Saving Residuals of a regression to the original dataframe

Question

I can't comment directly on the page in question but essentially I'm trying to get the code from Combining dplyr::do() with dplyr::mutate? to work.

dat <- mtcars

dat %>% 
    group_by(gear) %>% 
    mutate(res = residuals(lm(deparse(substitute(mpg ~ disp)))))

Running the above code, I get:

Error in eval(substitute(expr), envir, enclos) : object 'mpg' not found

Am I missing something?

Answer 1

There are a lot of options here, including modelr::add_residuals (see @LmW's answer), broom::augment, and plain old residuals. If you're working with grouped models, nesting models in a list column is handy, and naturally leads to iterating over the list of models to calculate residuals and such.

---

residuals

Plain old base R works neatly with some purrr (use lapply if you prefer):

library(tidyverse)

mtcars %>% 
    rownames_to_column('car') %>% 
    nest(-gear) %>% 
    mutate(model = map(data, ~lm(mpg ~ disp, data = .x)),
           resid = map(model, residuals)) %>%
    unnest(data, resid)

#> # A tibble: 32 × 13
#>     gear       resid            car   mpg   cyl  disp    hp  drat    wt
#>    <dbl>       <dbl>          <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1      4  0.98649891      Mazda RX4  21.0     6 160.0   110  3.90 2.620
#> 2      4  0.98649891  Mazda RX4 Wag  21.0     6 160.0   110  3.90 2.875
#> 3      4 -3.56856040     Datsun 710  22.8     4 108.0    93  3.85 2.320
#> 4      4  2.76107028      Merc 240D  24.4     4 146.7    62  3.69 3.190
#> 5      4  0.44001547       Merc 230  22.8     4 140.8    95  3.92 3.150
#> 6      4  0.11531527       Merc 280  19.2     6 167.6   123  3.92 3.440
#> 7      4 -1.28468473      Merc 280C  17.8     6 167.6   123  3.92 3.440
#> 8      4  2.45060811       Fiat 128  32.4     4  78.7    66  4.08 2.200
#> 9      4  0.08397007    Honda Civic  30.4     4  75.7    52  4.93 1.615
#> 10     4  3.02179175 Toyota Corolla  33.9     4  71.1    65  4.22 1.835
#> # ... with 22 more rows, and 4 more variables: qsec <dbl>, vs <dbl>,
#> #   am <dbl>, carb <dbl>

You could wrap the lm call directly in residuals:

mtcars %>% 
    rownames_to_column('car') %>% 
    group_by(gear) %>% 
    mutate(resid = residuals(lm(mpg ~ disp)))

which obtains the same results, but such an approach is inadvisable unless you're sure you don't plan on doing anything else with the models. (Obviously unnesting loses the models as well, but you can control if and when you do so and whether you save a copy by breaking the chain earlier.)

---

broom::augment

augment adds lots of useful variables, including residuals, and can be used similarly:

mtcars %>% 
    rownames_to_column('car') %>%
    nest(-gear) %>% 
    mutate(model = map(data, ~lm(mpg ~ disp, data = .x)), 
           model_data = map(model, broom::augment)) %>% 
    unnest(model_data)

#> # A tibble: 32 × 10
#>     gear   mpg  disp  .fitted   .se.fit      .resid       .hat   .sigma
#>    <dbl> <dbl> <dbl>    <dbl>     <dbl>       <dbl>      <dbl>    <dbl>
#> 1      4  21.0 160.0 20.01350 0.9758770  0.98649891 0.16546553 2.503083
#> 2      4  21.0 160.0 20.01350 0.9758770  0.98649891 0.16546553 2.503083
#> 3      4  22.8 108.0 26.36856 0.7466989 -3.56856040 0.09687426 2.197330
#> 4      4  24.4 146.7 21.63893 0.8206560  2.76107028 0.11701449 2.331455
#> 5      4  22.8 140.8 22.35998 0.7674126  0.44001547 0.10232345 2.524090
#> 6      4  19.2 167.6 19.08468 1.0800836  0.11531527 0.20268993 2.528466
#> 7      4  17.8 167.6 19.08468 1.0800836 -1.28468473 0.20268993 2.482941
#> 8      4  32.4  78.7 29.94939 1.0762841  2.45060811 0.20126638 2.357875
#> 9      4  30.4  75.7 30.31603 1.1195513  0.08397007 0.21777368 2.528634
#> 10     4  33.9  71.1 30.87821 1.1879209  3.02179175 0.24518417 2.247410
#> # ... with 22 more rows, and 2 more variables: .cooksd <dbl>,
#> #   .std.resid <dbl>

If you'd like to save the unused variables from the original data, change model_data to model_data = map2(model, data, broom::augment)), passing augment a data parameter instead of letting it default to the data used by the model.