Using panelsummary with fixest (original) (raw)
This vignette focuses on using the panelsummary package with the fixest package. fixest has become one of the most popular regression packages for economists, and therefore further capabilities have been implemented in panelsummaryto work seamlessly with fixest. There will be two main capabilities that this vignette covers:
- Adding the mean of the dependent variable.
- Collapsing fixed effects.
Adding the Mean of the Dependent Variable
To start, load in the fixest, panelsummary, and kableExtra packages.
For demonstration purposes, four regressions are estimated with thefixest package using the feols function and the mtcars data:
## estimating two models for mpg
mpg_1 <- mtcars |>
feols(mpg ~ cyl | gear + carb, cluster = ~hp)
mpg_2 <- mtcars |>
feols(mpg ~ cyl | gear + carb + am, cluster = ~hp)
## estimating the same two models for disp
disp_1 <- mtcars |>
feols(disp ~ cyl | gear + carb, cluster = ~hp)
disp_2 <- mtcars |>
feols(disp ~ cyl | gear + carb + am, cluster = ~hp)To create a regression table with two panels (Panel A: MPG/Panel B: Displacement) and two models in each, the listslist(mpg_1, mpg_2) and list(disp_1, disp_2)can be passed into the arguments for .... However, since each of these models is of the class fixest (i.e., created with the fixest package), the mean_dependentargument can be set to TRUE to show the means of the dependent variable in each column:
panelsummary(list(mpg_1, mpg_2), list(disp_1, disp_2),
mean_dependent = TRUE,
panel_labels = c("Panel A: MPG", "Panel B: Displacement"),
caption = "Automated Mean of Dependent Variable") |>
kable_classic(full_width = F, html_font = "Cambria") Automated Mean of Dependent Variable
| | | | | | -------------------------- | -------- | ------- | | Panel A: MPG | | | | cyl | -0.907 | -1.093 | | | (0.789) | (0.801) | | | Mean of Dependent Variable | 20.091 | 20.091 | | Num.Obs. | 32 | 32 | | R2 | 0.818 | 0.835 | | R2 Adj. | 0.755 | 0.768 | | R2 Within | 0.058 | 0.088 | | R2 Within Adj. | 0.017 | 0.046 | | AIC | 168.3 | 167.1 | | BIC | 181.4 | 181.8 | | RMSE | 2.53 | 2.41 | | Std.Errors | by: hp | by: hp | | FE: gear | X | X | | FE: carb | X | X | | FE: am | | X | | Panel B: Displacement | | | | cyl | 43.306 | 43.905 | | | (13.075) | (13.489) | | | Mean of Dependent Variable | 230.722 | 230.722 | | Num.Obs. | 32 | 32 | | R2 | 0.909 | 0.910 | | R2 Adj. | 0.878 | 0.873 | | R2 Within | 0.399 | 0.401 | | R2 Within Adj. | 0.373 | 0.373 | | AIC | 339.5 | 341.4 | | BIC | 352.7 | 356.0 | | RMSE | 36.76 | 36.68 | | Std.Errors | by: hp | by: hp | | FE: gear | X | X | | FE: carb | X | X | | FE: am | | X |
Notice that a new row was placed before the “Num.Obs.” row in each panel with the mean of the dependent variable. This is the default setting. If you would like to change where the mean is placed, you will need to customize using the [panelsummary::panelsummary_raw](../reference/panelsummary%5Fraw.html)function (see the Adding Rows to a panelsummary Table vignette for more details). Now, usinggof_map (see the modelsummary website) the goodness-of-fit statistics will be mapped/renamed.Note that the mean of the dependent variable is not included in the mapped statistics and cannot be reordered without using panelsummary::panelsummary_raw.
## creating a renaming tibble to pass into panelsummary
gm <- tibble::tribble(
~raw, ~clean, ~fmt, ~omit,
"nobs", "Observations", 0, FALSE,
"FE: gear", "FE: Gear", 0, FALSE,
"FE: carb", "FE: Carb", 0, FALSE,
"FE: am", "FE: AM", 0, FALSE
)
## creating the regression table
panelsummary(list(mpg_1, mpg_2), list(disp_1, disp_2),
mean_dependent = TRUE,
panel_labels = c("Panel A: MPG", "Panel B: Displacement"),
gof_map = gm,
caption = "Automated Mean of Dependent Variable-Renamed and Reordered") |>
kable_classic(full_width = F, html_font = "Cambria") Automated Mean of Dependent Variable-Renamed and Reordered
| | | | | | -------------------------- | -------- | ------- | | Panel A: MPG | | | | cyl | -0.907 | -1.093 | | | (0.789) | (0.801) | | | Mean of Dependent Variable | 20.091 | 20.091 | | Observations | 32 | 32 | | FE: Gear | X | X | | FE: Carb | X | X | | FE: AM | | X | | Panel B: Displacement | | | | cyl | 43.306 | 43.905 | | | (13.075) | (13.489) | | | Mean of Dependent Variable | 230.722 | 230.722 | | Observations | 32 | 32 | | FE: Gear | X | X | | FE: Carb | X | X | | FE: AM | | X |
Lastly, as an important aside, any models that are not of classfixest will not return a dependent variable mean. There are plans to accommodate more models for this feature in the future.
Collapsing Fixed Effects
Notice that in the tables above that the fixed effects in columns (1) and (2) are the same across each panel. To reduce the number of rows, the collapse_fe argument can be used to show a separate panel containing only the fixed effects. Observe that there is also a warning message returned when this command is used; the warning message is telling the user that the function assumes both panels have the same fixed effects for the same models.
## collapsing fixed effects with collapse_fe = T
panelsummary(list(mpg_1, mpg_2), list(disp_1, disp_2),
mean_dependent = TRUE,
panel_labels = c("Panel A: MPG", "Panel B: Displacement"),
gof_map = gm,
caption = "Collapsed Fixed Effects",
collapse_fe = T) |>
kable_classic(full_width = F, html_font = "Cambria") |>
footnote("Collapsing the fixed effects assumes the fixed effects in all panels are the same!")
#> Warning in panelsummary(list(mpg_1, mpg_2), list(disp_1, disp_2),
#> mean_dependent = TRUE, : panelsummary does not check if the fixed effects in
#> each panel match. It always assumes they do!Collapsed Fixed Effects
| | | | | | ---------------------------------------------------------------------------------- | -------- | ------- | | Panel A: MPG | | | | cyl | -0.907 | -1.093 | | | (0.789) | (0.801) | | | Mean of Dependent Variable | 20.091 | 20.091 | | Observations | 32 | 32 | | Panel B: Displacement | | | | cyl | 43.306 | 43.905 | | | (13.075) | (13.489) | | | Mean of Dependent Variable | 230.722 | 230.722 | | Observations | 32 | 32 | | FE: Gear | X | X | | FE: Carb | X | X | | FE: AM | | X | | Note: | | | | Collapsing the fixed effects assumes the fixed effects in all panels are the same! | | |
Although it does not appear in HTML, a LaTeX version of this table will include a horizontal line separating the fixed effects from the other panels. This can be switched off using thehline_before_fe argument.