Scatterplots

We will use a dataset containing data on 1993 auto models, most of it collected by Consumers Union, and made available over the internet on the Statlib archive; paste in the source code at this link to load the data we will be using into your R session. The data frame is called "Cars93", use the attach command to allow you to access the columns by name.

Here are some examples of the basic plot features that will be useful for looking at relationships between variables.

• Plain vanilla scatterplot with title

  plot(Hpwr,Hmpg,main="Horsepower vs Miles/Gal")
  

• Labeled points, with a legend
  First, the ifelse() function will be useful: read the help file!

  ?ifelse
  ifelse(US,"gold","blue")
  #  that is really a shortcut for
  ifelse(US==1,"gold","blue")
  

  Now, make plots!

  plot(Hpwr,Hmpg,pch=19,
       col=ifelse(US,"red","darkgreen"))
  legend(locator(1),legend=c("Foreign","Domestic"),
         pch=c(19,19),col=c("red","darkgreen"))
  

  To use the locator function, move the cursor into the plot window, and click in an empty region where you would like to place the legend. To see what the locator function is doing, try using the locator function all by itself in an existing plot:

         locator(1)
  

• Labeled points, with a legend, take 2.
  Note: Often the labels will not print correctly unless you resize the graphics window! Grab the lower right corner and make the window bigger.

  plot(Hpwr,Hmpg,pch=ifelse(US,19,17))
  legend(locator(1),c("Foreign","Domestic"),pch=c(17,19))
  

• Labeled points, with a legend, take 3.
  

• Produce a plot of the numbered pch symbols; you might want to print this one, if you think you might make use of these.

  dev.off()
  plot.new()
  par(usr=c(-1,21,0,1))
  for(i in 0:20) {
  points(i, .5, pch=i); text(i, .35, i) }
  text(9, .75, 'Samples of "pch" Parameter')
  box()
  

  Now, let's add color:

  COL <- ifelse(US,"blue","red")
  PCH <- ifelse(US,9,5)
  plot(Hpwr,Hmpg,pch=PCH,col=COL)
  legend(locator(1),c("Foreign","Domestic"),pch=c(5,9),col=c("red","blue"))
  

• Use a quantitative variable as the plot character

  plot(Hpwr,Hmpg,type="n")
  text(Hpwr,Hmpg,Cyl)
  

• Use a quantitative variable to code by color. Let's plot highway vs city miles per gallon, coding by vehicle weight. The rgb color codes are numbers between zero and one in each component (red, green, blue). We must convert the weights to numbers between zero and one. Here is one way to do it:

  range(Wgt)
  w <- (Wgt-1695)/(4105-1695)
  

  We subtract the minimum, and divide by the range. The lightest car gets mapped to 0, the heaviest to 1. The plot uses the rgb function to color code the points, I have chosen the plot symbol numbered "19" (filled circle) instead of the default open circle to make the color more visible:

  plot(Cmpg,Hmpg,pch=19,col=rgb(w,0,0))
  

• Interactively identify points of interest.

  plot(Hpwr,Hmpg)
  identify(Hpwr,Hmpg)
  

  Click on a point with mouse button 1 to select it. To terminate the identify function on the Mac, hit the escape key or with a two button mouse, right click in the graphics windowy. On Unix systems, pressing mouse button 2 will terminate the function. Upon termination, identify() returns the case numbers of the selected cases. The identify function is useful for identifying interesting cases in a plot, such as outliers. It saves you the trouble of trying to read through the dataset manually to try to find a case with values matching one you see in a scatterplot.

• Identify cases by a label:

  plot(Hpwr,Hmpg)
  identify(Hpwr,Hmpg,Model)
  

  To save the case numbers returned by the identify function, just assign the output to a variable:

  plot(Hpwr,Hmpg)
  n <- identify(Hpwr,Hmpg,Model)
  n
  

• Add a straight line fit by least squares.

  plot(Hpwr,Hmpg)
  HH.lm <- lm(Hmpg ~ Hpwr)
  abline(coef(HH.lm))
  

• Add a curve which estimates the local average. The R commands below will add it to the previous plot.

  HH.lo <- loess(Hmpg~Hpwr)
  lines(HH.lo,lty=3)
  legend(locator(1),c("LSE","Lowess"),lty=c(1,3))
  

  Note that the plot function will work on transformed data too:

  plot(Hpwr,1/Hmpg)
  

  To produce what is known as a log-log plot, simply plot the logarithm of each variable:

  plot(log(Hpwr),log(Hmpg))
  

  In different scales, the relationship between the variables may appear quite different!

Example

  IceCoreCO2 <- read.csv("~/Desktop/merged_ice_core_yearly.csv",
		header=FALSE,skip=28)
head(IceCoreCO2)
names(IceCoreCO2) <- c("Year","CO2")
attach(IceCoreCO2)
plot(Year,CO2,pch=19,col="red",col.lab="green",col.axis="purple")
?loess
CO2.lo <- loess(CO2~Year,data=IceCoreCO2)
lines(CO2.lo,col="blue",lwd=2)
title("Atmospheric CO2, ppm",col.main="magenta")
  

Exercises Exercise Load the Brain/Body size dataset Brains <- read.table("http://people.reed.edu/~jones/141/Brains.dat", header=TRUE,as.is=TRUE) Make a scatterplot of log(Brain.WT) on log(Body.WT), using different plot symbols for the different classes of animal. Add a legend and title.