Parallel Coordinate

Instead of making coordinates orthogonal, we can make them parallel to adapt more than 3 coordinates/variables. Just like line charts, parallel coordinate plots are for EDAV - Multivariate Continuous Data.

Slope Graph

When there are only two parallel coordinates, the graph is called a slope graph.

To make the patterns in a slope graph more obvious, we can apply some data transformation

• Standardization: (x - mean(x)) / sd(x)
• Rescaling (to $[0,1]$): scales::rescale()

A drawback of these transformations is that we lose the actual units. A remedy is to label the units on each axis.

More Variables

We can use ggparcoord in GGally to create parallel coordinate plots.

• scale = "std" (standardization) is the default scale
• scale = "uniminmax" is rescaling
• scale = "globalminmax" is the scale without transformation

Other methods to reveal the patterns:

• Reorder the variables

  • Making lines cross may be better than making them parallel

• Make lines transparent (change alpha)

• Interpolation

  • Instead of using straight lines, we can use splines or other interpolated curves
  • This helps separate lines sharing the same value of some variables

• Highlight a trend

  • Use colors to separate the trend and other lines

Implementations

• ggplot2::geom_line()
  • Use tidyr to transform variable names to x-axis first
• GGally::ggparcoord() (static, ggplot2)
• MASS:: parcoord() (static, base)
• parcoords::parcoords() (interactive)
  • devtools::install_github(“timelyportfolio/parcoords”)

by zcysxy