Printing regression coefficients from multiple models to a shared data frame

Question

This is a little rudimentary, I know. Basically, I want to use the save data from the coef function to a shared data frame for models that all pull limited possible variable

Answer 1

The first step is to combine your coefficients into a data frame with one row per combination of model and term. Then you'll be able to spread it into a table with one row per model and one column per term.

My broom package has a useful function, tidy for turning a linear fit into a data frame of coefficients:

fit <- lm(mpg ~ wt + disp + qsec, mtcars)
library(broom)
tidy(fit)
#          term  estimate std.error statistic p.value
# 1 (Intercept) 19.777558    5.9383    3.3305 0.00244
# 2          wt -5.034410    1.2241   -4.1127 0.00031
# 3        disp -0.000128    0.0106   -0.0121 0.99042
# 4        qsec  0.926649    0.3421    2.7087 0.01139

(Note that unlike coef, this returns a data frame rather than a matrix, and incorporates the terms as a column rather than rownames). You can apply this function to each of your models and then recombine, for example with plyr's ldply. We generate an example using 20 of the same model as your "models":

models <- replicate(20, lm(mpg ~ wt + disp + qsec, mtcars), simplify = FALSE)
names(models) <- paste0("MODEL", 1:20)

Then our "tidy and recombine" code will be:

all_coefs <- plyr::ldply(models, tidy, .id = "model")
head(all_coefs)
#    model        term  estimate std.error statistic p.value
# 1 MODEL1 (Intercept) 19.777558    5.9383    3.3305 0.00244
# 2 MODEL1          wt -5.034410    1.2241   -4.1127 0.00031
# 3 MODEL1        disp -0.000128    0.0106   -0.0121 0.99042
# 4 MODEL1        qsec  0.926649    0.3421    2.7087 0.01139
# 5 MODEL2 (Intercept) 19.777558    5.9383    3.3305 0.00244
# 6 MODEL2          wt -5.034410    1.2241   -4.1127 0.00031

You then need to remove the std.error, statistic, and p.value columns and spread the estimate term out. This can be done with the dplyr and tidyr packages:

library(dplyr)
library(tidyr)
results <- all_coefs %>% select(-(std.error:p.value)) %>%
    spread(term, estimate)

This produces:

     model (Intercept)      disp  qsec    wt
1   MODEL1        19.8 -0.000128 0.927 -5.03
2   MODEL2        19.8 -0.000128 0.927 -5.03
3   MODEL3        19.8 -0.000128 0.927 -5.03
4   MODEL4        19.8 -0.000128 0.927 -5.03
5   MODEL5        19.8 -0.000128 0.927 -5.03

Which is your desired output. (This output is boring since all the models were the same, but presumably yours are different). If some models have coefficients others don't, the missing values will be filled in with NA.