I have a dataframe that is set up as follows:

set.seed(1234)
id <- rep(1:50, each = 3)
stimuli <- rep(c("a", "b", "c"), each = 1, times = 50)
dv_1 <- rnorm(150, mean = 2, sd = 0.7)
dv_2 <- rnorm(150, mean = 4, sd = 1.5)
dv_3 <- rnorm(150, mean = 7.5, sd = 1)

simdat <- data.frame(id, stimuli, dv_1, dv_2, dv_3)

simranks <- t(apply(simdat[,-1], 1, function(x) rank(x, ties.method = "min") ))
colnames(simranks) <- paste(colnames(simranks), "_rank", sep="")

simdat <- data.frame(simdat, simranks)

I then have split the dataframe according to the different types of stimuli, e.g.,

dat_a <- simdat %>%
  dplyr::filter(stimuli == "a") %>%
  select(id, dv_1_rank, dv_2_rank, dv_3_rank)

Then I would like to perform a bunch of different analyses on the subsetted data:

a_melt <- melt(dat_a, id.vars = c("id"), measure.vars = c("dv_1_rank", "dv_2_rank", "dv_3_rank"))

a_perc <- a_melt %>% 
  group_by(variable, value) %>% 
  summarise(count = n()) %>% 
  mutate(perc = count/sum(count))

ggplot(a_perc, aes(x = variable, y = perc, fill = value))   
  geom_col(position = "stack")  
  scale_y_continuous(labels=scales::percent) 

How can I write the code so that, rather than copying and pasting the code chunks for stimuli b and stimuli c, it loops over all of them (the "stimuli" column in the original dataset)

CodePudding user response：

Here is how I usually deal with this.

1. I split my dataset into chunks gathered in a list then,
2. I use lapply ou purrr::map function to apply a function, that does the analysis for one chunk, to each chunck.

Here you could do something like:

analyses <- function(.df){
  require(dplyr)
  require(ggplot2)
  df_melt <- data.table::melt(.df, id.vars = c("id"), measure.vars = c("dv_1_rank", "dv_2_rank", "dv_3_rank"))
  df_perc <- df_melt %>% 
    group_by(variable, value) %>% 
    summarise(count = n()) %>% 
    mutate(perc = count/sum(count))
  ggplot(df_perc, aes(x = variable, y = perc, fill = value))   
    geom_col(position = "stack")  
    scale_y_continuous(labels=scales::percent) 
}
 
lapply(split(simdat, ~ stimuli), analyses)

Note: this is just a proof of concept.

CodePudding user response：

Does this produce your desired output?

library(tidyverse)
simdat <- expand_grid(stimuli = c("a", "b", "c"), id = 1:20) %>%
  mutate(
    dv_1_rank = floor(runif(nrow(.), 1, 5)),
    dv_2_rank = floor(runif(nrow(.), 1, 5)),
    dv_3_rank = floor(runif(nrow(.), 1, 5)),
  )
a_perc <- simdat %>% 
  pivot_longer(dv_1_rank:dv_3_rank) %>%
  group_by(stimuli, name, value) %>% 
  summarise(count = n(), .groups = "drop") %>% 
  group_by(stimuli, name) %>% 
  mutate(perc = count/sum(count)) %>%
  ungroup() 
print(a_perc)
#> # A tibble: 36 x 5
#>    stimuli name      value count  perc
#>    <chr>   <chr>     <dbl> <int> <dbl>
#>  1 a       dv_1_rank     1     4  0.2 
#>  2 a       dv_1_rank     2     8  0.4 
#>  3 a       dv_1_rank     3     5  0.25
#>  4 a       dv_1_rank     4     3  0.15
#>  5 a       dv_2_rank     1     5  0.25
#>  6 a       dv_2_rank     2     6  0.3 
#>  7 a       dv_2_rank     3     4  0.2 
#>  8 a       dv_2_rank     4     5  0.25
#>  9 a       dv_3_rank     1     3  0.15
#> 10 a       dv_3_rank     2     5  0.25
#> # ... with 26 more rows

^{Created on 2022-03-14 by the reprex package (v2.0.1)}