I have the following data frame (df)
df <- data.frame( TTMENT = c("Group_1","Group_1","Group_1","Group_2","Group_2","Group_2","Group_3","Group_3","Group_3","Group_4","Group_4","Group_4"),
D8 = c(40.3,37.4,44.8,39.6,40.6,41.1,38.5,41.6,41.8,43.6,43.1,41.6),
D9 = c(41.1,36.9,44.1,39.6,41.2,42.1,39.2,41.6,41.8,42.0,42.7,40.8),
D10 = c(41.5,37.1,44.4,39.4,40.8,41.2,39.9,41.5,42.4,42.4,42.7,41.9))
# A tibble: 12 x 4
TTMENT D8 D9 D10
<chr> <dbl> <dbl> <dbl>
1 Group_1 40.3 41.1 41.5
2 Group_1 37.4 36.9 37.1
3 Group_1 44.8 44.1 44.4
4 Group_2 39.6 39.6 39.4
5 Group_2 40.6 41.2 40.8
6 Group_2 41.1 42.1 41.2
7 Group_3 38.5 39.2 39.9
8 Group_3 41.6 41.6 41.5
9 Group_3 41.8 41.8 42.4
10 Group_4 43.6 42 42.4
11 Group_4 43.1 42.7 42.7
12 Group_4 41.6 40.8 41.9
Group 1 is control data against which I need to compare (via t-test) data from groups 2, 3 and 4. I'm trying to use dplyr (groupby / summarize) to show mean and t-test p-value, and the mean should have an "*" if p<0.05. The intended output is as follows:
D8 D9 D10
Group_1 mean XX.X XX.X XX.X
Group_2 mean XX.X XX.X *XX.X
Group_2 p 0.XX 0.XX 0.03
Group_3 mean XX.X XX.X XX.X
Group_3 p 0.XX 0.XX 0.XX
Group_4 mean XX.X *XX.X XX.X
Group_4 p 0.XX 0.01 0.XX
I'm trying to achieve this with the following code:
# Save control data in a separate data frame
control <- df%>%
filter(TTMENT == "Group_1") %>%
select(-TTMENT)
# function to do the actual job
compare_means <- function(x) {
p <- t.test(control,x)$p.value # this bit is not doing what is intended
data_mean <- sprintf("%.1f", mean(x, na.rm=TRUE))
if (p < 0.05) {
significant = "*" } else { significant = "" }
data_mean <- paste(significant, data_mean, sep="")
# Next line also triggers error. It is not possible to return two lines to summarize?
result <- data.frame(c(data_mean, p), row.names("mean", "p"), stringAsFactors=FALSE)
return(result)
}
# dplyr part
stack %>%
group_by(TTMENT) %>%
summarize(across(1:3, ~ compare_means(.) ))
I don't know how to handle the t-test comparisons (each not Group_1 against Group_1 [control])
Furthermore, is it possible to achieve what I intend to in one go? or should I do a dplyr groupby/summrize step for each row of results I need, i.e. one for p values and another one for means, and possibly another one for any other statistic I might need?
CodePudding user response：
I suggest the following approach.
It doesn‘t give you the very exact structure of your desired output, but a clean / tidy version of all the t test results. From there it should be easy to e.g. add asterisks for significant result.
Not sure what you want to do with your output, but IMO I would skip the last lines of code where I pivot_longer and pivot_wider and just use the tidy output as is because that‘s the, well, most tidy output you can get.
library(tidyverse)
library(broom)
df_control <- df %>%
filter(TTMENT == 'Group_1') %>%
pivot_longer(cols = -TTMENT)
df_control
df %>%
filter(TTMENT != 'Group_1') %>%
pivot_longer(cols = -TTMENT) %>%
mutate(name_filter = name) %>%
group_by(TTMENT, name) %>%
group_modify(.f = ~tidy(t.test(df_control$value[df_control$name == .x$name_filter], .x$value, alternative = 'two.sided'))) %>%
arrange(TTMENT, match(name, str_sort(name, numeric = TRUE))) %>%
select(TTMENT, name, group_1_mean = estimate1, group_estimate = estimate2, p.value) %>%
pivot_longer(-c(TTMENT, name), names_to = 'statistic') %>%
pivot_wider()
which gives:
# A tibble: 9 x 5
# Groups: TTMENT [3]
TTMENT statistic D8 D9 D10
<chr> <chr> <dbl> <dbl> <dbl>
1 Group_2 group_1_mean 40.8 40.7 41
2 Group_2 group_estimate 40.4 41.0 40.5
3 Group_2 p.value 0.871 0.913 0.828
4 Group_3 group_1_mean 40.8 40.7 41
5 Group_3 group_estimate 40.6 40.9 41.3
6 Group_3 p.value 0.939 0.946 0.914
7 Group_4 group_1_mean 40.8 40.7 41
8 Group_4 group_estimate 42.8 41.8 42.3
9 Group_4 p.value 0.467 0.646 0.595
CodePudding user response：
Thanks to @deschen for his answer.
Just posting the complete working code (based on @deschen suggestion) to obtain the intended summary output. I'm certain there's much leaner solutions, but I think this (newbie) code is easy to understand and probably helpful to others.
(df_control <- df %>%
filter(TTMENT == 'Group_1') %>%
pivot_longer(cols = -TTMENT)
)
get_summary_statistics <- function(x) {
control <- df_control$value[df_control$name == x$name_filter]
# t.test and assigns p-value
p_value <- t.test(control, x$value)$p.value
# Compose mean /- SD (with asterisk if significant)
significant <- case_when(p_value < 0.01 ~"**",
p_value < 0.05 ~"*",
p_value >= 0.05 ~"")
mean <- mean(x$value, na.rm=TRUE)
sd <- sd(x$value, na.rm=TRUE)
a <- paste (significant, sprintf("%.1f", mean), sep="")
b <- "\u00B1"
c <- sprintf("%.1f", sd(x$value, na.rm=TRUE))
mean_and_sd <- paste(a, b, c, sep=" ")
# SD over mean as %
sd_over_mean = sprintf("%.1f", sd/mean*100)
# N (excluding NAs)
n = sprintf("%d", sum(!is.na(x$value)) )
# Returns a tibble with the required statistics
tibble(p = sprintf("%.3f", p_value),
"Mean \u00B1 SD" = mean_and_sd,
"SD (% of Mean)" = sd_over_mean,
"n" = n)
}
df %>%
pivot_longer(cols = -TTMENT) %>% # columns "name" and "value" are defaults from pivot_longer
mutate(name_filter = name) %>% # name_filter is needed to get the necessary values from control data
group_by(TTMENT, name) %>%
group_modify(.f = ~get_summary_statistics(.x)) %>%
pivot_longer(-c(TTMENT, name), names_to = 'Statistic') %>%
pivot_wider() %>%
arrange(TTMENT, match(Statistic, c("n", "Mean ± SD","SD (% of Mean)","p")))
Output:
# A tibble: 16 x 5
# Groups: TTMENT [4]
TTMENT Statistic D10 D8 D9
<chr> <chr> <chr> <chr> <chr>
1 Group_1 n 3 3 3
2 Group_1 Mean ± SD 41.0 ± 3.7 40.8 ± 3.7 40.7 ± 3.6
3 Group_1 SD (% of Mean) 9.0 9.1 8.9
4 Group_1 p 1.000 1.000 1.000
5 Group_2 n 3 3 3
6 Group_2 Mean ± SD 40.5 ± 0.9 40.4 ± 0.8 41.0 ± 1.3
7 Group_2 SD (% of Mean) 2.3 1.9 3.1
8 Group_2 p 0.828 0.871 0.913
9 Group_3 n 3 3 3
10 Group_3 Mean ± SD 41.3 ± 1.3 40.6 ± 1.9 40.9 ± 1.4
etc.