Table generation in LaTeX using R's stargazer works great. But I can't figure how to format my numbers to whole numbers (this post is related, but different).

Due to data scale, I want coefficients and standard errors to show no decimal. I therefore use the option digits=0. However, this option also drops decimals from the model-fit statistics (eg. R2 = 0 rather than R2 = 0.42). How can different rounding rules be set for the "top" and "bottom" parts of the table?

Using regular expressions, we could replace the particular numbers from LaTeX output s by their rounded values.

Since the texreg package seems to be better maintained at the moment than stargazer, I include both in the answer.

Although it will be difficult to generalize a solution, because the regression results may always be different, here is one idea.

(fit <- lm(mpg ~ hp + cyl, mtcars))
# Call:
#   lm(formula = mpg ~ hp + cyl, data = mtcars)
# 
# Coefficients:
# (Intercept)           hp          cyl  
#    36.90833     -0.01912     -2.26469  

s <- stargazer::stargazer(fit, header=FALSE, digit.separator="")

texreg users would do:

tmp <- tempfile()
texreg::texreg(fit, file=tmp)
s <- readLines(tmp)

For this, first, we want to subset s to not include the GOFs. Also we don't want to replace LaTeX format commands within brackets.

s.sub1 <- 1:(grep("Observations", s) - 1)  ## stargazer
s.sub1 <- 1:(grep("R$^2$", s, fixed=TRUE) - 1)  ## texreg
re <- regexpr("(\\d+\\.\\d+)(?![^[]*\\])", s[s.sub1], perl=TRUE)
toInt <- regmatches(s[s.sub1], re)
v.repl <- round(as.double(toInt), digits=0)

s.sub2 <- which(attr(re, "match.length")[s.sub1] > 1)

After we obtained the replacements v.repl we use mapply in a second step to replace decimals with values rounded to whole numbers.

s[s.sub2] <- mapply(function(x, y) gsub("(\\d+\\.\\d+)(?![^[]*\\])", x, y, perl=TRUE),
    v.repl, s[s.sub2])

Results

stargazer

# \begin{table}[!htbp] \centering 
#   \caption{} 
#   \label{} 
# \begin{tabular}{@{\extracolsep{5pt}}lc} 
# \\[-1.8ex]\hline 
# \hline \\[-1.8ex] 
#  & \multicolumn{1}{c}{\textit{Dependent variable:}} \\ 
# \cline{2-2} 
# \\[-1.8ex] & mpg \\ 
# \hline \\[-1.8ex] 
#  hp & $-$0 \\ 
#   & (0) \\ 
#   & \\ 
#  cyl & $-$2$^{***}$ \\ 
#   & (1) \\ 
#   & \\ 
#  Constant & 37$^{***}$ \\ 
#   & (2) \\ 
#   & \\ 
# \hline \\[-1.8ex] 
# Observations & 32 \\ 
# R$^{2}$ & 0.741 \\ 
# Adjusted R$^{2}$ & 0.723 \\ 
# Residual Std. Error & 3.173 (df = 29) \\ 
# F Statistic & 41.422$^{***}$ (df = 2; 29) \\ 
# \hline 
# \hline \\[-1.8ex] 
# \textit{Note:}  & \multicolumn{1}{r}{$^{*}$p$<$0.1; $^{**}$p$<$0.05; $^{***}$p$<$0.01} \\ 
# \end{tabular} 
# \end{table} 

texreg

# \begin{table}
# \begin{center}
# \begin{tabular}{l c}
# \hline
#  & Model 1 \\
# \hline
# (Intercept) & $37^{***}$ \\
#             & $(2)$      \\
# hp          & $-0$       \\
#             & $(0)$      \\
# cyl         & $-2^{***}$ \\
#             & $(1)$      \\
# \hline
# R$^2$       & $0.74$        \\
# Adj. R$^2$  & $0.72$        \\
# Num. obs.   & $32$          \\
# \hline
# \multicolumn{2}{l}{\scriptsize{$^{***}p<0.001$; $^{**}p<0.01$; $^{*}p<0.05$}}
# \end{tabular}
# \caption{Statistical models}
# \label{table:coefficients}
# \end{center}
# \end{table}

Note: To get the coefficients aligned, you may want to look into the siunitx package for LaTeX. Here a starter on tex.stackexchange. Read here how to include LaTeX packages in Rmarkdown.