This H2x2Factorial package implements the sample size
methods for hierarchical 2x2 factorial trials with unequal cluster
sizes. The sample size calculations support two types of treatment
effect estimands and five types of hypothesis tests based on the two
measures. The two estimands are named as the controlled effect and the
natural effect, as formally defined in Tian et al. (under review); The
hypotheses include (A1) test for the cluster-level controlled effect,
(A2) test for the individual-level controlled effect, (B1) test for the
cluster-level natural effect, (B2) test for the individual-level natural
effect, (C) interaction test for the two treatments, (D1) joint test for
the two controlled treatment effects, (D2) joint test for the two
natural treatment effects, (E1) intersection-union test for the two
controlled treatment effects, (E2) intersection-union test for the two
natural treatment effects. Finite-sample considerations are included for
the tests involving either cluster-level treatment effect, due to the
degree of freedom issues. Three functions are currently contained for
predicting the power or sample size based on given design parameters as
well as delivering illustrative tables or line plots. Specifically, the
calc.H2x2Factorial function calculates required number of
clusters for a specific test to achieve a given power, or predicts the
actual power given specified sample size resources, with or without
finite-sample considerations. The table.H2x2Factorial
function creates a data frame to show a series of sample size
predictions by providing varying mean cluster sizes, intraclass
correlation coefficients, or coefficient of variations of cluster sizes
(CV). The graph.H2x2Factorial function plots sample size
requirements under different CV in the form of the combinations of mean
cluster sizes and number of clusters. All of the hypothesis tests and
sample size methodologies are formalized in “Sample size calculation in
hierarchical 2x2 factorial trials with unequal cluster sizes” (under
review).
The released version of H2x2Factorial can be installed from CRAN with:
install.packages("H2x2Factorial")This is an example for predicting the required number of clusters based on fixed design parameters:
library(H2x2Factorial)
#> Warning: package 'H2x2Factorial' was built under R version 4.0.5
example("calc.H2x2Factorial")
#>
#> c.H22F> #Predict the actual power of a joint test when the number of clusters is 10
#> c.H22F> joint.power <- calc.H2x2Factorial(n_input=10,
#> c.H22F+ delta_x=0.2, delta_z=0.1,
#> c.H22F+ rho=0.1, CV=0.38,
#> c.H22F+ test="joint", correction=TRUE, seed_mix=123456, verbose=FALSE)
#>
#> c.H22F> print(joint.power)
#> [1] 0.2131This is an example for displaying a series of sample size predictions in a table format based on varying design parameters:
example("table.H2x2Factorial")
#>
#> t.H22F> #Make a result table by providing three mean cluster sizes, three CV, and three ICC
#> t.H22F> table.cluster <- table.H2x2Factorial(delta_x=0.2, delta_z=0.1,
#> t.H22F+ m_bar=c(10,50,100), CV=c(0, 0.3, 0.5), rho=c(0.01, 0.1),
#> t.H22F+ test="cluster", verbose=FALSE)
#>
#> t.H22F> table.cluster
#> m_bar rho CV n predicted power
#> 1 10 0.01 0.0 86 0.8020410
#> 2 10 0.01 0.3 87 0.8036148
#> 3 10 0.01 0.5 88 0.8027978
#> 4 10 0.10 0.0 150 0.8022800
#> 5 10 0.10 0.3 153 0.8011498
#> 6 10 0.10 0.5 160 0.8023522
#> 7 50 0.01 0.0 24 0.8100115
#> 8 50 0.01 0.3 24 0.8021486
#> 9 50 0.01 0.5 25 0.8036072
#> 10 50 0.10 0.0 93 0.8016170
#> 11 50 0.10 0.3 94 0.8012229
#> 12 50 0.10 0.5 96 0.8011854
#> 13 100 0.01 0.0 16 0.8093656
#> 14 100 0.01 0.3 16 0.8005201
#> 15 100 0.01 0.5 17 0.8078552
#> 16 100 0.10 0.0 86 0.8020410
#> 17 100 0.10 0.3 87 0.8038824
#> 18 100 0.10 0.5 88 0.8035513This is an example for plotting the sample size requirements under varying coefficients of variation of cluster sizes:
example("graph.H2x2Factorial")
#>
#> g.H22F> #Make a plot under the test for marginal cluster-level treatment effect
#> g.H22F> graph.H2x2Factorial(power=0.9, test="cluster", rho=0.1, verbose=FALSE)