Regression tables in Markdown format (for flexible use in R Markdown v2)

Question

The new version of R Markdown is based on pandoc, so you can easyly change the output format.

My Problem is to get markdown formated tables from e.g. regression models, because LATEX and HTML tables do not survive the pandoc conversion.

I know packages that generate LATEX/HTML output from a variety of models (stargazer, texreg, asprtable...) and I'm aware of functions/packages, that generate markdown tables from data frames and matrices but not from other objects.

Any suggestions?

Halo

Try pander that I started for exactly this purpose: rapporter.github.io/pander And Roman is currently working on implementing bunch of new S3 methods in GSoC 2014, so lots of models are already included -- but pls feel free to propose further feature requests on GH. – Stipel

Thanks that fits some of my needs! Is there a way in pander to show results from a series of regression models? – Halo

@user3762565 yes, see e.g. github.com/Rapporter/pander/pull/80 – Stipel

Answer 1

My above comment in more details:

1. Define a few models for reproducible example:

   lm0 <- lm(hp ~ wt, mtcars)
   lm1 <- lm(qsec ~ hp, mtcars)
   lm2 <- lm(qsec ~ wt, mtcars)
   

2. Create a comparative table from those:

   require(memisc)
   mtable123 <- mtable('Model 1' = lm0,
               'Model 2' = lm1,
               'Model 3' = lm2,
               summary.stats = c('R-squared','F','p','N'))
   

3. Render markdown table with a simple call to pander:

   pander(mtable123)
   

4. Enjoy the result:

   --------------------------------------------------
                 Model 1    Model 2    Model 3  
   ----------------- ---------- ---------- ----------
    **(Intercept)**   -1.821\   20.556***\ 18.875***\
                      (32.325)   (0.542)    (1.103)  
   
        **wt**       46.160***\     \       -0.319\  
                      (9.625)               (0.328)  
   
        **hp**           \      -0.018***\     \     
                                 (0.003)             
   
     **R-squared**     0.434      0.502      0.031   
   
         **F**         22.999     30.190     0.945   
   
         **p**         0.000      0.000      0.339   
   
         **N**           32         32         32    
   --------------------------------------------------
   

---

Thanks for Roman Tsegelskyi for implementing this nice feature in GSoC 2014.

Stipel

is it possible to change the format of the table? For example, increase font size? – Mariner

@Mariner it's markdown, thus there is no font size. If you want to convert the document to LaTex/pdf/html or other document formats, then of course there are some options. What format would you like to change? – Stipel

I'm doing a presentation with R Markdown v2 and ioslides. The font that I get in the tables by default is a little too small. Thanks! – Mariner

@Mariner I see, thanks for the details. Well, it's definitely not a markdown question and rather not relevant in this thread, so if I were you, I'd start a new question on SO. Otherwise search for the following keywords: custom CSS with knitr. – Stipel

I tried this solution but I got this error message: "Error in sink(file) : sink stack is full", ..., anyone had this problem? I found a related post, but the solution was to pass the model instead of the model's summary link, but such a solution only works for one model in the table and not for several model using mtable. Any ideas? – Pasahow

@HernandoCasas please verify that you are using the most recent version of pander, as it's very outdated on CRAN. I will push an update there in the next few weeks. – Stipel

Many thx @Stipel that was indeed the problem. It works interactively, but within r code chunk it shows this error: could not find function "asis_output" This is the example: ````{r results='asis'} lm0 <- lm(hp ~ wt, mtcars) lm1 <- lm(qsec ~ hp, mtcars) lm2 <- lm(qsec ~ wt, mtcars) require(memisc) mtable123 <- mtable('Model 1' = lm0, 'Model 2' = lm1, 'Model 3' = lm2, summary.stats = c('R-squared','F','p','N')) require(pander) pander(mtable123) ``` ` – Pasahow

@HernandoCasas are you knitting from RStudio? Probably it starts a vanilla R session without loading knitr or something. I've just pushed a possible update to GH, please reinstall the dev version of pander and let me know if it works. – Stipel

great! thanks a lot @daroczig. That was the problem; knitr was not loaded. Sorry for that!, maybe your update is to require knitr at the time of loading pander?, that would work and it would be helpful for other users to avoid such mistakes in the future. – Pasahow

@HernandoCasas yeah, that's what I did yesterday :) And due to a recent update in pander, there's no need to specify results='asis' any more in your chunk options when calling pander. – Stipel

I seem to be getting: "Warning in pander.default(m): No pander.method for "memisc_mtable"" when I try to do as you suggest. – Consanguinity

@Consanguinity the most recent version of the pander package dropped support for mtable but there are plans to re-integrate it at github.com/Rapporter/pander/issues/304 – Stipel

Answer 2

Just generate the HTML or LATEX tables. All you have to do is to just add results='asis' to the code chunk. It will leave the output as it is.

For example this code using xtable works for me.

```{r,results='asis'}
x<-rnorm(100)
y<-rnorm(100)
lm <- lm(y~x)
library(xtable)
print(xtable(summary(lm)),type='html')
```

Hassi

could you include a minimal working example, together with explanation on how to compile it? – Immediacy

Answer 3

Here what I did some hours ago:

1. Some data:

   ```{r}
   lm1 <- lm(qsec ~ hp, mtcars)
   lm2 <- lm(qsec ~ wt, mtcars)
   ```
   

2. We use the package sjPlot

   ```{r}
   library(sjPlot)
   tab_model(lm1,lm2, file="output.html")# You have to save the table in html format.
   ```
   

   The next part needs to be outside the chunk in markdown:

htmltools::includeHTML("output.html")

Coppins

That does not work for me, could you insert a minimal example, please :) – Appointee

Fixed. Sorry for the mistake – Coppins

'sjt.lm' is deprecated. Use 'tab_model' instead. – Coppins

Answer 4

Another way to use sjPlot (great library for easy presentation of regression output) is to use the no.output feature:

     library(sjmisc)
     library(sjPlot)
     library(magrittr)

     lm(qsec ~ wt, mtcars) %>% 
       sjt.lm(no.output = TRUE, show.se = TRUE) %>% 
       return() %>% .[["knitr"]] %>% asis_output

Nabonidus

Answer 5

I'm updating this with an example that uses the popular (and newer) knitr and kableExtra packages.

library(knitr)
library(xtable)

lm(hp ~ wt, mtcars) %>%
summary() %>%
xtable() %>%
kable()

Now you can access all the cool table-formatting features available in Hao Zhu's kableExtra package.

Midvictorian

Can this do model comparison? – Henig

Answer 6

The huxtable package can now print nicely formatted regression table. See the documentaton https://hughjonesd.github.io/huxtable/huxreg.html

Eaglewood