CODEWITHSUNDEEP
rfunctionloopslapply
rnorouzian
rnorouzian

Reputation: 7517

Looping all pairwise comparisons from a list in R

In my R code below, suppose I want to compare all unique 2 m objects using a similar R routine. For example, to compare m1 and m2, my routine is:

pchisq(2 * (logLik(m2) - logLik(m1)), df = abs(m1$df.residual - m2$df.residual), lower = F)

Question:

I was wondering how I could make a function to make all unique pairwise comparisons for all m objects using my routine?

Here is what I've tried without success:

## Suppose we have 4 `m` objects: `m1...m4` (6 unique pairwise comparisons possible)

m1 <- lm(hp ~ vs, mtcars)
m2 <- lm(hp ~ vs*wt, mtcars)
m3 <- lm(hp ~., mtcars)
m4 <- lm(hp ~ vs * gear * wt, mtcars)


 compare <- function(...){

   m <- list(...)
   L <- length(m) - 1

   lapply(1:L, function(i) pchisq(2 * (logLik(m[[i+1]]) - logLik(m[[i]])), df = abs(m[[i]]$df.residual - m[[i+1]]$df.residual), lower.tail = FALSE) )

  }
# Example of use:
compare(m1, m2, m3, m4)

Upvotes: 1

Views: 2434

Answers (1)

Douglas Mesquita
Douglas Mesquita

Reputation: 1021

You can use combn to get all combinations you want to compare:

compare <- function(...){

  m <- list(...)
  n_mod <- length(m)
  names(m) <- sapply(substitute(list(...))[-1], deparse)
  combs <- t(combn(x = names(m), m = 2))

  comp_value <- apply(X = combs, MARGIN = 1, function(ind) pchisq(2 * (logLik(m[[ind[2]]]) - logLik(m[[ind[1]]])), df = abs(m[[ind[1]]]$df.residual - m[[ind[2]]]$df.residual), lower.tail = FALSE))
  df_out <- data.frame(combs, comp_value)
  names(df_out) <- c("mod_1", "mod_2", "comp_value")

  return(df_out)
}

So, to make it easier to read the result, you can return a data.frame with all the comparisons.

Then 

compare(m1, m2, m3, m4)
  mod_1 mod_2   comp_value
1    m1    m2 2.391012e-02
2    m1    m3 7.253068e-08
3    m1    m4 1.248692e-06
4    m2    m3 2.735901e-07
5    m2    m4 4.256098e-06
6    m3    m4 1.000000e+00

Upvotes: 3

Related Questions