R: Apply function to data frame including only specific rows - r

Question

I have tried to apply a function to a data.frame including only specific rows.
My aim is to have a fifth column which includes a function which varies according to the group and func. Say I would like to perform a t-test in the case that func=a and to calculate a mean difference in the case func=b. In other words, the first three rows in the fifth column should include the result of a t.test (t.test(n1[1:3],n2[1:3])$p.value) comparing the n1 and n2 in the group 1. How is this possible?
n1<-c(58,94,58,94,65,87,65,91,20,16)
n2<-c(37,34,88,23,86,37,80,34,24,67)
group<-c(1,1,1,2,2,2,2,3,3,3)
func<-c('a','a','a','b','b','b','b','a','a','a')
data<-data.frame(n1,n2,group,func)
data
n1 n2 group func
1 58 37 1 a
2 94 34 1 a
3 58 88 1 a
4 94 23 2 b
5 65 86 2 b
6 87 37 2 b
7 65 80 2 b
8 91 34 3 a
9 20 24 3 a
10 16 67 3 a
EDIT:
Manually I can do it like this. But is could I do it if I had +1000 rows with 100+ groups?
pvalue1<-t.test(c(58,94,58),c(37,34,88))$p.value
pvalue2<-chisq.test(c(94,65,87,65),c(23,86,37,80))$p.value
pvalue3<-t.test(c(91,20,16),c(34,24,67))$p.value
pvalue<-c(rep(pvalue1,3),rep(pvalue2,4),rep(pvalue3,3))
cbind(data,pvalue)
n1 n2 group func pvalue
1 58 37 1 a 0.4737073
2 94 34 1 a 0.4737073
3 58 88 1 a 0.4737073
4 94 23 2 b 0.2381033
5 65 86 2 b 0.2381033
6 87 37 2 b 0.2381033
7 65 80 2 b 0.2381033
8 91 34 3 a 0.9822272
9 20 24 3 a 0.9822272
10 16 67 3 a 0.9822272

Answer 1

You can do the calculations with dplyr like this:
library(dplyr)
my_df %>%
group_by(group) %>%
mutate(p_value = ifelse(func == 'a', t.test(n1, n2)$p.value, chisq.test(n1, n2)$p.value))
# A tibble: 10 x 5
# Groups: group [3]
# n1 n2 group func p_value
# <dbl> <dbl> <dbl> <fct> <dbl>
# 1 58. 37. 1. a 0.474
# 2 94. 34. 1. a 0.474
# 3 58. 88. 1. a 0.474
# 4 94. 23. 2. b 0.238
# 5 65. 86. 2. b 0.238
# 6 87. 37. 2. b 0.238
# 7 65. 80. 2. b 0.238
# 8 91. 34. 3. a 0.982
# 9 20. 24. 3. a 0.982
# 10 16. 67. 3. a 0.982

Answer 2

I've seen some cool stuff along these lines in Hadley's R4DS book. Check this out for an example and some discussion around my approach below.
The following goes some way to achieving what you'd like:
library(dplyr)
library(purrr)
library(tidyr)
test_function <- function(func, data) {
if (func == "a") {t.test(data$n1, data$n2)$p.value}
else if (func == "b") {chisq.test(data$n1, data$n2)$p.value}
}
df %>%
group_by(group, func) %>%
nest() %>%
mutate(p_value = map2_dbl(func, data, function(x, y) test_function(x, y)))
%>% unnest()

Answer 3

Consider base R's underused by() which can split dataframes by one or more factors and then pass subsets into a defined or anonymous function, returning a list of function's output.
Data (assuming functions are strings)
n1 <- c(58,94,58,94,65,87,65,91,20,16)
n2 <- c(37,34,88,23,86,37,80,34,24,67)
group <- c(1,1,1,2,2,2,2,3,3,3)
func < -c('t.test','t.test','t.test','chisq.test','chisq.test',
'chisq.test','chisq.test','t.test','t.test','t.test')
data <- data.frame(n1,n2,group,func)
By processing (using get() to retrieve actual function):
data_list <- by(data, data$group, function(sub){
func <- print(as.character(sub$func[[1]]))
f <- get(func)
sub$pvalue <- f(sub$n1, sub$n2)$p.value
return(sub)
})
final_df <- do.call(rbind, data_list)
final_df
# n1 n2 group func pvalue
# 1.1 58 37 1 t.test 0.4737073
# 1.2 94 34 1 t.test 0.4737073
# 1.3 58 88 1 t.test 0.4737073
# 2.4 94 23 2 chisq.test 0.2381033
# 2.5 65 86 2 chisq.test 0.2381033
# 2.6 87 37 2 chisq.test 0.2381033
# 2.7 65 80 2 chisq.test 0.2381033
# 3.8 91 34 3 t.test 0.9822272
# 3.9 20 24 3 t.test 0.9822272
# 3.10 16 67 3 t.test 0.9822272