It's often necessary to produce an output table that doesn't fit any of the format provided by the standard tools outreg, esttab, etc.. Some examples:
- Showing p-values for differences of coefficients
- Putting certain coefficients in bold
- Putting different outcome variables in the same row of a regression table
- Multi-panel tables with different formatting in each panel
Stata-tex allows you to separate the LaTeX table template from the table data. This lets you set up and compile exactly the LaTeX table you want, with placeholders for the data. Then you can generate the data separately, and transfer it into the LaTeX table automatically.
- Arbitrary customization of Stata/LaTeX tables
- Iterate on the LaTeX table without regenerating estimates
- Stata code 770C is much much cleaner
- Can get LaTeX table looking right without using Stata
- Easier to copy table templates to new contexts
Create a latex table, with placeholders for all the numbers you want
to put in later. Placeholders are marked with $$, e.g. $$beta1$$. So
you create a regular latex table, with lines like this:
New Road & $$app1_beta$$ & $$app2_beta$$ \\ & $$app1_se$$ & $$app2_se$$ \\
In this example, app1 and app2 are estimates from different
regressions -- app1_beta is a placeholder for the regression estimate,
and app1_se for the standard error.
Compile this table in LaTeX until satisfied with the format.
In Stata, generate a data file with the table estimates in CSV format:
app1_p, 0.71
app1_beta, -0.000
app1_starbeta, -0.000
app1_se, 0.001
app1_n, 143998
app1_r2, 0.00
app2_p, 0.00
app2_beta, 0.009
app2_starbeta, 0.009***
app2_se, 0.001
app2_n, 147143
app2_r2, 0.00
The post-estimation command store_est_tpl generates this file
automatically. Following an estimation, the command:
store_est_tpl using table_data.csv, coef(treatment_comp) name(app1) all
writes the top six lines to the above file for you. Note that the
placeholder beta holds just the beta coefficient, while starbeta
calculates a p-value and shows stars for p<0.1, p<0.05,
p<0.01. e.g. beta contains 0.06, and starbeta contains 0.06**.
store_est_tpl takes an optional format() parameter that lets you
specify a different format for the coefficient and standard error
(e.g. "%5.2f" -- the default is 3 decimal points). p-values and r2
always have 2 decimal points.
The all parameter is short for beta se n p r2 -- the latter let
you choose exactly what estimation statistics you want to
output. Outputting statistics you don't use is fine, so all is
usually the right answer.
If you want to store some other value in this file, e.g. a p-value from an F test or significance test for a difference between two coefficients, you can store an arbitrary string using:
append_to_file using table_data.csv, s("app2_ftest, 0.35")
Finally, use table_from_tpl to transfer the estimates into the LaTeX
template (via Python):
table_from_tpl, t(two_panel.tex) r(table_data.csv) o(output_table.tex)
t = template file, r = replacement data file, o = output file
The easiest way to get this running may be to work through the example
files table_tpl.do and two_panel.tex which create a contrived two
panel output table from a system dataset.
-
Unlike
outreg,estout, etc., you can't add and remove columns from these tables without modifying the LaTeX template. This is an inherent limitation of having a totally customizable table template. -
If a placeholder appears multiple times in the output file, the first appearance will be used. So you need to delete the output file each time you generate data. (It would probably be better to use the last appearance, then you could just keep adding to the same estimate data file.)
table_from_tpl() needs to find table_from_tpl.py either in the current folder, or in the folder specified by the global $PYTHONPATH.