William S. Cleveland

William S. Cleveland’s approach is graphical perception: encode the comparison that matters with the task readers judge most accurately. Books: The Elements of Graphing Data (1985; revised 1994) and Visualizing Data (1993).

Use it where readers must judge values accurately: ranking, small differences, relationships and distributions. For other pages, see Choose an expert.

Illustrative example with synthetic data. Ordered dots on a common scale make close comparisons easier.

Encode the comparison as position on a common scale

Readers judge these tasks from most to least accurately. Put the main comparison as high as possible.

Rank	Task	Types that use it
1	Position on a common scale	dot, scatter, line, unstacked bar
2	Position on identical, non-aligned scales	One chart repeated in a card grid
3	Length	Stacked bar segments above the first
4	Angle and slope	pie; the slope of a line
5	Area	treemap; scatter with a size measure
6	Colour	heatMap; setColoringType gradient

• Compare parts with setBarChartLayout groups (bar), so every bar starts on the common baseline.
• Keep compared measures on one axis; do not move one with changeMeasureAxis (bar, line).

Prefer dot plots when zero is not part of the comparison

A dot marks a value by position only, so its scale need not include zero. dot takes exactly two dimensions and one measure; see Visualization definitions. Use a bar when the value is an amount whose zero matters.

Check	Operation (dot)
Category labels read horizontally	setOrientation
Dots ordered by value	setSortingEnabled, setSortingDirection
One colour when the second dimension need not be told apart	setColoringType single, setColoringColor

Order categories by value

Sort unordered categories by the measure with setSortingEnabled and setSortingDirection (bar, dot). Keep the natural order for months, sizes and other ordered categories; group dates with setDimensionTemporalGrouping (Charts).

Choose the aspect ratio for slopes

The slope of a line on screen depends on the plot’s width-to-height ratio, and slopes near 45 degrees are judged most accurately. Give the chart host an explicit ratio (Keep proportions fixed inside a card) and recheck it at each reflow tier.

Chart choices

Prefer	For
Sorted dot	One measure for many categories, split by a second dimension
Sorted bar	Amounts whose zero matters
line	Change over time
scatter	Two measures against each other
histogram	Distribution of one numeric field

Avoid	Because
pie	Angle and area are judged less accurately than position
Stacked bars and setLineChartType stacked	Only the bottom series sits on the common baseline
A secondary measure axis	Two scales in one plot
treemap and heatMap for precise comparison	Area and colour rank lowest

Pairs with card grid, compare and table-first layouts, and the Economist and explainer styles.

Not available

A one-dimension dot plot, trellis panels, aspect-ratio banking, smoothers and quantile plots are not available; see App limitations. Fixed measure-axis ranges and logarithmic scales are supported by the axis controls.