Automate statistical analyses using apply/apply-like function

user_13675032 0
Viewed 39

I have a dataframe that contains 5 columns, each corresponding to a survey item, and a grouping variable. There are a total of 300 observations in my dataframe and each cell entry represents the response given by a student on a given item. I constructed the following reproducible dataframe:

set.seed(14)
Group <- rep(c(0, 1), each = 150)
mydf <- data.frame(replicate(5, sample(0:1,300,rep=TRUE)))
mydf$Group <- Group
mydf$Group <- factor(mydf$Group, levels = c(0, 1), labels = c("Group A", "Group B"))
head(mydf); tail(mydf)
> head(mydf); tail(mydf)
  X1 X2 X3 X4 X5   Group
1  0  1  1  1  0 Group A
2  1  1  1  1  1 Group A
3  1  1  0  1  0 Group A
4  1  0  0  1  1 Group A
5  1  1  0  1  0 Group A
6  1  0  1  1  1 Group A
    X1 X2 X3 X4 X5   Group
295  0  1  1  0  1 Group B
296  0  0  1  0  0 Group B
297  1  1  0  1  0 Group B
298  1  1  0  0  1 Group B
299  0  0  1  0  0 Group B
300  1  1  1  1  1 Group B

What I would like to do is perform chi-square test of independence on each survey item X1 to X5. So far I have been doing the following [for item 1 (X1)]:

mydf$X1 <- factor(mydf$X1, levels = c(0, 1), labels = c("AGREE", "DISAGREE"))
MyTable <- table(mydf$Group, mydf$X1)
addmargins(MyTable)
chisq.test(MyTable, correct = FALSE)

and I would like to use the lapply function (or something similar) to automate this process so that I do not have to repeat the preceding code for each of the 5 items. This is particualry important because I have a similar dataframe that contains 50 items and I would like to use the same code to automate those analyses. Any advice on how to proceed? I am having the most trouble with calling out each variable (i.e. X1, X2, etc.) and I am not sure how to do so. I recently started using R so I do not have a firm understanding of these functions and commands. Any help is greatly appreciated.

lapply apply statistics r
2 Answers

We can use lapply to loop over the column 'X1' to 'X5', then create a table with 'Group' column and o the chisq.test to return a list of test results

out <- lapply(mydf[paste0("X", 1:5)], function(x) 
    chisq.test(table(mydf$Group,
            factor(x, levels = 0:1, labels = c("AGREE", "DISAGREE"))), 
           correct = FALSE))

sapply(out, `[[`, "p.value")
#       X1         X2         X3         X4         X5 
#0.72875061 0.72888976 0.90732945 0.01525704 0.08243538
user_37322710

Here's a little fancier a solution that helps keep things organized...

library(dplyr)
library(purrr)
library(infer)


# make a simple vector of the variables you want on one side
exexs <- c("X1", "X2", "X3")
# an easy way to keep track of names
names(exexs) <- exexs

# make our own simple function
MyFunction <- function(data, groupvar, var) 
{ 
  infer::chisq_test(formula = as.formula(paste(var, "~", groupvar)), 
                    x = data, 
                    correct = FALSE) # you can pass options to chisq.test
}

# purrr::map(exexs, ~ MyFunction(data = mydf, groupvar = .x, var = "Group"))
purrr::map_dfr(exexs, ~ MyFunction(data = mydf, groupvar = .x, var = "Group"), .id = "Variable") %>% rbind()
#> # A tibble: 3 x 4
#>   Variable statistic chisq_df p_value
#>   <chr>        <dbl>    <int>   <dbl>
#> 1 X1          0.120         1   0.729
#> 2 X2          0.120         1   0.729
#> 3 X3          0.0136        1   0.907

Based on your data

set.seed(14)
Group <- rep(c(0, 1), each = 150)
mydf <- data.frame(replicate(5, sample(0:1,300,rep=TRUE)))
mydf$Group <- Group
mydf$Group <- factor(mydf$Group, levels = c(0, 1), labels = c("Group A", "Group B"))
user_91900340
Related