Wrapping custom notes in texreg output

Question

I'm trying to add a fairly long note to the bottom of a table created by texreg; I want this to simply wrap around, but there doesn't seem to be any functionality built into the function for doing so.

Take, e.g.:

texreg(
  lm(speed ~ dist, data = cars),
  custom.note = paste(
    "%stars. This regression should be",
    "intepreted with strong caution as",
    "it is likely plagued by extensive", 
    "omitted variable bias"
  )
)

Which, when compiled, gives something like:

The formatting is atrocious; much better would be something like replacing the standard output:

\multicolumn{2}{l}{\scriptsize{$^{***}p<0.001$, $^{**}p<0.01$, $^*p<0.05$. This regression should be intepreted with strong caution as it is likely plagued by extensive omitted variable bias}}

With more digestible wrapping:

\multicolumn{2}{l}{\scriptsize{$^{***}p<0.001$, $^{**}p<0.01$, $^*p<0.05$.}} \\
\multicolumn{2}{l}{\scriptsize{This regression should be intepreted with}} \\
\multicolumn{2}{l}{\scriptsize{strong caution as it is likely plagued by}} \\
\multicolumn{2}{l}{\scriptsize{extensive omitted variable bias}}

Which gives output much closer to what I'm looking for:

Is there a way to do this programatically?

Answer 1

With version 1.37.1 (released in May 2020), texreg introduces the threeparttable argument, which uses the threeparttable LaTeX package, which was designed for this purpose.

Example R code:

texreg(lm(speed ~ dist, data = cars),
       custom.note = paste("\\item %stars. This regression",
                           "should be interpreted with strong",
                           "caution as it is likely plagued by",
                           "extensive omitted variable bias."),
       single.row = TRUE,
       threeparttable = TRUE)

Output:

\begin{table}
\begin{center}
\begin{threeparttable}
\begin{tabular}{l c}
\hline
 & Model 1 \\
\hline
(Intercept) & $8.28 \; (0.87)^{***}$ \\
dist        & $0.17 \; (0.02)^{***}$ \\
\hline
R$^2$       & $0.65$                 \\
Adj. R$^2$  & $0.64$                 \\
Num. obs.   & $50$                   \\
\hline
\end{tabular}
\begin{tablenotes}[flushleft]
\scriptsize{\item $^{***}p<0.001$; $^{**}p<0.01$; $^{*}p<0.05$. This regression should be interpreted with strong caution as it is likely plagued by extensive omitted variable bias}
\end{tablenotes}
\end{threeparttable}
\caption{Statistical models}
\label{table:coefficients}
\end{center}
\end{table}

Which is rendered as:

Note that the custom note must start with \\item. It is also possible to have multiple items and/or use bullet points to format multiple notes like in a list:

texreg(lm(speed ~ dist, data = cars),
       custom.note = paste("\\item[$\\bullet$] %stars.",
                           "\\item[$\\bullet$] This regression",
                           "should be interpreted with strong",
                           "caution as it is likely plagued by",
                           "extensive omitted variable bias."),
       single.row = TRUE,
       threeparttable = TRUE)

The formatting is not perfect as you cannot set the desired width of the table; the note just adjusts to the width of the respective table. But I think it should be less of a problem in a realistic usage scenario where more than one model is displayed at a time and some of the coefficient names are longer than in the example. This solution also supports longtable environments, in which case the threeparttablex package is used instead.

Here is an example of how you could make it look nice with two models:

fit <- lm(speed ~ dist, data = cars)
texreg(list(fit, fit),
       custom.note = paste("\\item[\\hspace{-5mm}] %stars.",
                           "\\item[\\hspace{-5mm}] This regression",
                           "should be interpreted with strong",
                           "caution as it is likely plagued by",
                           "extensive omitted variable bias."),
       single.row = TRUE,
       threeparttable = TRUE)

This yields:

\begin{table}
\begin{center}
\begin{threeparttable}
\begin{tabular}{l c c}
\hline
 & Model 1 & Model 2 \\
\hline
(Intercept) & $8.28 \; (0.87)^{***}$ & $8.28 \; (0.87)^{***}$ \\
dist        & $0.17 \; (0.02)^{***}$ & $0.17 \; (0.02)^{***}$ \\
\hline
R$^2$       & $0.65$                 & $0.65$                 \\
Adj. R$^2$  & $0.64$                 & $0.64$                 \\
Num. obs.   & $50$                   & $50$                   \\
\hline
\end{tabular}
\begin{tablenotes}[flushleft]
\scriptsize{\item[\hspace{-5mm}] $^{***}p<0.001$; $^{**}p<0.01$; $^{*}p<0.05$. \item[\hspace{-5mm}] This regression should be interpreted with strong caution as it is likely plagued by extensive omitted variable bias.}
\end{tablenotes}
\end{threeparttable}
\caption{Statistical models}
\label{table:coefficients}
\end{center}
\end{table}

Which renders as:

Answer 2

I've come up with a workaround so far by rewriting the texreg function by adding a custom.note.wrap argument and changing:

note <- paste0("\\multicolumn{", length(models) + 1, 
               "}{l}{\\", notesize, "{", custom.note, "}}")
note <- gsub("%stars", snote, note, perl = TRUE)

To:

if (custom.note.wrap){
  note<-paste(paste0("\\multicolumn{", length(models) + 1L,"}{l}{\\",notesize,"{",
                     strwrap(custom.note, width=custom.note.wrap), "}}"),
              collapse = " \\ \n")
  note <- gsub("%stars", snote, note, perl = TRUE)
}else{
  note <- paste0("\\multicolumn{", length(models) + 1L, 
                 "}{l}{\\", notesize, "{", custom.note, "}}")
  note <- gsub("%stars", snote, note, perl = TRUE)
}

The idea is to pick a maximum string length for each line (custom.note.wrap) and then split the supplied note into strings of at most that length which end in a space, finally concatenating everything into a bunch of multicolumns with each split substring.

This is not optimal, as it would be better for texreg (to have the ability) to automatically set custom.note.wrap given the lengths of the model names, etc. But my raw LaTeX abilities are lacking, so I'm not sure how I would do this.

Answer 3

I might point out a neat alternative solution I received which could be of interest to you, at the latest when you need to update the texreg package.

Accordingly the custom note ends in a \multicolumn in the LaTeX code, thus we cannot use line break commands like par or \\. But we can achieve the automatic line break with \parbox. If we still want a custom line break we can use four backslashes \\\\. For better formatting we can use \\vspace{2pt} just at the beginning of the text content:

texreg(lm(speed ~ dist, data = cars),
       custom.note = ("\\parbox{.4\\linewidth}{\\vspace{2pt}%stars. \\\\
       This regression should be intepreted with strong caution as it is 
       likely plagued by extensive omitted variable bias.}"))