How to use this document.

This is an extremely condensed introduction to R’s base graphics and—more importantly—the powerful data-visualization package ggplot2, developed by Hadley Wickham. Run the codes shown and study the outputs to learn about these tools. When questions are posed, do your best to answer them.

For your convenience, the R codes for this document are provided in an R script which you can download, edit, and run.

Getting started: R’s base graphics

Transgenic mosquito experiment

Let’s load the data on transgenic mosquito survival time.

dat <- read.csv("http://kingaa.github.io/short-course/hadley/mosquitoes.csv")

Let’s compare the average lifespan of transgenic vs wildtype mosquitoes from this experiment. The following split the data into two subsets, one for each genetic type.

wt <- subset(dat,type=="wildtype",select=lifespan)
tg <- subset(dat,type=="transgenic",select=-type)

Let’s try and visualize the data.

plot(dat)
op <- par(mfrow=c(1,2))
hist(tg$lifespan,breaks=seq(0,55,by=5),ylim=c(0,40))
hist(wt$lifespan,breaks=seq(0,55,by=5),ylim=c(0,40))
par(op)

Question: What does the second par command accomplish?

Another way to visualize a distribution is via the empirical cumulative distribution plot.

plot(sort(dat$lifespan),seq(1,nrow(dat))/nrow(dat),type='n')
lines(sort(wt$lifespan),seq(1,nrow(wt))/nrow(wt),type='s',col='blue')
lines(sort(tg$lifespan),seq(1,nrow(tg))/nrow(tg),type='s',col='red')

Question: What does type="n" do in the first line above?

Mammal body and brain sizes

The data on mammal body and brain sizes is included in the MASS package:

library(MASS)

plot(mammals)
plot(mammals,log='x')
plot(mammals,log='xy')
plot(mammals$body,mammals$brain,log='xy')
plot(brain~body,data=mammals,log='xy')

Oil production

read.csv("http://kingaa.github.io/short-course/hadley/oil_production.csv",
         comment.char="#") -> oil
head(oil)
summary(oil)
plot(oil)
plot(Gbbl~year,data=oil,subset=region=="North.America",type='l')
lines(Gbbl~year,data=oil,subset=region=="Eurasia",type="l",col='red')
library(reshape2)

dcast(oil,year~region) -> wideOil
names(wideOil)
wideOil$total <- wideOil$Africa+wideOil$Asia+wideOil$Central+wideOil$Eurasia+wideOil$Europe+wideOil$Middle+wideOil$North.America
wideOil$total <- apply(wideOil[,-1],1,sum)
plot(wideOil$year,wideOil$total,type='l')

A systematic approach to visualization: the Grammar of Graphics

Parts of a graphic:

  1. Data
  2. Geometrical object: point, line, box, bar, density plot, contours, ribbons
  3. Statistical transformations: bins, mean, median, quantile, ECDF, identity
  4. Aesthetic attributes: x and y position, color, fill, size, shape, line type, transparency
  5. Scales: map the data onto the aesthetic attributes
  6. A coordinate system: maps x and y position onto the page
  7. A faceting system: multiple plots

You construct a graphical visualization by choosing the constituent parts. This is implemented in the ggplot2 package.

Examples

Energy production

read.csv("http://kingaa.github.io/short-course/hadley/energy_production.csv",
         comment.char="#") -> energy

library(ggplot2)

ggplot(data=energy,mapping=aes(x=year,y=TJ,color=region,linetype=source))+geom_line()
ggplot(data=energy,mapping=aes(x=year,y=TJ,color=region))+geom_line()+facet_wrap(~source)
ggplot(data=energy,mapping=aes(x=year,y=TJ,color=source))+geom_line()+facet_wrap(~region,ncol=2)

What can you conclude from the above? Try plotting these data on the log scale (scale_y_log10()). How does your interpretation change?

ggplot(data=energy,mapping=aes(x=year,y=TJ))+geom_line()
ggplot(data=energy,mapping=aes(x=year,y=TJ,group=source))+geom_line()

Question: How do you account for the appearance of the two plots immediately above?

ggplot(data=energy,mapping=aes(x=year,y=TJ,group=source:region))+geom_line()

Question: What does the group aesthetic do?

Let’s aggregate across regions by year and source of energy.

library(reshape2)

tot <- dcast(energy,year+source~'TJ',value.var="TJ",fun.aggregate=sum)
ggplot(data=tot,mapping=aes(x=year,y=TJ,color=source))+geom_line()
ggplot(data=tot,mapping=aes(x=year,y=TJ,fill=source))+geom_area()

Now let’s aggregate across years by region and source.

reg <- dcast(energy,region+source~'TJ',value.var="TJ",fun.aggregate=mean)
ggplot(data=reg,mapping=aes(x=region,y=TJ,fill=source))+
   geom_bar(stat="identity")+coord_flip()

An even better way to manipulate the data is to use the plyr package. See the data munging tutorial.

library(plyr)

ddply(energy,~region+source,summarize,TJ=mean(TJ)) -> x

ggplot(data=x,mapping=aes(x=region,y=TJ,fill=source))+
   geom_bar(stat="identity")+coord_flip()
ddply(x,~region,mutate,frac=TJ/sum(TJ)) -> y

ggplot(data=y,mapping=aes(x=region,y=frac,fill=source))+
   geom_bar(stat="identity")+coord_flip()+labs(x="fraction of production")

In the above, we first average across years for every region and source. Then, for each region, we compute the fraction of the total production due to each source. Finally, we plot the fractions using a barplot. The coord_flip coordinate specification gives us horizontal bars instead of the default vertical bars. Fancy!

Let’s compare fossil fuel production to renewable.

library(plyr)

mutate(energy,
       source=as.character(source),
       source1=mapvalues(source,
                         from=c("Hydro","Other Renewables","Coal","Oil","Gas"),
                         to=c("Renewable","Renewable","Carbon","Carbon","Carbon"))
       ) -> energy

ddply(energy,~source1+region+year,summarize,TJ=sum(TJ)) -> x

ggplot(data=x,mapping=aes(x=year,y=TJ,fill=source1))+
    geom_area()+
    facet_wrap(~region,scales="free_y",ncol=2)
ddply(energy,~source1+year,summarize,TJ=sum(TJ)) -> x

ggplot(data=x,mapping=aes(x=year,y=TJ,fill=source1))+
    geom_area()

Exercise

Ask a question regarding one of the datasets shown here and devise a visualization to answer it.

Back to course homepage

R codes for this document