This vignette shows how to add rows to panelsummary
tables. Since panelsummary::panelsummary returns
kableExtra objects, adding rows is impossible using the
base function panelsummary::panelsummary. This can be a
nuisance when you want to include additional statistics to your table.
However, the workaround is to use the
panelsummary::panelsummary_raw function.
panelsummary::panelsummary_raw
The panelsummary::panelsummary_raw function returns a
dataframe that can be manipulated. The trade-off is that the table will
need to be customized using the kableExtra package,
although some of this customization can be completed using the
panelsummary::clean_raw function (see example).
Workflow
If you want to create a table with panels that has additional statistical rows, the workflow will be the following:
- Estimate your models (preferably with the
fixestpackage). - Pass your models into
panelsummary::panelsummary_rawto get a manipulable dataframe. - Manipulate the dataframe (preferably with
dplyr::add_row). - Use
panelsummary::clean_rawto return akableExtraobject. - Use
kableExtra::pack_rowsto create panel labels.
Example: Adding Additional Rows to Each Panel
To demonstrate how to add a row to each panel of a
panelsummary::panelsummary_raw table, I will assume that
each panel needs one additional row for the wild-cluster-bootstrap
p-values of the cyl column from the mtcars
data. While I will not explicitly calculate these (in fact, they will be
completely made up), the following steps will show how to complete such
task.
Consider the following models, estimated using the
fixest package:
library(panelsummary)
## Panel A: mpg dependent variable
mpg_1 <- fixest::feols(mpg ~cyl + disp, data = mtcars)
mpg_2 <- fixest::feols(mpg ~ cyl + am, data = mtcars)
## Panel B: wt dependent variable
wt_1 <- fixest::feols(wt ~ cyl + disp, data = mtcars)
wt_2 <- fixest::feols(wt ~ cyl + am, data = mtcars)Recall that putting each of these two models in a list will result in multiple specifications per-panel.
## putting each panel's models into a list—one for each panel
panel_a_mpg <- list(mpg_1, mpg_2)
panel_b_wt <- list(wt_1, wt_2)Now, I will pass these two lists into
panelsummary::panelsummary_raw. Note that this function
shares many of the same arguments as
panelsummary::panelsummary. In the following, I will also
use the arguments stars (significance stars),
mean_dependent (for dependent variable mean), and
gof_omit (to omit \(R^2\)
statistics):
## creating the dataframe
panelsummary_raw(panel_a_mpg, panel_b_wt,
stars = "econ",
mean_dependent = T,
gof_omit = "^R")
#> term Model 1 Model 2
#> 1 (Intercept) 34.661*** 34.522***
#> 2 (2.547) (2.603)
#> 3 cyl -1.587** -2.501***
#> 4 (0.712) (0.361)
#> 5 disp -0.021*
#> 6 (0.010)
#> 7 am 2.567*
#> 8 (1.291)
#> 9 Mean of Dependent Variable 20.091 20.091
#> 10 Num.Obs. 32 32
#> 11 AIC 165.1 165.2
#> 12 BIC 169.5 169.6
#> 13 Std.Errors IID IID
#> 14 (Intercept) 1.773*** 1.566***
#> 15 (0.386) (0.453)
#> 16 cyl -0.054 0.317***
#> 17 (0.108) (0.063)
#> 18 disp 0.008***
#> 19 (0.002)
#> 20 am -0.765***
#> 21 (0.225)
#> 22 Mean of Dependent Variable 3.217 3.217
#> 23 Num.Obs. 32 32
#> 24 AIC 44.4 53.3
#> 25 BIC 48.8 57.7
#> 26 Std.Errors IID IIDThe return value of this output is a dataframe without panel labels.
Given that it is a dataframe, you can use dplyr functions
to easily manipulate. For instance, I will use the
dplyr::add_rows function to add the pseudo
wild-cluster-bootstrap p-values to the table:
## adding rows using dplyr::add_row
bootstrap_table <- panelsummary_raw(panel_a_mpg, panel_b_wt,
stars = "econ",
mean_dependent = T,
gof_omit = "R") |>
dplyr::add_row(term = "Wild Cluster Boot P-value (cyl)",
`Model 1` = ".001",
`Model 2` = ".002",
.before = 10) |>
dplyr::add_row(term = "Wild Cluster Boot P-value (cyl)",
`Model 1` = ".001",
`Model 2` = ".005",
.before = 22)
bootstrap_table
#> term Model 1 Model 2
#> 1 (Intercept) 34.661*** 34.522***
#> 2 (2.547) (2.603)
#> 3 cyl -1.587** -2.501***
#> 4 (0.712) (0.361)
#> 5 disp -0.021*
#> 6 (0.010)
#> 7 am 2.567*
#> 8 (1.291)
#> 9 Mean of Dependent Variable 20.091 20.091
#> 10 Wild Cluster Boot P-value (cyl) .001 .002
#> 11 Num.Obs. 32 32
#> 12 AIC 165.1 165.2
#> 13 BIC 169.5 169.6
#> 14 Std.Errors IID IID
#> 15 (Intercept) 1.773*** 1.566***
#> 16 (0.386) (0.453)
#> 17 cyl -0.054 0.317***
#> 18 (0.108) (0.063)
#> 19 disp 0.008***
#> 20 (0.002)
#> 21 am -0.765***
#> 22 Wild Cluster Boot P-value (cyl) .001 .005
#> 23 (0.225)
#> 24 Mean of Dependent Variable 3.217 3.217
#> 25 Num.Obs. 32 32
#> 26 AIC 44.4 53.3
#> 27 BIC 48.8 57.7
#> 28 Std.Errors IID IIDNow the dataframe is ready to be passed into
kableExtra::kbl to further customize it with panel
headers/column names etc. Alternatively, you can use the
panelsummary::clean_raw function to get a few desirable
traits and speed up the process. In particular, the
panelsummary::clean_raw function does the following by
default:
- Passes the dataframe into
kableExtra::kbl - Changes the alignment to Left for the first column, and Center for all remaining columns.
- Changes the names of the columns to a blank white space for the first column, and consecutive numbers in the remaining columns.
However, you can also pass in a caption using the
caption argument:
## creating a kableExtra object with a few desirable defaults
bootstrap_table |>
clean_raw(caption = "A customized panelsummary table with added rows.")|
|
|
|
|---|---|---|
| (Intercept) | 34.661*** | 34.522*** |
| (2.547) | (2.603) | |
| cyl | -1.587** | -2.501*** |
| (0.712) | (0.361) | |
| disp | -0.021* | |
| (0.010) | ||
| am | 2.567* | |
| (1.291) | ||
| Mean of Dependent Variable | 20.091 | 20.091 |
| Wild Cluster Boot P-value (cyl) | .001 | .002 |
| Num.Obs. | 32 | 32 |
| AIC | 165.1 | 165.2 |
| BIC | 169.5 | 169.6 |
| Std.Errors | IID | IID |
| (Intercept) | 1.773*** | 1.566*** |
| (0.386) | (0.453) | |
| cyl | -0.054 | 0.317*** |
| (0.108) | (0.063) | |
| disp | 0.008*** | |
| (0.002) | ||
| am | -0.765*** | |
| Wild Cluster Boot P-value (cyl) | .001 | .005 |
| (0.225) | ||
| Mean of Dependent Variable | 3.217 | 3.217 |
| Num.Obs. | 32 | 32 |
| AIC | 44.4 | 53.3 |
| BIC | 48.8 | 57.7 |
| Std.Errors | IID | IID |
Finally, to get the desired panel labels, you will need to use
kableExtra::pack_rows:
## creating the final panels.
bootstrap_table |>
clean_raw(caption = "A customized panelsummary table with added rows.") |>
kableExtra::pack_rows("Panel A: MPG", 1, 14,
italic = T,
bold = F) |>
kableExtra::pack_rows("Panel B: Cyl", 15, 28,
italic = T,
bold = F) |>
kableExtra::footnote(list("* p < 0.1, ** p < 0.05, *** p < 0.01")) |>
kableExtra::kable_classic(full_width = F, html_font = "Cambria") |
|
|
|
|---|---|---|
| Panel A: MPG | ||
| (Intercept) | 34.661*** | 34.522*** |
| (2.547) | (2.603) | |
| cyl | -1.587** | -2.501*** |
| (0.712) | (0.361) | |
| disp | -0.021* | |
| (0.010) | ||
| am | 2.567* | |
| (1.291) | ||
| Mean of Dependent Variable | 20.091 | 20.091 |
| Wild Cluster Boot P-value (cyl) | .001 | .002 |
| Num.Obs. | 32 | 32 |
| AIC | 165.1 | 165.2 |
| BIC | 169.5 | 169.6 |
| Std.Errors | IID | IID |
| Panel B: Cyl | ||
| (Intercept) | 1.773*** | 1.566*** |
| (0.386) | (0.453) | |
| cyl | -0.054 | 0.317*** |
| (0.108) | (0.063) | |
| disp | 0.008*** | |
| (0.002) | ||
| am | -0.765*** | |
| Wild Cluster Boot P-value (cyl) | .001 | .005 |
| (0.225) | ||
| Mean of Dependent Variable | 3.217 | 3.217 |
| Num.Obs. | 32 | 32 |
| AIC | 44.4 | 53.3 |
| BIC | 48.8 | 57.7 |
| Std.Errors | IID | IID |
| Note: | ||
| * p < 0.1, ** p < 0.05, *** p < 0.01 | ||