The kernel density estimator

Visualization Best Practices in R

Nick Strayer

Instructor

Where histograms struggle

Data with multiple strong peaks
Small data

Kernel density plots

Place "kernel" on top of every data point
Add up heights of all overlapping kernels

Making a KDE in ggplot

Just swap geom_histogram() for geom_density()

sample_n(md_speeding, 100) %>%
  ggplot(aes(x = percentage_over_limit)) +
  # Swap out geom_histogram() 
    geom_density(
      # Fill in curve with color
      fill = 'steelblue',
      # Standard deviation of kernel
      bw = 8 
    )

A new width to worry about

Need to adjust the standard deviation of the kernel placed on each point

Show all the data

Use geom_rug() to show all data below KDE with lines

p <-sample_n(md_speeding, 100) %>%
  ggplot(aes(x = percentage_over_limit)) +
  geom_density(
    fill = 'steelblue', # fill in curve with color
    bw = 8 # standard deviation of kernel
  ) 

p + geom_rug(alpha = 0.4)

Let's stack some gaussians!

Visualization Best Practices in R

Preparing Video For Download...