clusterPower (version 0.6.111)

power.sim.normal: Power simulations for cluster-randomized crossover study designs.

Description

These functions run simulations to calculate power for a given cluster-randomized crossover study design. The user can specify a function which runs the desired method of analysis. The function make.base.data() is not meant to be called directly by users, but is used in the data generation algorithms employed by the other functions.

Usage

power.sim.normal(n.sim = 10, effect.size, alpha = 0.05, n.clusters,
  n.periods, cluster.size, btw.clust.var, indiv.var = NULL, ICC = NULL,
  period.effect, period.var, estimation.function, permute = FALSE,
  verbose = FALSE)
power.sim.binomial(n.sim = 10, effect.size, alpha = 0.05, n.clusters,
  n.periods, cluster.size, btw.clust.var, period.effect, period.var,
  estimation.function, permute = FALSE, verbose = FALSE)
power.sim.poisson(n.sim = 10, effect.size, alpha = 0.05, n.clusters,
  n.periods, cluster.size, btw.clust.var, period.effect, period.var,
  estimation.function, at.risk.params, permute = FALSE, verbose = FALSE)
make.base.data(n.obs, n.clusters, cluster.size, n.periods)

Arguments

n.sim

number of datasets to simulate

effect.size

effect size, specified on the GLM link scale

alpha

desired type I error rate

n.clusters

number of clusters

n.periods

number of periods of study

cluster.size

either a numeric vector length one or of length(n.clusters) defining the number of individuals in each cluster.

btw.clust.var

the between-cluster variance

indiv.var

for normal outcomes only, the individual level variance

ICC

for normal outcomes only, the ICC may be specified instead of indiv.var

period.effect

period effect, on the link scale. See details.

period.var

the period effects are drawn from a normal distribution centered at period.effect with variance period.var. If period.var = 0, period effect is assumed to be the same for all periods.

estimation.function

function to run the data analysis

permute

indicator of whether to run permutation inferences. Defaults to FALSE.

verbose

indicator of whether to print out updates as the simulator is running. Defaults to FALSE

at.risk.params

a numeric vector of length 1 or 2. See details.

n.obs

(for make.base.data() only) - the total number of observations in the data set

Value

A list with the following components

results

matrix with columns "dataset", "beta.est", "beta.cil", "beta.cih", "reject.null", "pval.permute"

power

numeric, the estimated power

permute.power

numeric, the estimated power using the permutation inference

sample.data

a data frame containing the final simulated data set from the simulation run

Details

Runs the power simulation.

The period.effect parameter needs to be specified on the "link function scale". Meaning that if the average baseline risk for a Poisson model is 4/1000, then the period.effect should be specified as log(.004). Similarly, the baseline risk in a logistic model should be specified on the logit scale. The period effect can have length of 1, in which case it is treated as the average period effect across all periods, or it can have length equal to n.periods, in which case it is assumed that the investigator is specifying exact period effects s/he wishes to simulate.

For the Poisson simulations, at risk time is computed for each individual in the simulation. If at.risk.time is specified as a numeric vector of length 1, then the given number is the constant atrisk time which every individual is assumed to have. If length(at.risk.time)==2, the values are taken as the mean and size parameters of a negative binomial distribution (used as mu and size in the rnbinom() function) from which an at-risk time is drawn for each individual. Specifically, the at risk times are drawn as at.risk.time = 1 + rnbinom(1, size=at.risk.params[2], mu=at.risk.params[1]).

References

Turner RM, White IR, Croudace T, PIP Study Group. Analysis of cluster randomized cross-over trial data: a comparison of methods. Stat Med. 2007 Jan 30;26(2):274-89.

Examples

# NOT RUN {
a <- power.sim.normal(n.sim=10, effect.size=5, alpha=.05, n.clusters=2, n.periods=2, 
  cluster.size=20, btw.clust.var=5, ICC=1/20, period.effect=2, 
  estimation.function=fixed.effect, verbose=TRUE, period.var=0)
  
b <- power.sim.binomial(n.sim=10, effect.size=log(.75), alpha=.05, n.clusters=20, n.periods=2, 
  cluster.size=50, btw.clust.var=.2, period.effect=logit(.2), 
  estimation.function=random.effect, verbose=TRUE, period.var=0)

c <- power.sim.poisson(n.sim=10, effect.size=log(.75), alpha=.05, n.clusters=100, n.periods=2, 
  cluster.size=10, btw.clust.var=.4, period.effect=log(.2), 
  estimation.function=random.effect, verbose=TRUE, period.var=0, at.risk.params=10)
# }
# NOT RUN {
# }