Create t.test table with dplyr? - r

Question

Suppose I have data that looks like this:
set.seed(031915)
myDF <- data.frame(
Name= rep(c("A", "B"), times = c(10,10)),
Group = rep(c("treatment", "control", "treatment", "control"), times = c(5,5,5,5)),
X = c(rnorm(n=5,mean = .05, sd = .001), rnorm(n=5,mean = .02, sd = .001),
rnorm(n=5,mean = .08, sd = .02), rnorm(n=5,mean = .03, sd = .02))
)
I want to create a t.test table with a row for "A" and one for "B"
I can write my own function that does that:
ttestbyName <- function(Name) {
b <- t.test(myDF$X[myDF$Group == "treatment" & myDF$Name==Name],
myDF$X[myDF$Group == "control" & myDF$Name==Name],
conf.level = 0.90)
dataNameX <- data.frame(Name = Name,
treatment = round(b$estimate[[1]], digits = 4),
control = round(b$estimate[[2]], digits = 4),
CI = paste('(',round(b$conf.int[[1]],
digits = 4),', ',
round(b$conf.int[[2]],
digits = 4), ')',
sep=""),
pvalue = round(b$p.value, digits = 4),
ntreatment = nrow(myDF[myDF$Group == "treatment" & myDF$Name==Name,]),
ncontrol = nrow(myDF[myDF$Group == "control" & myDF$Name==Name,]))
}
library(parallel)
Test_by_Name <- mclapply(unique(myDF$Name), ttestbyName)
Test_by_Name <- do.call("rbind", Test_by_Name)
and the output looks like this:
Name treatment control CI pvalue ntreatment ncontrol
1 A 0.0500 0.0195 (0.0296, 0.0314) 0.0000 5 5
2 B 0.0654 0.0212 (0.0174, 0.071) 0.0161 5 5
I'm wondering if there is a cleaner way of doing this with dplyr. I thought about using groupby, but I'm a little lost.
Thanks!

Answer 1

Not much cleaner, but here's an improvement:
library(dplyr)
ttestbyName <- function(myName) {
bt <- filter(myDF, Group=="treatment", Name==myName)
bc <- filter(myDF, Group=="control", Name==myName)
b <- t.test(bt$X, bc$X, conf.level=0.90)
dataNameX <- data.frame(Name = myName,
treatment = round(b$estimate[[1]], digits = 4),
control = round(b$estimate[[2]], digits = 4),
CI = paste('(',round(b$conf.int[[1]],
digits = 4),', ',
round(b$conf.int[[2]],
digits = 4), ')',
sep=""),
pvalue = round(b$p.value, digits = 4),
ntreatment = nrow(bt), # changes only in
ncontrol = nrow(bc)) # these 2 nrow() args
}
You should really replace the do.call function with rbindlist from data.table:
library(data.table)
Test_by_Name <- lapply(unique(myDF$Name), ttestbyName)
Test_by_Name <- rbindlist(Test_by_Name)
or, even better, use the %>% pipes:
Test_by_Name <- myDF$Name %>%
unique %>%
lapply(., ttestbyName) %>%
rbindlist
> Test_by_Name
Name treatment control CI pvalue ntreatment ncontrol
1: A 0.0500 0.0195 (0.0296, 0.0314) 0.0000 5 5
2: B 0.0654 0.0212 (0.0174, 0.071) 0.0161 5 5

Answer 2

An old question, but the broom package has since been made available for this exact purpose (as well as other statistical tests):
library(broom)
library(dplyr)
myDF %>% group_by(Name) %>%
do(tidy(t.test(X~Group, data = .)))
Source: local data frame [2 x 9]
Groups: Name [2]
Name estimate estimate1 estimate2 statistic p.value
(fctr) (dbl) (dbl) (dbl) (dbl) (dbl)
1 A -0.03050475 0.01950384 0.05000860 -63.838440 1.195226e-09
2 B -0.04423181 0.02117864 0.06541046 -3.104927 1.613625e-02
Variables not shown: parameter (dbl), conf.low (dbl), conf.high (dbl)

Answer 3

library(tidyr)
library(dplyr)
myDF %>% group_by(Group) %>% mutate(rowname=1:n())%>%
spread(Group, X) %>%
group_by(Name) %>%
do(b = t.test(.$control, .$treatment)) %>%
mutate(
treatment = round(b[['estimate']][[2]], digits = 4),
control = round(b[['estimate']][[1]], digits = 4),
CI = paste0("(", paste(b[['conf.int']], collapse=", "), ")"),
pvalue = b[['p.value']]
)
# Name treatment control CI pvalue
#1 A 0.0500 0.0195 (-0.031677109707283, -0.0293323994902097) 1.195226e-09
#2 B 0.0654 0.0212 (-0.0775829100729602, -0.010880719830447) 1.613625e-02
You can add ncontrol, ntreatment manually.

Answer 4

You can do it with a custom t.test function and do:
my.t.test <- function(data, formula, ...)
{
tt <- t.test(formula=formula, data=data, ...)
ests <- tt$estimate
names(ests) <- sub("mean in group ()", "\\1",names(ests))
counts <- xtabs(formula[c(1,3)],data)
names(counts) <- paste0("n",names(counts))
cbind(
as.list(ests),
data.frame(
CI = paste0("(", paste(tt$conf.int, collapse=", "), ")"),
pvalue = tt$p.value,
stringsAsFactors=FALSE
),
as.list(counts)
)
}
myDF %>% group_by(Name) %>% do(my.t.test(.,X~Group))
Source: local data frame [2 x 7]
Groups: Name
Name control treatment CI pvalue ncontrol ntreatment
1 A 0.01950384 0.05000860 (-0.031677109707283, -0.0293323994902097) 1.195226e-09 5 5
2 B 0.02117864 0.06541046 (-0.0775829100729602, -0.010880719830447) 1.613625e-02 5 5