Visualising many variables

• We can do more than visualise variables spatially
  • Colour
  • Size
  • Label
  • Facets

An example

Mpg data

• The variable cty measures fuel efficiency of different cars in the city, while displ measures the size of the engine.
• These are negatively correlated.
• We can also see how the non-metric variable drv interacts with these variables using the col (colour) aesthetic.

Using color

ggplot(data = mpg,mapping = 
         aes(x=displ,y=cty, col=drv))+geom_point()

Aes v geom

• Note that unlike the last lecture, color is being used here to display information about a variable in the dataset.
• Therefore instead of specifying color in the geom, it has to be specified in the aes function.
• Remember the aes function maps data to something we can perceive.

Text labels

• Another option is to plot text rather than points
  • This is in fact a different geom called geom_text
  • This was used on some of the plots demonstrating Zipf's Law
• A variable can be mapped to the actual text that appears
  • The aesthetic is label

With text

ggplot(data = mpg,mapping = 
         aes(x=displ,y=cty, label=drv))+geom_text()

The bubble chart

• To add a fourth variable we can manipulate the size of the points.
• This is known as a bubble chart.
• The aesthetic in question is size
• The following plot maps the number of cylinders to the size of points.

Bubble plot

ggplot(data = mpg,mapping = 
         aes(x=displ,y=cty, col=drv,size=cyl))+
  geom_point()

Color scales

• Suppose we are mapping metric or ordinal data to a colormap. The colormap should be
  • Sequential
  • Perceptually uniform
  • Work when printed in black and white
  • Accessible to colorblind people
  • Colorful and pretty
• The viridis colormap was developed with this in mind

Jet v Viridis

A popular palette is jet.

A better palette (by the above criteria) is viridis

Problems with jet

• Colors close to one another should be similar.
• On jet, in some parts the color changes dramatically over a small range.
• Also colorblind people (about 8% of the population) can have difficulty with the red colors in jet.
• For more on this see this talk by the creators of viridis.

Jet Colormap

Viridis colormap

In ggplot2

Ordered factors now use viridis by default.

ggplot(diamonds,aes(y=price,x=carat,col=cut))+
  geom_point(size=0.2)

Continous color

ggplot(data = mpg,mapping = 
         aes(x=displ,y=cty, col=hwy))+geom_point()

Continous color

• To use viridis for a continous variable simply add scale_color_viridis_c().
• Scale is another element of the grammar of graphics.

ggplot(data = mpg,mapping = 
         aes(x=displ,y=cty, col=hwy))+
  geom_point()+scale_color_viridis_c()

Viridis

Variations on Viridis

ggplot(data = mpg,mapping = 
         aes(x=displ,y=cty, col=hwy))+
  geom_point()+scale_color_viridis_c(option = 'C')

Caution

• There are some situations where viridis may not be ideal.
  • Nominal variables
  • Divergent scales
• Divergent scales can be used when there is a natural middle point for the data (usually zero).
• For when plotting budget or trade balances using color, red can be used to show deficit and blue can be used to show surplus.

Divergent Scale

ggplot(data = mpg,mapping = 
         aes(x=displ,y=cty, col=hwy))+
  geom_point()+scale_color_distiller(type = 'div')

Facetting

• Sometimes we cannot display everything on a single plot
• In this case facetting can be used to construct multiple plots
• For the next example we look at the txhousing dataset

Code for facetting

ggplot(data = txhousing, 
       mapping = aes(x=date, y=sales))+
  geom_line()+
  facet_wrap(~city)

Note the tilde (~) in ~city

Texas Housing

Scales

• A problem here is that due to the scaling on the y axis, only the large cities display anything interesting.
• The option scales in the facet_wrap function allows each plot to have its own scale.
• Use this with caution!

Free scales

ggplot(data = txhousing, 
       mapping = aes(x=date, y=sales))+
  geom_line()+
  facet_wrap(~city,scales = 'free_y')

Texas Housing

Change number of rows

• The number of rows or columns can be changed with the nrow or ncol arguments

ggplot(data = txhousing, 
       mapping = aes(x=date, y=sales))+
  geom_line()+
  facet_wrap(~city,scales = 'free_y',nrow = 12)

Changing number of rows

Facet grid

• We can also facet so that the rows correspond to one categorical variable and the columns to another.
• Lets try this with the diamonds dataset

ggplot(data = diamonds, 
       mapping = aes(x=carat, y=price))+
  geom_point()+
  facet_grid(rows = vars(cut), cols = vars(color))

Facet grid

Your Turn

• Plot a scatterplot with
  • Sales on the x axis
  • Median on the y axis
  • Facet by year on the rows
  • Facet by month in the columns

Solution

ggplot(data = txhousing, 
       mapping = aes(x=sales, y=median))+
  geom_point()+
  facet_grid(rows = vars(year), cols = vars(month))

Pairs plot

• A pairs plot gives an array of plots
  • On the diagonal there are kernel densities or barplots
  • On the lower diagonal are scatterplots or facetted histograms
  • On the upper diagonal are correlations or boxplots.
• This can be implemented using the ggpairs function in the GGally package.

Economics data

ggpairs(economics)

The Iris data

• The iris data is an old dataset on three species of flower with different measurements of the flower.
• The aim is to classify each flower into its species.
• However since it has a mix of metric and non-metric variables it is often used an an example for demonstration.

Iris data

ggpairs(iris)

Parallel Coordinates

• A parallel coordinates plots the variables of all values along the y axis.
• The variables themselves appear along the x axis.
• Values corresponding to the same observation are joined up by lines.
• They can often look messy but sometimes provide insight.

Parallel Coordinates

ggparcoord(iris)