CODEWITHSUNDEEP
rindexinglinear-regression
Abed
Abed

Reputation: 325

How can I access the name of an lm() object in a list of other objects?

I have some unruly data which I have tried to model with a number of varying linear models, which I have given various names. I then created a data frame called "compared fits". I am able to store some of the more relevant output from the various models using the code below. 

for(i in 1:length(regressionmodels.listed))
{
  comparedfits[i, 2] = summary(regressionmodels.listed[[i]])$df[2]
  comparedfits[i, 3] = summary(regressionmodels.listed[[i]])$r.squared 
  comparedfits[i, 4] = summary(modelnow)$adj.r.squared 
#See code snippet below for explanation  
#name = deparse(get(substitute(regressionmodels.listed[[i]])))
#names_models[i] = name
}

The "regression models listed" is an object of type list which I create using simply: 

regressionmodels.listed <- list(fit1, fit2 ...etc)

What I cannot do in any kind of iterative way is the following: 

for(i in 1:length(regressionmodels.listed))
{
namesofmodels[i] = deparse(substitute(regressionmodels.listed[[i]]))
}

Note that if I run the deparse code on an individual fitted model, say fit1, it does work. I only cannot get it to work with the elements of a list of fitted models. This is obviously a "nice to have" kind of thing, but since I will be comparing maybe 10 or more models, it would really make my life easier if I could find a way to do this and add it to my other data frame above.

EDIT This is what the regressionmodels.listed looks like:

regressionmodels.listed <- list(lmobject1, lmobject2, lmobject3)

The lm objects are created simply by running a linear model on various combinations of the data.

Upvotes: 1

Views: 1384

Answers (2)

StupidWolf
StupidWolf

Reputation: 46888

Your list of models are not named so there's no way of getting back "lmobject1" etc, one way is to name them at the start:

regressionmodels.listed <- list(lmobject1, lmobject2, lmobject3)
names(regressionmodels.listed) = c("lmobject1","lmobject2","lmobject3")
#or
names(regressionmodels.listed) = paste0("lmobject",1:length(regressionmodels.listed))

namesofmodels = names(regressionmodels.listed)

One other option is to use broom:

library(tibble)
library(broom)
library(tidyr)
library(dplyr)
library(purrr)

lm1 = lm(mpg ~ cyl,data=mtcars)
lm2 = lm(mpg ~ disp,data=mtcars)
lm3 = lm(mpg ~ hp,data=mtcars)

res = tibble(models=list(lm1,lm2,lm3))
res = res %>% mutate(names=paste0("model",1:n()),stats=map(models,glance))

# A tibble: 3 x 3
  models names  stats            
  <list> <chr>  <list>           
1 <lm>   model1 <tibble [1 × 11]>
2 <lm>   model2 <tibble [1 × 11]>
3 <lm>   model3 <tibble [1 × 11]>

we can unnest this:

res %>% unnest(cols=stats)

# A tibble: 3 x 13
  models names r.squared adj.r.squared sigma statistic  p.value    df logLik
  <list> <chr>     <dbl>         <dbl> <dbl>     <dbl>    <dbl> <int>  <dbl>
1 <lm>   mode…     0.726         0.717  3.21      79.6 6.11e-10     2  -81.7
2 <lm>   mode…     0.718         0.709  3.25      76.5 9.38e-10     2  -82.1
3 <lm>   mode…     0.602         0.589  3.86      45.5 1.79e- 7     2  -87.6
# … with 4 more variables: AIC <dbl>, BIC <dbl>, deviance <dbl>,
#   df.residual <int>

and display the columns you like:

res %>% unnest(cols=stats) %>% select(names,df,r.squared,adj.r.squared)
# A tibble: 3 x 4
  names     df r.squared adj.r.squared
  <chr>  <int>     <dbl>         <dbl>
1 model1     2     0.726         0.717
2 model2     2     0.718         0.709
3 model3     2     0.602         0.589

Upvotes: 2

Francesco Grossetti
Francesco Grossetti

Reputation: 1595

I came up with this fast solution. You want to store some summary statistics from a list of models (e.g. linear models from lm()) in a data.frame. I call this object result which will be of dimensions NxP where N is the number of models you stored in your list and P is the number of statistics you want to collect. In this case, P=3 for d.o.f., R-squared and Adjusted R-Squared.

Here I am using the built-in data mtcars.

data("mtcars")
mod1 = lm( mpg ~ hp, data = mtcars )
mod2 = lm( mpg ~ wt, data = mtcars )
mod3 = lm( mpg ~ qsec, data = mtcars )

regressionmodels.listed = list( mod1, mod2, mod3 )
N = length( regressionmodels.listed )

results = data.frame( df = rep( NA_integer_, N ),
                      r.squared = rep( NA_real_, N ),
                      adj.r.squared = rep( NA_real_, N ) )

for ( i_mod in seq_along( regressionmodels.listed ) ) {

  results[ i_mod, "df" ] = regressionmodels.listed[[ i_mod ]]$df
  results[ i_mod, "r.squared" ] = summary(regressionmodels.listed[[ i_mod ]])$r.squared
  results[ i_mod, "adj.r.squared" ] = summary(regressionmodels.listed[[ i_mod ]])$adj.r.squared

}

Could this be what you wanted? If so, then substitute the mod* with your models and you should be fine.

Upvotes: 1

Related Questions