CODEWITHSUNDEEP
rsubsetlevels
barerd
barerd

Reputation: 845

How can I compare one level of a factor with all remaining levels in R

I have a dataframe similar to built-in InsectSprays (with factor and numeric data), but it contains 10+ numeric and 20+ factor vectors with few NAs. When I boxplot(numeric ~ factor), I notice that some levels stand out, and I want to be able to compare them with the rest.

As an example: InsectSprays contains a numeric vector called count (0:26), and a factor vector called sprays with levels: A, B, C, D, E and F. In InsectSprays, C is lowest, so I want to cbe able to compare C with all others.

I wrote a function for such numeric vectors:

num_interlevel <- function(df, variable, category){
  #find the levels of the categorizing parameter
  level.list <- levels(category)
  #build enough columns in the plot area
  par(mfrow=c(1,length(level.list)))
  for(i in 1:length(level.list)){
    #subset the df containing only the level in question
    variable.df <- na.omit(df[which(category == level.list[i]),])
    #subset the df containing all other levels
    category.df <- na.omit(df[which(category != level.list[i]),])
    boxplot(variable.df[, variable], category.df[, variable])
    p <- t.test(variable.df[, variable], category.df[, variable])$p.value
    title(paste(level.list[i], "=", p))
  }
}

and num_interlevel(InsectSprays, "count", InsectSprays$spray) gives me the result I want.

But when it comes to comparing factor vectors with each other (and I used tables for that), it doesn't work, simply because the dataframes are of different size, and more importantly, because this is a wrong way.

Then I thought that there may be an existing function for that, but couldn't find any. Can anyone suggest a way/function to create one subset containing exactly one level and another subset containing all the other levels?

#dput:
structure(list(Yas = c(27, 18, 17, 18, 18), Cinsiyet = structure(c(2L, 
2L, 2L, 1L, 1L), .Label = c("Erkek", "Kadın"), class = "factor"), 
Ikamet = structure(c(5L, 4L, 3L, 3L, 5L), .Label = c("Aileyle", 
"Akrabayla", "Arkadaşla", "Devlet yurdu", "Diğer", "Özel yurt", 
"Tek başına"), class = "factor"), Aile_birey = c(13, 9, 6, 
10, 6), Aile_gelir = c(700, 1000, 1500, 600, 800)), .Names = c("Yas", 
"Cinsiyet", "Ikamet", "Aile_birey", "Aile_gelir"), row.names = c(NA, 
5L), class = "data.frame")

Edit

I reformed my functions after James's answer. This is certainly not an elegant solution, but I put it here for future reference:

n.compare <- function(df, variable, category){
  level.list <- levels(df[,category])
  par(mfrow=c(1,length(level.list)))
  for(i in 1:length(level.list)){
    boxplot(df[,variable] ~ (df[,category] == level.list[i]))
    p <- t.test(df[,variable] ~ (df[,category] == level.list[i]))$p.value
    title(paste(level.list[i], "=", p))
  }
}

f.compare <- function(df, variable, category){
  level.list <- levels(df[,category])
  par(mfrow=c(1,length(level.list)))
  for(i in 1:length(level.list)){
    print(paste(level.list[i]))
    print(table((df[,category] == level.list[i]), df[,variable]))
    writeLines("\n")
  }
}

Upvotes: 0

Views: 1308

Answers (1)

James
James

Reputation: 66834

To split up a data.frame, use split:

lapply(split(InsectSprays,InsectSprays$spray=="A"),summary)
$`FALSE`
     count       spray 
 Min.   : 0.00   A: 0  
 1st Qu.: 3.00   B:12  
 Median : 5.00   C:12  
 Mean   : 8.50   D:12  
 3rd Qu.:13.25   E:12  
 Max.   :26.00   F:12  

$`TRUE`
     count       spray 
 Min.   : 7.00   A:12  
 1st Qu.:11.50   B: 0  
 Median :14.00   C: 0  
 Mean   :14.50   D: 0  
 3rd Qu.:17.75   E: 0  
 Max.   :23.00   F: 0

Also consider the following:

boxplot(count~(spray=="A"),InsectSprays)

Upvotes: 2

Related Questions