Print model in multi column using modelsummary in R

Question

I like to print the output of modelsummary in a two-column format. When I do a regression interaction model with a factor variable with several levels, I get quite a few regression coefficients. The R console prints them side-by-side. My final output should be html for r markdown / bookdown.

interaction <- lm(mpg ~ factor(hp)*factor(am), data=mtcars)
interaction

When I do this in modelsummary

library(modelsummary)

modelsummary(title = 'Interaction Model.',
             list("MPG" = interaction), 
             gof_omit = 'R2|AIC|BIC|RMSE|Log.Lik.|F')

it's just too long.

Any alternative?

Answer 1

You can use broom::tidy and dplyr::select to get a 44 row table with 3 columns:

library(broom)
library(dplyr)
interaction <- lm(mpg ~ factor(hp)*factor(am), data=mtcars)
tidy(interaction) %>% select(term, estimate, std.error)

knitr::kable can format the table if you need, and you can filter the rows, etc. For example:

library(dplyr)
library(broom)
library(knitr)
interaction <- lm(mpg ~ factor(hp)*factor(am), data=mtcars)
tidy(interaction) %>% 
  select(term, estimate, std.error) %>% 
  filter(! is.na(estimate)) %>% 
  kable(digits=3)

results in

term	estimate	std.error
(Intercept)	29.65	2.296
factor(hp)62	-5.25	2.717
factor(hp)65	3.50	2.054
factor(hp)66	-0.55	1.779
factor(hp)91	-4.40	2.054
factor(hp)93	-7.60	2.054
factor(hp)95	-6.85	2.717
factor(hp)97	-8.15	2.717
factor(hp)105	-11.55	2.717
factor(hp)109	-9.00	2.054
factor(hp)110	-8.25	2.717
factor(hp)113	0.00	2.054
factor(hp)123	-11.15	2.516
factor(hp)150	-14.30	2.516
factor(hp)175	-10.70	2.054
factor(hp)180	-13.35	2.445
factor(hp)205	-19.25	2.717
factor(hp)215	-19.25	2.717
factor(hp)230	-14.95	2.717
factor(hp)245	-15.85	2.516
factor(hp)264	-14.60	2.054
factor(hp)335	-15.40	2.054
factor(am)1	0.75	1.779
factor(hp)110:factor(am)1	-1.15	2.516

Answer 2

You can use the shape argument. See the ?modelsummary documentation and the many examples on the website.

library(modelsummary)
interaction <- lm(mpg ~ factor(hp)*factor(am), data=mtcars)
modelsummary(interaction, shape = term ~ statistic)

	(1) / Est.	(1) / S.E.
(Intercept)	29.650	2.296
factor(hp)62	-5.250	2.717
factor(hp)65	3.500	2.054
factor(hp)66	-0.550	1.779
factor(hp)91	-4.400	2.054
factor(hp)93	-7.600	2.054
factor(hp)95	-6.850	2.717
factor(hp)97	-8.150	2.717
factor(hp)105	-11.550	2.717
factor(hp)109	-9.000	2.054
factor(hp)110	-8.250	2.717
factor(hp)113	0.000	2.054
factor(hp)123	-11.150	2.516
factor(hp)150	-14.300	2.516
factor(hp)175	-10.700	2.054
factor(hp)180	-13.350	2.445
factor(hp)205	-19.250	2.717
factor(hp)215	-19.250	2.717
factor(hp)230	-14.950	2.717
factor(hp)245	-15.850	2.516
factor(hp)264	-14.600	2.054
factor(hp)335	-15.400	2.054
factor(am)1	0.750	1.779
factor(hp)110 × factor(am)1	-1.150	2.516
:—————————-	———–:	———–:
Num.Obs.	32
R2	0.985
R2 Adj.	0.942
AIC	120.3
BIC	157.0
Log.Lik.	-35.168
RMSE	0.73

Answer 3

building on James_D's answer, here's a way to reshape the long table into a wide columns-of-columns format (using an eclectic mix of base R and tidyverse pieces):

library(dplyr)
library(tidyr)
library(broom)
library(knitr)

parts = 3 ## multi-column parts to display alongside

tidy(interaction) |>
  select(term, coefficient = estimate) |>
  mutate(part = rep(1:parts, each = ceiling(n()/parts), length.out = n()),
         row_index = rep(1:ceiling(n()/parts), length.out = n())
         ) |>
  split(~ part) |>
  Reduce(f = \(x, y) x |> select(-part) |> left_join(y, by = 'row_index')) |>
  select(-c(row_index, part)) |>
  kable()

|term.x        | coefficient.x|term.y                   | coefficient.y|term                      | coefficient|
|:-------------|-------------:|:------------------------|-------------:|:-------------------------|-----------:|
|(Intercept)   |         29.65|factor(hp)180            |        -13.35|factor(hp)105:factor(am)1 |          NA|
|factor(hp)62  |         -5.25|factor(hp)205            |        -19.25|factor(hp)109:factor(am)1 |          NA|
|factor(hp)65  |          3.50|factor(hp)215            |        -19.25|factor(hp)110:factor(am)1 |       -1.15|
|factor(hp)66  |         -0.55|factor(hp)230            |        -14.95|factor(hp)113:factor(am)1 |          NA|
|factor(hp)91  |         -4.40|factor(hp)245            |        -15.85|factor(hp)123:factor(am)1 |          NA|
## [...]

You can polish up above output with {htmlTables} as described in this SO answer.