How scales are shared¶
A scale maps data values to positions. Sharing one across panels is what makes them comparable: the same value is at the same place. anytrellis shares every scale by default, and lets you free one per row, per column or per panel.
What each choice compares¶
| Scales | You compare | You lose |
|---|---|---|
| Shared | Levels across panels, at a glance | Detail in panels with small values |
| Free | Each panel's shape and change | Levels: the axes must be read |
| Sliced | Each panel's change, in the same units: a rise of 5 is as tall everywhere | Levels: the axes must be read |
| Shared per row or column | Levels along a row or a column | Levels across them |
| Log, shared | Ratios and growth rates across panels, at a glance | Differences in absolute terms |
A free scale is honest when each panel draws its own axis, and anytrellis does: wherever a panel's scale differs from its neighbour's, it draws its own ticks.
Zooming follows the sharing¶
A shared scale is one scale: zooming it in one panel zooms it in all the panels that share
it. A panel with a free scale zooms alone. In a grid with share_y="row", zooming y in one
panel zooms its row. A sliced scale keeps every panel as wide: zooming one panel zooms all
of them to its width, each about its own centre, and panning moves one panel alone.
Fitting y to the data in view (y) fits each scale to the data of all its panels, so panels that share y stay comparable while it fits. A sliced y fits each panel to its own data, every one as wide as the widest.
Why log scales are often better than free ones¶
Population, income or counts often differ between groups by orders of magnitude. Free scales then make every panel look alike: each fills its own panel. A shared log scale keeps the levels comparable and shows each group's change, because on a log scale a doubling is the same distance at any size.
A log scale has no place for zero or below, so marks drawn from zero (bars, areas, histograms, densities) don't go on a log y, and values of 0 or less are left out.
Summaries follow the scale¶
hist, density and smooth are computed on the scale they are drawn on: on a log x, a
histogram's bins are equally wide on the log scale, as in ggplot2. Their bins and x values
are the same in every panel, so the summaries compare as the marks do.