Generating regression tables in R

2019/11/25

I often need to document the statistical results I estimate in table format. I have tried many, many things over the years, and none of my solutions are perfect, including the one I’m about to describe. But, it is now… pretty good.

First, I define a function that takes a list of fitted models (models) and some other variables and outputs a list of pieces that I can create a table with. See below for the function definition.

library(stargazer)

#' @param models A list of fitted models that stargazer can process
#' @param keep Length 1 character vector of variables to display in table
#' @param covariate.labels Labels for keep
#' @param digits Number of digits to use for numbers in the table
#'
#' @return List of pieces of a tabular with named items header, inner, and footer
make_tex_pieces <- function(models, keep, covariate.labels, digits = 2) {
# models: a
# Use stargazer, but keep as little extra stuff as possible
tex_raw <- stargazer(models,
keep = keep, covariate.labels = covariate.labels,
digits = digits,
table.layout = "t", no.space = T, align = T)

# Split up into header, footer, and inner
idx0 <- grep("begin{tabular}", tex_raw, fixed = T) # Start of \begin{tabular}
idx1 <- grep("end{tabular}", tex_raw, fixed = T) # End of \begin{tabular}

tex_footer <- c("\\bottomrule", tex_raw[idx1])

# Remove [-1.8ex] and get the inside of the tabular
tex_inner <- gsub("\\\\[-[\\.0-9]+ex]", "", tex_raw[(idx0+1):(idx1-1)])

# Return these as a 3 element list so that the user can insert header rows (column labels)
# and footer rows (summary statistics, fixed effects)
}

Once I have that function defined, I can use it to create the inside part of the table: the tabular command.

# Load a sample dataset and run regression
data(cars)
fit <- lm(speed ~ dist, data = cars)

# Use the function we defined above to split the regression output into different pieces of a tabulr
pieces <- make_tex_pieces(fit, "dist", "distance")

# Put the pieces back together, adding a short panel with the count of observations
latex_output <- c(pieces$header, pieces$inner,
"\\midrule",
sprintf("Observations & %.0f \\\\", length(fit$model$dist)),
pieces$footer) # Write to file write(latex_output, "model-tabular.tex") Next, I use the LaTeX threeparttable package (also used in this post) to display the table. Here’s a minimum example. \documentclass{article} \usepackage{booktabs} % Nice-looking tables \usepackage{dcolumn} % Booktabs column spacing \usepackage{threeparttable} % Align column caption, table, and notes % Flexible notes environment based on minipage \newenvironment{notes}[1][Notes]{\begin{minipage}[t]{\linewidth}\normalsize{\itshape#1: }}{\end{minipage}} \begin{document} \begin{table} \centering \begin{threeparttable} \caption{My table} \input{model-tabular.tex} \begin{notes} * p$<$0.1, ** p$<$0.05, *** p$<\$ 0.01. This regression is not confounded at all.
\end{notes}
\end{threeparttable}
\end{table}

\end{document}

And here’s the result.

Other packages you might find useful:

• huxtable is a good solution for generating quick regression tables for export to Markdown or HTML. I find its LaTeX output functions fairly cumbersome.
• kable/kableExtra are great for general purpose table creation, but can’t easily process fitted model output.