• Today we're going to move away from scatter plots and lines graphs, and use the ToothGrow data set to look at:
  • Histograms
  • Boxplots
  • Barplots
• Along the way, we'll learn more graphical parameters that control the appearance of our figures
  • Focus on settings that change the global layout of the plots, rather than appearance of the data itself

• The ToothGrow data set measures the effects of Vitamin C on the length of Guinea Pig teeth.
• 2 Factors:
  1. Dose Level: .5, 1, or 2 mg.
  2. Delivery Method: O.J. or vitamin supplements

head(ToothGrowth)

##    len supp dose
## 1  4.2   VC  0.5
## 2 11.5   VC  0.5
## 3  7.3   VC  0.5
## 4  5.8   VC  0.5
## 5  6.4   VC  0.5
## 6 10.0   VC  0.5

In the previous example, the bins were too wide and made the histogram appear rectangular. We can change this with the breaks argument:

hist(ToothGrowth$len, breaks = 20, main = NULL, xlab = "Tooth Length")

We'll compare how well different breaks describe the data by plotting several histograms in the same figure. We can do this by changing the mfrow (multiple figures by row) graphics setting.

The mfrow value is set by using the par() function, and is used to divide the figure window up into a grid of different regions, allowing a different plot to be drawn in each region.

When using par, remember that the changes you make affect all subsequent plots, which is not usually what you want.

One solution is to make a copy of the default par() settings before you modify them, then use that copy to reset the defaults after your plotting.

par_defaults <- par(no.readonly=TRUE) # copy the current settings
## Do some plotting stuff 
par(par_defaults) # Reset the  defaults 

If you happen to forget this, another trick is to restart your R session (Ctrl + Shift + f10) or "Restart R" under the "Session" menu.

With the warnings finished, lets change the plot window layout.

• mfrow must be a vector of length 2.
  • The first element gives the desired number of rows in the grid layout
  • The second element gives the number of columns

With par(), you modify the settings using keyname/value pairs. So, the syntax for setting the layout style with mfrow is:

par(mfrow = c(2,2)) # sets a 2x2 grid of plots

Since bar plots only plot 1 value per group, we need to summarise the ToothGrowth data before plotting.

# Don't worry about this just do it =)
library(dplyr); library(reshape2)
TG <- ToothGrowth %>% group_by(supp,dose) %>% 
  summarise(y = mean(len)) %>% acast(supp~dose) %>%
  `dimnames<-`(list("Supp"=c("O.J.","Vit. C"), "Dose"=c(.5,1,2)))

Reshaping from data.frame to matrix is necessary because barplot can only work with vectors or matrices.

In a barplot, its hard to know what x and y coordinates to use for your legend if the position keywords are not appropriate for your plot (e.g. the "topright" corner has data in it).

One trick is to use the locator() function.

1. Call locator() with no arguments, and your cursor should turn into cross-hairs.
2. Click those cross hairs on the plot window and hit escape
3. In the console, you should see the x and y position of the point you clicked on.
4. Yay!

One thing is missing from our barplots that was included in our other 2 types of figures: An estimate of variability!

We'll add errors bars equal to 1 Standard Error of the Mean (SEM) to our barplot, using the matrix made below

TGerr <- ToothGrowth %>% group_by(supp,dose) %>% 
  summarise(sd = sd(len)/sqrt(n())) %>%
  acast(supp~dose,value.var="sd")

## Warning: package 'bindrcpp' was built under R version 3.4.4

The key to plotting error bars is capturing the return value of barplot, which tells you the x-axis coordinates the bars are centered on.

x <- barplot(TG, xlab="Dosage (mg)", ylab="Tooth Length",
             legend.text=TRUE, beside=TRUE, ylim=c(0,35),
             plot=FALSE, args.legend =list(x=3,y=30,bty="n"))
print(x)

##      [,1] [,2] [,3]
## [1,]  1.5  4.5  7.5
## [2,]  2.5  5.5  8.5

Then, you can use this matrix to tell the arrows function where to plot the error bars along the x axis.

Lets imagine we want to put our histogram, box plot and barplot on the same figure.

We could use mfrow or mfcol to do that, but we have an odd number of plots, and would end up with a blank spot in the grid of plot locations.

A better choice would be to use the layout function, which lets you divide the plot window into rows and columns for multiple plots, but also lets you specify how many rows and columns each plot takes up.

• The primary argument to layout is a mat, a numeric matrix
• The dimensions (e.g. 2 by 2) determine the number of rows and column in the plot window
• The value in each cell of the matrix determine which rows and columns the next figure is drawn across

For example, passing the matrix \(\left[\begin{array}{cc}1 & 1 \\ 2 & 3\end{array}\right]\) to layout means:

1. Create a 2x2 grid layout
2. Draw the first plot made across both columns of the first row
3. Draw the second and third plots in one column each of the second row

You might notice that there is still some "dead space" in this layout (i.e. space where it is not useful) and a lack of space where we might want some to be.

To fix this, we often want to change the:

• inner margin size (the distance between the plot axis and the figure limits)
• the outer margin size (the padding between the figure limits and the window edges)

Using the warpbreaks data set:

1. Make a box plot for each combination of wool type (A and B) and tensions (Low, Medium, and High), all on the same axis.
2. Make a bar plot for each combination of wool type and tensions with errors bars (summary code below), be sure to include a legend.
3. Put these two figures in one plot window, with layout or par(mfrow).

library(dplyr); library(reshape2)
WB <- warpbreaks %>% 
  group_by(wool,tension) %>% 
  summarise(y = mean(breaks),
            se = sd(breaks)/sqrt(n()))
WB_means <- acast(WB, wool~tension,value.var="y")
WB_err <- acast(WB, wool~tension,value.var="se")