map a vector of characters to lm formula in r

Question

I'm trying to make a list of lm object using purrr::map. use mtcars as an example:

vars <- c('hp', 'wt', 'disp')
map(vars, ~lm(mpg~.x, data=mtcars))

error: Error in model.frame.default(formula = mpg ~ .x, data = mtcars, drop.unused.levels = TRUE) : variable lengths differ (found for '.x')

I also tried:

map(vars, function(x) {x=sym(x); lm(mpg~!!x, data=mtcars)})

I got error message:

Error in !x : invalid argument type

Can anyone tell what I did wrong? Thanks in advance.

Answer 1

The usual way is to paste together formulas as strings, convert them by mapping as.formula (you can't make a vector of formulas; it has to be a list), and then map lm. You can combine it all to a single call if you like, but I've come to prefer mapping single functions, which makes code easier to read:

library(purrr)

c('hp', 'wt', 'disp') %>% 
    paste('mpg ~', .) %>% 
    map(as.formula) %>% 
    map(lm, data = mtcars)
#> [[1]]
#> 
#> Call:
#> .f(formula = .x[[i]], data = ..1)
#> 
#> Coefficients:
#> (Intercept)           hp  
#>    30.09886     -0.06823  
#> 
#> 
#> [[2]]
#> 
#> Call:
#> .f(formula = .x[[i]], data = ..1)
#> 
#> Coefficients:
#> (Intercept)           wt  
#>      37.285       -5.344  
#> 
#> 
#> [[3]]
#> 
#> Call:
#> .f(formula = .x[[i]], data = ..1)
#> 
#> Coefficients:
#> (Intercept)         disp  
#>    29.59985     -0.04122

It's actually unnecessary to call map(as.formula) as lm will coerce it into a formula, but not all models are so generous (e.g. mgcv::gam).

A downside of this approach are that the call listed in the object looks funky, but the coefficients tell you which is which easily enough anyway. A useful alternative is to keep the formula as a string in one column of a data.frame and the model in a list column, e.g.

library(tidyverse)

data_frame(formula = paste('mpg ~', c('hp', 'wt', 'disp')), 
           model = map(formula, lm, data = mtcars))
#> # A tibble: 3 x 2
#>      formula    model
#>        <chr>   <list>
#> 1   mpg ~ hp <S3: lm>
#> 2   mpg ~ wt <S3: lm>
#> 3 mpg ~ disp <S3: lm>

Answer 2

The elegant tidyverse approach demonstrated by @alistaire worked well for me until I tried to pass the list column to the stargazer package and received "% Error: Unrecognized object type."

In case it is helpful for anyone else trying to use purrr map and stargazer, this slight modification solved the issue:

models_out <- data_frame(
    formula = paste('mpg ~', c('hp', 'wt', 'disp')), 
    model = map(
                .x = formula, 
                .f = function(x) lm(x, data = mtcars))
                )

stargazer(models_out$model, type = 'text')

===========================================================
                                   Dependent variable:     
                              -----------------------------
                                           mpg             
                                 (1)       (2)       (3)   
-----------------------------------------------------------
hp                            -0.068***                    
                               (0.010)                     
                                                           
wt                                      -5.344***          
                                         (0.559)           
                                                           
disp                                              -0.041***
                                                   (0.005) 
                                                           
Constant                      30.099*** 37.285*** 29.600***
                               (1.634)   (1.878)   (1.230) 
                                                           
-----------------------------------------------------------
Observations                     32        32        32    
R2                              0.602     0.753     0.718  
Adjusted R2                     0.589     0.745     0.709  
Residual Std. Error (df = 30)   3.863     3.046     3.251  
F Statistic (df = 1; 30)      45.460*** 91.375*** 76.513***
===========================================================
Note:                           *p<0.1; **p<0.05; ***p<0.01