroups

Number of groups, if unspecified standard errors are not computed.

Nsim

Number of simulations for computing the moments of the logistic-normal distribution.

Details

Consider the following model, where Y_ij is the jth observation from the ith group:

Y_ij|P_i ~ Bernoulli(P_i)

logit(P_i) = mu+ U_i

U_i ~ N(0, sigma^2)

logitMean is mu, the conditional mean on the logit scale. The sigma argument is sigma above.

Value

A vector with the following elements

ICC

The correlation cor(Y_ij,Y_ik))

DEFF

The design effect, (the total sample size divided by the effective sample size)

SE

The standard error of the estimate of mu

The first and second moments of P_i are returned as an attribute

Author(s)

Patrick Brown

References

Brown and Jiang (2009), "Intraclass Correlation and the Design Effect for Binary Random Effects Models", unpublished.

Examples

1
2
3
4
5
6
# Design effect with conditional mean 0.5, standard deviation 1,
# 10 groups and 10 observations per group
DEFFbinary(0, 1, NperGroup=10, Ngroups = 10, Nsim=10000)

# the same with conditional mean 0.1
DEFFbinary(log(0.1/0.9), 1, NperGroup=10, Ngroups = 10, Nsim=10000)

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.