Parallel Coordinates

Use parallel coordinates to inspect multi-variable patterns across observations, and read it as an exploratory line-based summary rather than an automatic multivariate finding.

Figure purpose

Parallel coordinates let you view several numeric dimensions at once by drawing one line per observation across multiple vertical axes. This can help you spot broad profile similarities, crossings, and unusual patterns that are harder to see in a set of separate pairwise plots.

In the figure, each row of data becomes one connected line. The line passes through each vertical axis at the value for that variable, so the output is a set of per-observation trajectories across many numeric dimensions rather than a matrix of separate scatter plots.

When to use or avoid

Use parallel coordinates when the main goal is to compare many numeric variables for the same set of observations. It is most helpful for exploratory reading, especially when you want to see whether some observations follow similar multi-axis patterns.

Avoid overreading it as a precise inferential display. The chart can become crowded quickly, and the order of axes changes what the reader notices first.

Scale comparability also matters. If variables live on very different ranges, the figure can suggest stronger or weaker contrasts than the raw data alone would imply.

Use this figure when the main question is "do some observations follow similar multi-axis profiles?" rather than "what is the exact relationship between one specific pair of variables?" If pairwise interpretation or exact point locations matter most, a simpler figure is usually safer.

Required columns

  • Multiple numeric columns that can be shown as comparable axes
  • A reading task where per-observation trajectories across variables matter

The current implementation supports a direct multivariate rendering path, but docs should avoid promising richer grouped encoding or advanced figure-specific metadata beyond that core surface.

This is currently a fixed exploratory rendering surface, not a promise of rich axis-specific tuning, automated grouping discovery, or a full multivariate diagnostics workflow on this page.

Parallel coordinates are best read as exploratory. They can help you notice pattern candidates, but they do not prove separation, latent structure, or automatic grouping.

Axis order and scale choices matter. Readers should treat those as visualization choices, not as evidence that the product discovered one definitive multivariate story.

Licklider does not automatically determine from this figure whether visually similar lines form a real cluster, whether one variable is the main driver of separation, or whether a crossing pattern is scientifically meaningful rather than a scale artifact. Those are follow-up interpretation questions, not guarantees provided by the plot itself.

Alternative figures

  • Use Heatmap when color-based matrix comparison is easier to scan than many crossing lines.
  • Use PCA Biplot when a reduced two-dimensional summary is easier to communicate.
  • Prefer smaller subsets or multiple simpler figures if the display becomes visually overloaded.

Use Heatmap when the main reading task is matrix-style comparison across many rows and columns. Use PCA Biplot when you need a compact two-dimensional summary of multivariable structure. Stay with parallel coordinates when per-observation profiles across many axes are the thing you want readers to inspect directly.

TODO (Phase02+)

  • Expand only if the public product surface later exposes clearer scale or grouping controls for this figure.