clusterPower: clusterPower: power analysis and sample size calculations for...

In clusterPower: Power Calculations for Cluster-Randomized and Cluster-Randomized Crossover Trials

Description

\loadmathjax

The clusterPower package is design for experiments with correlated, A.K.A. clustered observations. It contains many functions for calculating the power, sample size, and parameters necessary for achieving desired power and sample size from analytic equations. It also estimation of power via Monte Carlo methods via very flexible functions for a wide range of methods for which analytic methods or approximations may not exist.

Introduction

Most of the functions in this package are based on the generalized linear mixed model approach to the analysis of cluster randomized trials.

For example, for approximately normal outcomes in a simple parallel design, the data generating model is:

\mjsdeqn

y_ij|b_i = \beta_0 + \beta_1 x_ij + b_i + e_ij

where i indexes cluster and j indexes subject within cluster. In this case we assume random effect \mjseqnb_i, the cluster means, are distributed Normal \mjseqn(0, \sigma^2_b) and the residual error \mjseqne_ij ~ N(0,\sigma^2). In this special case, we define the "total variance" as \mjseqn\sigma^2_b + \sigma^2 and the Intracluster Correlation coefficient as \mjseqn\sigma^2_b)/(\sigma^2_b + \sigma^2). The ICC is useful in some special cases as a simplifying statistic, though it has no natural analogue in generalized linear mixed models when the distribution lacks a variance that is independent of the mean.

For example, for dichotomous outcomes, one convenient data generating model is:

\mjsdeqn

logit(Pr(y_ij|b_i = \beta_0 + \beta_1 x_ij + b_i

clusterPower documentation built on Jan. 29, 2021, 1:06 a.m.