2

The sjPlot package (http://www.strengejacke.de/sjPlot) has the tab_model() function to create beautiful html tables for lots of model types. I am trying to use those tables in Overleaf, but I am not sure how to "easily" convert them to latex without losing some of the formatting, etc.

My very hacky approach has been the following (posted it in https://github.com/Rapporter/pander/issues/298).

Any help improving this would be appreciated.

Below, a reproducible example:

library(lme4)
library(sjPlot)
library(XML)
library(RCurl)
library(rlist)
library(janitor)
library(dplyr)
library(knitr)

# This is a terrible model
model = lmer(mpg ~ cyl * disp + (1|vs), mtcars)

# We save the sjPlot table to an .html file (see the table below)
sjPlot::tab_model(
  model,
  show.r2 = TRUE,
  show.icc = FALSE,
  show.re.var = FALSE,
  p.style = "scientific",
  emph.p = TRUE,
  file = "Downloads/temp.html")

enter image description here

Now we can read the .html file and clean it up a bit:

tables <- list.clean(readHTMLTable("~/Downloads/temp.html"), fun = is.null, recursive = FALSE)
tables2 = tables[[1]] %>% janitor::row_to_names(row_number = 1)
tables2 <- as.matrix(tables2) %>% as_tibble()
tables2[is.na(tables2)] <- ""

So now we have the html table in a "clean" dataframe, and we can use kable() to see it in the terminal:

knitr::kable(tables2, format = "pipe")

With this final kable() call we can create the latex code below, which is a reasonable approximation to the initial table... although some important things are missing (bold p values, VD top row...)

kable(
  tables2,
  format = "latex",
  booktabs = TRUE,
  col.names = names(tables2),
  align = c("l", rep("c", length(names(tables2)) - 1)),
  caption = "Means and Standard Deviations of Scores on Baseline Measures"
)

Latex code:

\begin{table}

\caption{\label{tab:}Means and Standard Deviations of Scores on Baseline Measures}
\centering
\begin{tabular}[t]{lccc}
\toprule
Predictors & Estimates & CI & p\\
\midrule
(Intercept) & 49.04 & 39.23 – 58.85 & 1.144e-22\\
cyl & -3.41 & -5.05 – -1.76 & 5.058e-05\\
disp & -0.15 & -0.22 – -0.07 & 2.748e-04\\
cyl * disp & 0.02 & 0.01 – 0.03 & 1.354e-03\\
N vs & 2 &  & \\
\addlinespace
Observations & 32 &  & \\
Marginal R2 / Conditional R2 & 0.809 / NA &  & \\
\bottomrule
\end{tabular}
\end{table}

This is the latex end result:

enter image description here

Of course, this is a toy example, with more complex models, the formatting issues pile up a bit...

Gorka
  • 3,555
  • 1
  • 31
  • 37

2 Answers2

7

In case is useful to someone (and for my future self), I created a Github repo with a function that gets a sjPlot::tab_model() html output and builds tex (and pdf) versions of it. So, using the table above:

# Load html2pdf.R function
source("R/html2pdf.R")
    
# Create tex and pdf
html2pdf(filename = "temp.html", page_width = 13, build_pdf = TRUE, silent = TRUE)

The end result is:

enter image description here

This seems to work for more complex tables too.

Thanks to tjebo, now you can install this as a package and works on Linux and Mac: remotes::install_github("gorkang/html2latex")

Gorka
  • 3,555
  • 1
  • 31
  • 37
2

I don't think it is possible to produce LaTeX result directly from sJPlot. Perhaps the stargazer package can be a suitable alternative while providing direct LaTeX output:

library(lme4)
library(stargazer)

# This is a terrible model
model = lmer(mpg ~ cyl * disp + (1|vs), mtcars)
stargazer(model, title="Regression Results", align=TRUE)

For the following latex result:

enter image description here

flafont11
  • 137
  • 7
  • Thanks for your response! stargazer is OK, but sjPlot has a lot of interesting features I would like to keep. As you say, it is not possible to directly create a latex output with sjPlot,... but as they say, when there is a will, there is a way :) – Gorka Jul 23 '20 at 12:51