power.dunnett.test {PMCMRplus} | R Documentation |
Compute average per-pair power of Dunnetts's multiple comparison test with one control.
power.dunnett.test(n, groups, delta, within.var, sig.level = 0.05)
n |
Number of observations (per group) |
groups |
Number of groups (including control) |
delta |
true difference in means |
within.var |
Within group variance |
sig.level |
Significance level (Type I error probability) |
The function has implemented the following Eq. to estimate average per-pair power for two-sided tests:
1 - β = 1 - t( T_{α ρ v}, v, \mathrm{ncp}) + t(-T_{α ρ v}, v, \mathrm{ncp}),
with T_{α ρ v} the two-sided α quantile of the multivariate t-distribution, with v = k (n - 1) degree of freedom, k the number of groups and correlation matrix ρ_{ij} = 0.5 ~ (i \neq j).
The non-centrality parameter for the non-central student t-distribution is
\mathrm{ncp} = |Δ| / √{s_{\mathrm{in}}^2 ~ 2 / n }.
Object of class ‘power.htest
’,
a list of the arguments
(including the computed one) augmented with
method
and note
elements.
The results for power are seed depending.
The Eqs. were taken from Lecture 5, Determining Sample Size, Statistics 514, Fall 2015, Purdue University, IN, USA.
set.seed(113) power.dunnett.test(n = 9, groups = 5, delta = 30, within.var = 333.7) ## compare with t-test, bonferroni corrected power.t.test(n = 9, delta = 30, sd = sqrt(333.7), sig.level = 0.05 / 4) ## Not run: ## asymptotic Monte-Carlo power analysis set.seed(113) powerMCTests(mu = c(rep(0,4), 30), n = 9, parms = list(mean = 0, sd = sqrt(333.7)), test = "dunnettTest", alternative = "two.sided") ## End(Not run)