CODEWITHSUNDEEP
rdplyrtidyversepurrr
zesla
zesla

Reputation: 11813

how to run models on combinations of two variables and return a dataframe with p-values and r-square for each model using tidyverse

I'm trying to run models on different combinations of variables. I want to have a dataframe with 3 column: variables, p-value and r-square for each model. I'm using mtcars dataset as an example. Here are my codes:

c('wt', 'hp', 'qsec') %>% 
    combn(2, paste, collapse='*') %>% 
    structure(., names=.) %>% 
    map(~formula(paste('mpg~', .x))) %>% 
    map(lm, data=mtcars) %>%
    map(~coef(summary(.x))[4,4]) %>% 
    unlist() %>% 
    data.frame(p.value=.) %>% 
    rownames_to_column(var='interaction')

Here is my output:

interaction      p.value
1       wt*hp 0.0008108307
2     wt*qsec 0.2652596233
3     hp*qsec 0.0001411028

The question is how I can add another column to the dataframe after extracting r-square from each model? I want to achieve that in the chaining operation above. Since I hope to generalize the approach to other type of models, so I want to do that without using broom package. Appreciate it if anyone can help me with that. Thanks a lot.

Upvotes: 1

Views: 354

Answers (2)

mt1022
mt1022

Reputation: 17299

I would try following

library(tidyverse)

reg.vars <- c('wt', 'hp', 'qsec')

tibble(interaction = combn(reg.vars, 2, paste, collapse = '*')) %>%
    mutate(fit = map(interaction, ~ summary(lm(paste('mpg ~', .), data = mtcars))),
           pval = map_dbl(fit, ~ coef(.)[4, 4]),
           rsq = map_dbl(fit, ~ .$r.squared)) %>%
    select(-fit)
# # A tibble: 3 x 3
#   interaction         pval       rsq
#         <chr>        <dbl>     <dbl>
# 1       wt*hp 0.0008108307 0.8847637
# 2     wt*qsec 0.2652596233 0.8340742
# 3     hp*qsec 0.0001411028 0.7854734

Upvotes: 5

akrun
akrun

Reputation: 887691

We can use broom package functions like glance, tidy

library(broom)
library(tidyverse)
v1 %>% 
     combn(2, paste, collapse='*') %>% 
     structure(., names=.) %>% 
      map(~summary(lm(formula(paste('mpg~', .x)), data = mtcars))) %>% 
      map(~ data.frame( tidy(.)[4,]['p.value'], glance(.)['r.squared'])) %>%    
      bind_rows(., .id = 'interaction') 
#  interaction      p.value r.squared
#1       wt*hp 0.0008108307 0.8847637
#2     wt*qsec 0.2652596233 0.8340742
#3     hp*qsec 0.0001411028 0.7854734

data

v1 <- c('wt', 'hp', 'qsec')

Upvotes: 2

Related Questions