Overview

In this lesson we hope to learn:

• Quick and basic export of results
• Writing a basic report
• Exporting graphics for use in other documents
• Reproducible research

Why does Export Matter?

• Need to be able to share work with others and present it
• This can be tricky in R because R has so many choices for export
• R can talk to a number of outside formats including Excel, Word, PDF, PNG, HTML
• R can even be used as a web service: https://public.opencpu.org/pages/
• We’ll cover the basics, but there are a lot of outside resources for doing what you need in your own environment

Generating a basic report

• There are a few key concepts that R allows that we should follow when preparing a report on data

1. Include the data, source code, and output together in one package
2. Have the source code available for raw data to finished product
3. Present figures, tables, and code in a single document

• Why do we do this?

• Transparency
• Reproducibility

• It isn’t much harder than the basic analysis itself

A few terms

• Rmarkdown extension .Rmd and used for R flavored markdown in RStudio
• R script extension .R used to store R code only
• R HTML extension .RHtml used to store R HTML
• Text extension .txt used to write plain text
• * R Sweave* extension .Rnw used to write LaTeX reports with embedded R code

Beginning

• Open a new script file in RStudio–“R Markdown”
• This opens a template file for an HTML report, the easiest type to create
• In R there are a number of packages designed to help create reports and place them on a scale like so:

Get the tools

• install.packages('knitr','markdown'); sweave is part of R base already
• We need a LaTeX distribution for Tex documents
• What is LaTeX?
• An open source typsetting framework that can be used to produce complex technical documents
• We won’t mess with LaTeX today - instead we will focus on HTML and markdown

A Simple Example

• Paste this in a new R script
• Save the script in your working directory as myscript.R

#' This is some text
#'

# + myplot, dev='svg',out.width='500px',out.height='400px'

library(ggplot2)
data(diamonds)
qplot(carat, price, data = diamonds, alpha = I(0.3), color = clarity)

#' Diamond size is clearly related to price, but not in a linear fashion.
#'

Converting the Script

• This happens in two steps to make an HTML file

o <- spin("C:/Path/To/myscript.R", knit = FALSE)
knit2html(o, envir = new.env())

Example Script II

• Create myscript2.R

#' This is some text that I want to explain
#' For example, this plot is important, let's look below

# + myplot,
# dev='svg',out.width='500px',out.height='400px',warning=FALSE,message=FALSE

library(ggplot2)
load("PATH/TO/MY/DATA.rda")
qplot(readSS, mathSS, data = df, alpha = I(0.2)) + geom_smooth()

#' There is not a linear relationship, but it sure is close.
#' Let's do some regression
#'

test <- lm(mathSS ~ readSS + factor(grade), data = df)
summary(test)

#' It's all statistically significant

Spin 2

o <- spin("C:/Path/To/myscript2.R", knit = FALSE)
knit2html(o, envir = new.env())
# We specify that new environment is used to carry out the analysis, not
# the current environment

Stitch

• Stiching is similar to spinning, but we don’t even have to mess around with anything but our R script
• Just write the R code
• Any comments you write preceeded by the # denote something we want printed in the text
• Even more simplified than spin

Stitch Example

## title: My Super Report ## Author: Mr. Data ##

# A plot and some text
library(ggplot2)
load("PATH/TO/MY/DATA")
qplot(readSS, mathSS, data = df, alpha = I(0.2)) + geom_smooth()

# Now a linear model
test <- lm(mathSS ~ readSS + factor(grade), data = df)
summary(test)

## 
## Call:
## lm(formula = mathSS ~ readSS + factor(grade), data = df)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -170.47  -43.35    1.21   45.45  194.45 
## 
## Coefficients:
##                Estimate Std. Error t value Pr(>|t|)    
## (Intercept)     198.343      7.252   27.35  < 2e-16 ***
## readSS            0.478      0.015   31.96  < 2e-16 ***
## factor(grade)4   30.837      4.324    7.13  1.3e-12 ***
## factor(grade)5   34.225      4.112    8.32  < 2e-16 ***
## factor(grade)6   62.517      4.418   14.15  < 2e-16 ***
## factor(grade)7   72.468      4.265   16.99  < 2e-16 ***
## factor(grade)8   96.530      4.650   20.76  < 2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 
## 
## Residual standard error: 64.4 on 2693 degrees of freedom
## Multiple R-squared: 0.497,   Adjusted R-squared: 0.496 
## F-statistic:  444 on 6 and 2693 DF,  p-value: <2e-16

# Ok!

Stitch Spin

• A number of different outputs

# Markdown
stitch("PATH/TO/MY/SCRIPT", system.file("misc", "knitr-template.Rmd", package = "knitr"))
knit2html("Path/To/My/Markdown.md")

# Direct 2 Html
stitch("PATH/TO/MY/SCRIPT", system.file("misc", "knitr-template.html", package = "knitr"))

# Direct to PDF Requires LaTeX
stitch("PATH/TO/MY/SCRIPT")

R Flavored Markdown

• Creating a .Rmd file allows us to use the knitr package to give us in-depth control over R HTML files
• Along the way we learn how to make great PDF files using the same toolkit
• This unified platform greatly simplifies the way R can be used in the workflow to make analyses a single coherent package from raw data to outputs

R Report Concepts

• a chunk is the technical term (no really) for a piece of R code inserted in a document
• These slides are made up of text with some markup + chunks of R code
• opts tell R how to use the chunk
• Options allow us to keep or discard the chunk, display the results, source code, or both, display a figure or graph, and how to format the output we get from R
• This is all incredibly flexible and incredibly easy to implement with a few basic parameters
• You can view the source code of each one of these tutorials online to get a feel for the format (.Rmd)

A quick example

# Start .Rmd file on next line
My Super Report on Student Testing
------------------------------------
Dr. Debateman
==============

In this report I plan to show you all the results of student testing in Myoming.

#```{r chunksetup, include=FALSE} (remove # in actual document)
load("PATH/TO/MY/DATA.rda")
source("myscript.R")
library(ggplot2)
#```

The most important thing to look at is this plot:
#```{r plot1,dev='png',fig.width=9,fig.height=6}
qplot(readSS,mathSS,data=df)
#```

And my model output can be included a few ways because it is so great.
#```{r mystatmodel,results='markup'}
mymod<-lm(readSS~mathSS+factor(grade),data=df)
summary(mymod)
#```

#```{r mystatmodel2,results='asis'}
mymod<-lm(readSS~mathSS+factor(grade),data=df)
print(xtable(summary(mymod)),type='html')
#```

And because I am awesome, I am done.

Check for Understanding

• Paste the text into a .Rmd file
• Delete the first line
• Change “PATH/TO/MY/DATA” to the path to our data
• Click “Knit HTML” in RStudio
• Or:

knit("PATH/TO/myscript.Rmd", envir = new.env())
knit2html("Path/To/Myscript.md")

Making the Report a Bit Prettier

• A frustrating thing about R reports can be that they don’t fit into our institutional reporting requirements
• The HTML framework is simple, but doesn’t give us much flexibility (we need LaTeX for total control)
• But, with a bit of CSS hacking, we can get somewhere in between by pasting this at the head of our document

<style type="text/css">
body, td {
   font-size: 14px;
}
r.code{
  font-size: 10px;
}
pre {
  font-size: 10px
}
</style>

Exporting a Single Plot

• Sometimes we don’t need to produce a whole report, but just save a particular plot and share it
• We do this with the dev argument where dev is equivalent to the graphic type we want to export

# PDF
pdf(file = "PATH/TO/MYPLOT.PDF", width = 10, heigh = 8)
print(qplot(readSS, mathSS, data = df, alpha = I(0.2)))
dev.off()
# PNG
png(file = "PATH/TO/MYPLOT.png", width = 1200, heigh = 900)
print(qplot(readSS, mathSS, data = df, alpha = I(0.2)))
dev.off()

• We can also always directly export from RStudio, which gives us a lot of options!

Exporting data

• Of course, other times we need to export raw data or results of our analyses
• R is very flexible in these cases, and depending on what you want to do, you are probably best served by Googling a specific results
• However, a few functions are indispensible for such work
• These are the foreign library, save, write.csv, and write.dta
• Note that R can also write SPSS files, SAS files, and files for just about any statistics program (sometimes even Excel, but CSV is preferred for this purpose)

Here’s an Example

• When you want to save data you can use a simple line of code

write.csv(df, file = "PATH/TO/MY.csv")
write.dta(df, file = "PATH/TO/MY.dta")
# save in the R file
save(df, file = "PATH/TO/MY.rda", compress = "xz")

Exporting Tables

• Sometimes we need to export a table that is attractive

table(df$female, df$schid)

##    
##       6  15  45  66  75 105
##   0 219 222 225 225 234 243
##   1 231 228 225 225 216 207

• This is ugly and hard to tell what it means
• We need a better way

We need XTABLE

• The xtable package provides a good way to do this output

	6	15	45	66	75	105
0	369	387	384	378	375	390
1	81	63	66	72	75	60

	schoolhigh	schoolavg	schoollow	readSS	mathSS
schoolhigh	1.00	-0.52	-0.23	-0.03	0.02
schoolavg	-0.52	1.00	-0.71	0.04	-0.07
schoollow	-0.23	-0.71	1.00	-0.02	0.06
readSS	-0.03	0.04	-0.02	1.00	0.63
mathSS	0.02	-0.07	0.06	0.63	1.00

How?

require(xtable)
print(xtable(table(df$ell, df$schid)), type = "html")

• A bit annoyingly we have to wrap our table in two functions:
• xtable reformats the table to an xtable object
• print exports the xtable object to the screen so whatever document processing system we use can grab it
• Lots of objects can be coerced into xtable objects, which helps
• xtable can be HTML or LaTeX output, giving flexibility for how we build our document

Other options

• apsrtable for beautiful looking regression tables
• Hmisc has some nice output functions as well for HTML in particular
• Many specialty packages have their own way of wrapping some mark up language around their output
• We can always export with CSV directly to Excel

Reproducibility Matters

• Point and click interfaces do not allow work to be reproduced
• Time is wasted when work cannot be repeated and produce the same results
• Not valid and reliable for decision making

Advanced Topics

• Using pandoc or slidify to create HTML5 slides in R
• Writing full blown LaTeX reports
• Embedding video and visualization
• Changes CSS styles of reports

References

1. How to Use knitr
2. CRAN Taskview: Reproducible Research
3. A Sweave Demo
4. Donald Knuth on Literate Programming

Session Info

It is good to include the session info, e.g. this document is produced with knitr version 0.8. Here is my session info:

print(sessionInfo(), locale = FALSE)

## R version 2.15.2 (2012-10-26)
## Platform: i386-w64-mingw32/i386 (32-bit)
## 
## attached base packages:
## [1] splines   grid      stats     graphics  grDevices utils     datasets 
## [8] methods   base     
## 
## other attached packages:
##  [1] snow_0.3-10      gbm_1.6-3.2      survival_2.36-14 caret_5.15-044  
##  [5] foreach_1.4.0    cluster_1.14.3   reshape_0.8.4    lme4_0.999999-0 
##  [9] Matrix_1.0-10    lattice_0.20-10  xtable_1.7-0     gridExtra_0.9.1 
## [13] sandwich_2.2-9   quantreg_4.91    SparseM_0.96     mgcv_1.7-22     
## [17] eeptools_0.1     mapproj_1.1-8.3  maps_2.2-6       proto_0.3-9.2   
## [21] stringr_0.6.1    plyr_1.7.1       ggplot2_0.9.2.1  lmtest_0.9-30   
## [25] zoo_1.7-9        knitr_0.8       
## 
## loaded via a namespace (and not attached):
##  [1] codetools_0.2-8    colorspace_1.2-0   compiler_2.15.2   
##  [4] dichromat_1.2-4    digest_0.5.2       evaluate_0.4.2    
##  [7] formatR_0.6        gtable_0.1.1       iterators_1.0.6   
## [10] labeling_0.1       markdown_0.5.3     MASS_7.3-22       
## [13] memoise_0.1        munsell_0.4        nlme_3.1-105      
## [16] RColorBrewer_1.0-5 reshape2_1.2.1     scales_0.2.2      
## [19] stats4_2.15.1      tools_2.15.1

Attribution and License

This work (R Tutorial for Education, by Jared E. Knowles), in service of the Wisconsin Department of Public Instruction, is free of known copyright restrictions.