| Title: | Robust Rank-Based Tests |
|---|---|
| Description: | Implements two-sample tests for paired data with missing values (Fong, Huang, Lemos and McElrath 2018, Biostatics, <doi:10.1093/biostatistics/kxx039>) and modified Wilcoxon-Mann-Whitney two sample location test, also known as the Fligner-Policello test. |
| Authors: | Youyi Fong <[email protected]> |
| Maintainer: | Youyi Fong <[email protected]> |
| License: | GPL-2 |
| Version: | 2024.1-28 |
| Built: | 2024-10-25 02:44:18 UTC |
| Source: | https://github.com/cran/robustrank |
Performs simulations to compare the power of different tests
choose.test(Xpaired, Ypaired, Xextra = NULL, Yextra = NULL, mc.rep = 1000)choose.test(Xpaired, Ypaired, Xextra = NULL, Yextra = NULL, mc.rep = 1000)
Xpaired |
Xpaired |
Ypaired |
Ypaired |
Xextra |
Xextra |
Yextra |
Yextra |
mc.rep |
mc.rep |
# There are unpaired observations from both samples dat=sim.partially.matched(m=20,n.x=40,n.y=5,distr="normal", params=c(loc.2=.8,rho=.1,scale.2=1),seed=1) choose.test(dat$X, dat$Y, dat$Xprime, dat$Yprime) ## There are unpaired observations from only one sample #dat=sim.partially.matched(m=20,n.x=0,n.y=10,distr="normal", # params=c(loc.2=.5,rho=.8,scale.2=1),seed=1) #choose.test(dat$X, dat$Y, dat$Xprime, dat$Yprime)# There are unpaired observations from both samples dat=sim.partially.matched(m=20,n.x=40,n.y=5,distr="normal", params=c(loc.2=.8,rho=.1,scale.2=1),seed=1) choose.test(dat$X, dat$Y, dat$Xprime, dat$Yprime) ## There are unpaired observations from only one sample #dat=sim.partially.matched(m=20,n.x=0,n.y=10,distr="normal", # params=c(loc.2=.5,rho=.8,scale.2=1),seed=1) #choose.test(dat$X, dat$Y, dat$Xprime, dat$Yprime)
from MTCT correlates study, C-section only
data("dat.mtct.rob")data("dat.mtct.rob")
A data frame with 55 observations on the following 2 variables.
ya numeric vector
V3_BioV3B_500a numeric vector
Fong and Huang (2016) Modified Wilcoxon-Mann-Whitney Test and Power against Strong Null.
Also known as the Fligner-Policello test.
mod.wmw.test(X, Y, alternative = c("two.sided", "less", "greater"), correct = TRUE, perm = NULL, mc.rep = 10000, method = c("combine", "comb2", "fp", "wmw", "fplarge", "nsm3"), verbose = FALSE, mode = c("test", "var"), useC = TRUE)mod.wmw.test(X, Y, alternative = c("two.sided", "less", "greater"), correct = TRUE, perm = NULL, mc.rep = 10000, method = c("combine", "comb2", "fp", "wmw", "fplarge", "nsm3"), verbose = FALSE, mode = c("test", "var"), useC = TRUE)
X |
Samples from population 1. |
Y |
Samples from population 2. |
alternative |
Directon of the alternative hypothesis. |
correct |
Whether to do continutiy correction. |
perm |
Boolean, whether to do permutation to get p-value or use normal approximation. See details. |
mc.rep |
Default number of replicates when doing permutation. See details. |
method |
For development. |
verbose |
For development. Print some debug info. |
mode |
For development. |
useC |
For development. Run C or R implementation. |
When perm is null, we will compute permutation-based p values if either sample size is less than 20 and compute normal approximation-based p values otherwise. When doing permuation, if the possible number of combinations is less than mc.rep, every possible configuration is done.
A p value for now.
manuscript in preperation
# Example 4.1, Hollander, Wolfe and Chicken (2014) Nonparameteric Statistics X <- c(0.80, 0.83, 1.89, 1.04, 1.45, 1.38, 1.91, 1.64, 0.73, 1.46) Y <- c(1.15, 0.88, 0.90, 0.74, 1.21) mod.wmw.test(X, Y, method="wmw", alternative="greater") mod.wmw.test(X, Y, method="combine", alternative="greater", verbose=1) # Section 4.1 Problem 1, Hollander et al. X=c(1651,1112,102.4,100,67.6,65.9,64.7,39.6,31.0) Y=c(48.1,48.0,45.5,41.7,35.4,34.3,32.4,29.1,27.3,18.9,6.6,5.2,4.7) mod.wmw.test(X, Y, method="wmw") mod.wmw.test(X, Y, method="combine", verbose=1) # Section 4.1 Problem 5, Hollander et al. X=c(12 ,44 ,34 ,14 ,9 ,19 ,156,23 ,13 ,11 ,47 ,26 ,14 ,33 ,15 ,62 ,5 ,8 ,0 ,154,146) Y=c(37,39,30,7,13,139, 45,25,16,146,94,16,23,1,290,169,62,145,36, 20, 13) mod.wmw.test(X, Y, method="wmw", alternative="less") mod.wmw.test(X, Y, method="combine", alternative="less", verbose=1) # Section 4.1 Problem 15, Hollander et al. X=c(0.19,0.14,0.02,0.44,0.37) Y=c(0.89,0.76,0.63,0.69,0.58,0.79,0.02,0.79) mod.wmw.test(X, Y, method="wmw") mod.wmw.test(X, Y, method="combine", verbose=1) # Table 4.7, Hollander et al. X=c(297,340,325,227,277,337,250,290) Y=c(293,291,289,430,510,353,318) mod.wmw.test(X, Y, method="wmw", alternative="less") mod.wmw.test(X, Y, method="combine", alternative="less", verbose=1)# Example 4.1, Hollander, Wolfe and Chicken (2014) Nonparameteric Statistics X <- c(0.80, 0.83, 1.89, 1.04, 1.45, 1.38, 1.91, 1.64, 0.73, 1.46) Y <- c(1.15, 0.88, 0.90, 0.74, 1.21) mod.wmw.test(X, Y, method="wmw", alternative="greater") mod.wmw.test(X, Y, method="combine", alternative="greater", verbose=1) # Section 4.1 Problem 1, Hollander et al. X=c(1651,1112,102.4,100,67.6,65.9,64.7,39.6,31.0) Y=c(48.1,48.0,45.5,41.7,35.4,34.3,32.4,29.1,27.3,18.9,6.6,5.2,4.7) mod.wmw.test(X, Y, method="wmw") mod.wmw.test(X, Y, method="combine", verbose=1) # Section 4.1 Problem 5, Hollander et al. X=c(12 ,44 ,34 ,14 ,9 ,19 ,156,23 ,13 ,11 ,47 ,26 ,14 ,33 ,15 ,62 ,5 ,8 ,0 ,154,146) Y=c(37,39,30,7,13,139, 45,25,16,146,94,16,23,1,290,169,62,145,36, 20, 13) mod.wmw.test(X, Y, method="wmw", alternative="less") mod.wmw.test(X, Y, method="combine", alternative="less", verbose=1) # Section 4.1 Problem 15, Hollander et al. X=c(0.19,0.14,0.02,0.44,0.37) Y=c(0.89,0.76,0.63,0.69,0.58,0.79,0.02,0.79) mod.wmw.test(X, Y, method="wmw") mod.wmw.test(X, Y, method="combine", verbose=1) # Table 4.7, Hollander et al. X=c(297,340,325,227,277,337,250,290) Y=c(293,291,289,430,510,353,318) mod.wmw.test(X, Y, method="wmw", alternative="less") mod.wmw.test(X, Y, method="combine", alternative="less", verbose=1)
Perform multinom test.
multinom.test(X, Y, alternative = c("two.sided", "less", "greater"), correct = FALSE, perm = NULL, mc.rep = 10000, method = c("exact.2", "large.0", "large", "exact", "exact.0", "exact.1", "exact.3"), verbose = FALSE, mode = c("test", "var"), useC = TRUE)multinom.test(X, Y, alternative = c("two.sided", "less", "greater"), correct = FALSE, perm = NULL, mc.rep = 10000, method = c("exact.2", "large.0", "large", "exact", "exact.0", "exact.1", "exact.3"), verbose = FALSE, mode = c("test", "var"), useC = TRUE)
X |
X |
Y |
Y |
alternative |
alternative |
correct |
correct |
perm |
perm |
mc.rep |
mc.rep |
method |
method |
verbose |
verbose |
mode |
mode |
useC |
useC |
Use permutation-based reference distribution to obtain p values for a test that combines WMW for paired data and WMW for unpaired data
mw.mw.2.perm(X, Y, Xprime, Yprime, .corr, mc.rep = 10000, alternative = c("two.sided", "less", "greater"), verbose = FALSE)mw.mw.2.perm(X, Y, Xprime, Yprime, .corr, mc.rep = 10000, alternative = c("two.sided", "less", "greater"), verbose = FALSE)
X |
X |
Y |
Y |
Xprime |
Xprime |
Yprime |
Yprime |
.corr |
.corr |
mc.rep |
mc.rep |
alternative |
alternative |
verbose |
verbose |
Performs a WMW-type test of the strong null for paired data.
pair.wmw.test(X, Y, alternative = c("two.sided", "less", "greater"), correct = TRUE, perm = NULL, mc.rep = 10000, method = c("exact.2", "large.0", "large", "exact", "exact.0", "exact.1", "exact.3"), verbose = FALSE, mode = c("test", "var"), p.method = NULL, useC = TRUE)pair.wmw.test(X, Y, alternative = c("two.sided", "less", "greater"), correct = TRUE, perm = NULL, mc.rep = 10000, method = c("exact.2", "large.0", "large", "exact", "exact.0", "exact.1", "exact.3"), verbose = FALSE, mode = c("test", "var"), p.method = NULL, useC = TRUE)
X |
Sample 1. |
Y |
Sample 2. |
alternative |
Alternative hypothesis. |
correct |
Whether to apply continuity correction. |
perm |
Whether to use permutation distribution or normal approximation to find p-value. See details. |
mc.rep |
Number of Monte Carlo replicates for permutation test. |
method |
Choices of test statistics. |
verbose |
Print debug message when positive. |
mode |
For development used. |
useC |
For development used. |
p.method |
Method for obtaining p values. |
When perm is NULL, if (min(m,n)>=20) normal approximatino is used to find p value, otherwise permutation test is used. When permutation test is used, if the number of possible permutations is less than mc.rep, a test statistic is computed for all permutations; otherwise, Monte Carlo is done.
P value for now.
Under prep.
dat=sim.partially.matched(m=15,n.x=0,n.y=20,distr="mixnormal",params=c(p.1=0.3,p.2=0.3),seed=1) X=dat$X; Y=dat$Y pair.wmw.test(X, Y, perm=TRUE, method="large.0", verbose=1) pair.wmw.test(X, Y, perm=FALSE, method="large.0", verbose=1)dat=sim.partially.matched(m=15,n.x=0,n.y=20,distr="mixnormal",params=c(p.1=0.3,p.2=0.3),seed=1) X=dat$X; Y=dat$Y pair.wmw.test(X, Y, perm=TRUE, method="large.0", verbose=1) pair.wmw.test(X, Y, perm=FALSE, method="large.0", verbose=1)
Performs rank-based two sample test for partially matched two sample data by combining information from matched and unmatched data
pm.wilcox.test(Xpaired, Ypaired, Xextra = NULL, Yextra = NULL, alternative = c("two.sided", "less", "greater"), method = c("SR-MW", "MW-MW", "all"), mode = c("test", "var", "power.study"), useC = FALSE, correct = NULL, verbose=FALSE)pm.wilcox.test(Xpaired, Ypaired, Xextra = NULL, Yextra = NULL, alternative = c("two.sided", "less", "greater"), method = c("SR-MW", "MW-MW", "all"), mode = c("test", "var", "power.study"), useC = FALSE, correct = NULL, verbose=FALSE)
Xpaired |
Xpaired |
Ypaired |
Ypaired |
Xextra |
Xextra |
Yextra |
Yextra |
alternative |
alternative |
method |
String. SR-MW is recommended, all is for development only. |
mode |
String. Do not change it to var, for developpment only. |
useC |
Boolean. Do not set it to TRUE, for development only. |
verbose |
verbose |
correct |
Continuity correction. |
If Xpaired and Ypaired have NAs, the corresponding unpaired data in Ypaired and Xpaired will be combined with Yextra and Xextra.
An htest object.
set.seed(1) z=rnorm(20, sd=0.5) # induces correlation between X and Y X=rnorm(20)+z Y=rnorm(20,mean=0.8)+z X[1:10]=NA boxplot(X,Y,names=c("X","Y")) pm.wilcox.test(X,Y) # for comparison wilcox.test(X,Y,paired=TRUE) wilcox.test(X,Y,paired=FALSE)# often a conservative test due to the correlation # no paired data Y1=Y Y1[11:20]=NA pm.wilcox.test(X,Y1) # should match the following wilcox.test(X,Y1,paired=FALSE) # only 1 pair of matched data Y1=Y Y1[12:20]=NA pm.wilcox.test(X,Y1)set.seed(1) z=rnorm(20, sd=0.5) # induces correlation between X and Y X=rnorm(20)+z Y=rnorm(20,mean=0.8)+z X[1:10]=NA boxplot(X,Y,names=c("X","Y")) pm.wilcox.test(X,Y) # for comparison wilcox.test(X,Y,paired=TRUE) wilcox.test(X,Y,paired=FALSE)# often a conservative test due to the correlation # no paired data Y1=Y Y1[11:20]=NA pm.wilcox.test(X,Y1) # should match the following wilcox.test(X,Y1,paired=FALSE) # only 1 pair of matched data Y1=Y Y1[12:20]=NA pm.wilcox.test(X,Y1)
sim.partially.matched generates partially matched two-sample data. for Monte Carlo studies. r2sample is a wrapper for sim.partially.matched and generates indepenent two-sample data.
sim.partially.matched(m, n.x, n.y, distr = c("normal","logistic","student","mixnormal","gamma","lognormal","beta", "uniform","hybrid1","hybrid2","doublexp"), params, seed) r2sample(m, n, distr = c("normal", "logistic", "student", "mixnormal"), params, seed) sim.paired.with.replicates(m, meanRatio, sdRatio, within.sd, type, hyp, distr, seed)sim.partially.matched(m, n.x, n.y, distr = c("normal","logistic","student","mixnormal","gamma","lognormal","beta", "uniform","hybrid1","hybrid2","doublexp"), params, seed) r2sample(m, n, distr = c("normal", "logistic", "student", "mixnormal"), params, seed) sim.paired.with.replicates(m, meanRatio, sdRatio, within.sd, type, hyp, distr, seed)
m |
Number of pairs. |
n |
Number of Ys. |
n.x |
Number of extra Xs. |
n.y |
Number of extra Ys. |
distr |
Distributions. |
params |
Named vector. See details. |
seed |
Seed for random number generator. |
meanRatio |
meanRatio |
sdRatio |
sdRatio |
within.sd |
within.sd |
type |
type |
hyp |
hyp |
If the distribution is in c("normal","student","logistic"), params should have three fields: loc.2, rho and scale.2; loc.1 is set to 0 and scale.1 is set to 1.
If the distribution is mixnormal, params should have three fields: p.1, p.2 and sd.n.
If the distribution is gamma, params should have fix fields: loc.2, shape.1, shape.2, rate.1, rate.2 and rho.
For details on bivariate logistic distribution, see rbilogistic
sim.partially.matched return a list with the following components:
X |
m sample 1 that pair with Y |
Y |
m sample 2 that pair with X |
Xprime |
n.x sample 1 |
Yprime |
n.y sample 2 |
r2sample returns a list with the following components:
X |
m sample 1 that are independent of Y |
Y |
n sample 2 that are independent of X |
dat=sim.partially.matched(m=10,n.x=5,n.y=4,distr="normal", params=c("loc.2"=0,"rho"=0,"scale.2"=1),seed=1) X=dat$X; Y=dat$Y; Yprime=dat$Yprime #dat=sim.partially.matched(m=10,n.x=5,n.y=4,distr="logistic", # params=c("loc.2"=0,"rho"=0,"scale.2"=1),seed=1) #X=dat$X; Y=dat$Y; Yprime=dat$Yprimedat=sim.partially.matched(m=10,n.x=5,n.y=4,distr="normal", params=c("loc.2"=0,"rho"=0,"scale.2"=1),seed=1) X=dat$X; Y=dat$Y; Yprime=dat$Yprime #dat=sim.partially.matched(m=10,n.x=5,n.y=4,distr="logistic", # params=c("loc.2"=0,"rho"=0,"scale.2"=1),seed=1) #X=dat$X; Y=dat$Y; Yprime=dat$Yprime
Perform WMW paired replicates test.
wmw.paired.replicates.test(X, Y, alternative = c("two.sided", "less", "greater"), correct = FALSE, perm = NULL, mc.rep = 10000, method = c("exact.2", "large.0", "large", "exact", "exact.0", "exact.1", "exact.3"), verbose = FALSE, mode = c("test", "var"), useC = TRUE)wmw.paired.replicates.test(X, Y, alternative = c("two.sided", "less", "greater"), correct = FALSE, perm = NULL, mc.rep = 10000, method = c("exact.2", "large.0", "large", "exact", "exact.0", "exact.1", "exact.3"), verbose = FALSE, mode = c("test", "var"), useC = TRUE)
X |
X |
Y |
Y |
alternative |
alternative |
correct |
correct |
perm |
perm |
mc.rep |
mc.rep |
method |
method |
verbose |
verbose |
mode |
mode |
useC |
useC |