Tables - norrissam/ra-manual GitHub Wiki

Capitalization

Variables, titles and panel headers should capitalize only the first word, not each word. For example, "The effect of parental incarceration on child crime," not "The Effect of Parental Incarceration on Child Crime."

Observations

Observations should be reported with commas at the bottom of the table and labelled "Observations" not "N".

esttab, ... stats(N, fmt(%9.0fc) labels("Observations"))

Fixed Effects

Fixed effects should be indicated with Yes and No.

Significance

Stars should indicate 10, 5, and 1% significance. The level of clustering should be indicated in the table notes.

Common statistics we add

• p-values are usually reported with up to 3 decimal points. The "p" should be italicized.
• F-stat: for IV regressions, we want to report the F-stat on the first stage. We usually report up to 2 decimal points. The F is not italicized.
• J-test: We also report the chi squared statistic from the J-test of overidentification (italicize the J) and the corresponding p-value:

ivreghdfe y (x = z), absorb(fe) cluster(id)
estadd scalar fstat = `e(F)'
estadd scalar jtest = e(j)
estadd scalar jpval = e(jp)

esttab, //
...
stats(fstat newline jtest jpval obs, ///
      fmt(%9.2f %9.0f %9.2f %9.3f %9.0fc) ///
      labels("First-stage F-stat" " " "\(J\) test of overidentification" "\quad \(\chi^2\) statistic" "\quad \(p\)-value" "Observations"))

Here is a tex snippet of statistics added to the bottom of a table:

First-stage F-stat  &      &      177.76    &      177.76    &      177.76 \\
\(J\) test of overidentification&      &      &      &      &      &      &      \\
\quad \(\chi^2\) statistic&      &      11.087    &      15.092    &  1.450  \\
\quad \(p\)-value   &      &  0.026    &  0.005    &  0.835  \\

Table notes

Table notes are added in the draft. We use threeparttable in tex to do this.

Multipanel tables

Some tables require more formatting than esttab allows for. For multipanel tables, we rely on the suite of MultiPartTab ado files. For each program, here is an example of the tex output produced:

MultiPartTabStart allows you to specify column names and supercolumn headers + lines (\cmidrule) to group similar outcomes.

\begin{tabular}{l*{6}{c}} \toprule
 &\multicolumn{2}{c}{Call responses}  &\multicolumn{3}{c}{Call outcomes} \\\
 cmidrule(lr){2-3}\cmidrule(lr){4-6}
 &\multicolumn{1}{c}{Police}  &\multicolumn{1}{c}{Fire}  &\multicolumn{1}{c}{Arrest}  &\multicolumn{1}{c}{Involuntary hold} &\multicolumn{1}{c}{Warrant} \\
 &\multicolumn{1}{c}{(1)}  &\multicolumn{1}{c}{(2)}  &\multicolumn{1}{c}{(3)}  &\multicolumn{1}{c}{(4)}  &\multicolumn{1}{c}{(5)} \\
\midrule

For each panel, use MultiPartTabPanelStart to name panel headers. Panel headers should be named alphabetically and labelled (eg. "Panel A: OLS estimates" "Panel B: IV estimates").

\multicolumn{6}{c}{\emph{Panel B: IV estimates of effect of police response}} \\

Use MultiPartTabPanelEnd to output results as a fragment (esttab output without the header and footer), and choose which statistics (dependent mean, F-stat, etc.) to report in the panel.

Police    &  -0.378\sym{***}&   0.055    &  -0.232\sym{***}&  -0.151\sym{***}&  -0.050\sym{***}&   0.025    \\
   &  (0.083)    &  (0.081)    &  (0.061)    &  (0.023)    &  (0.013)    &  (0.025)    \\
\addlinespace
Dependent mean & 0.945 & 0.945 & 0.309 & 0.171 & 0.054 \\

MultiPartTabEnd closes the table.