Using step function in R part of the tidyverse workflow, in particular, using the nest and map functions

0
votes

I am trying to fit nested models over which I would like to apply the step function and get a subset of potential predictors for each nested element.

I am currently running into the following error specifically when running the step function:

Error: Problem with mutate() input step_null_to_full. x cannot coerce class ‘"lm"’ to a data.frame i Input step_null_to_full is map2(...). i The error occurred in group 1: ID = 1.

Here is a generated code example:

library(tidyverse)

data <- data.frame(ID = rep(1:10,30),
                   Year = rep(1981:2010,10),
                   x1 = rnorm(300,0,1),
                   x2 = rnorm(300,0,1),
                   x3 = rnorm(300,0,1),
                   x4 = rnorm(300,0,1),
                   x5 = rnorm(300,0,1),
                   x6 = rnorm(300,0,1),
                   Y =  rnorm(300,0,1))


model_pipe <- data %>% 
  group_by(ID) %>% 
  nest() %>% 
  mutate(mod_intercept = map(data,~ lm(Y ~ 1, data=.)),
         mod_full = map(data,~ lm(Y ~ ., data=.)),
         mod_full_int = map(data,~ lm(Y ~ .*., data=.)),
         step_null_to_full = map2(mod_intercept,mod_full, 
                                  ~ step(.x, scope = formula(.y), direction = 'forward',trace = 0)))
1
rtidyverselinear-regressionstep

1 Answers

1
votes

To make this work you need to pass data in the step function with the same name as you used to fit model. While fitting the model we used data = . so rename data to . and apply the function.

library(tidyverse)

data %>% 
  group_by(ID) %>% 
  nest() %>% 
  mutate(mod_intercept = map(data,~ lm(Y ~ 1, data=.)),
         mod_full = map(data,~ lm(Y ~ ., data=.)),
         mod_full_int = map(data,~ lm(Y ~ .*., data=.)),
         step_null_to_full = pmap(list(mod_intercept,mod_full, data), 
                                  function(x, y, z) {
               `.` <- z
               step(x, scope = formula(y), direction = 'forward',trace = 0)
          })) -> result
result

#     ID data              mod_intercept mod_full mod_full_int step_null_to_full
#   <int> <list>            <list>        <list>   <list>       <list>           
# 1     1 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
# 2     2 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
# 3     3 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
# 4     4 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
# 5     5 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
# 6     6 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
# 7     7 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
# 8     8 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
# 9     9 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>             
#10    10 <tibble [30 × 8]> <lm>          <lm>     <lm>         <lm>