Position along a common scale is the most accurate channel

What it is

Shared-scale positional judgment maps multiple marks through one coordinate axis and the same numerical transformation, enabling direct comparison of their locations on that axis. In the conditions studied by classic graphical-perception proportion experiments and a later crowdsourced replication, this encoding generally yielded lower estimation error than length, angle, and area. “Most accurate” names an empirical result for particular tasks and stimuli, not an optimum for every chart, reader, and analytical goal.

A common scale does not inherently require a zero baseline. Point and line positions can use a clearly disclosed truncated domain. Bar height also encodes length, so a bar detached from a meaningful zero changes what its length means. The core condition is a shared data-to-screen mapping for the marks being compared.

Why it happens

Marks on one axis provide an external reference for alignment or scale reading, avoiding reconstruction across separate baselines, areas, or angles. A common domain, transformation, direction, and unit make an equal screen displacement represent an equal numerical change. Pixel resolution, overlap, density, axis labeling, viewing conditions, and task demands still constrain performance. The advantage describes the opportunity afforded by an encoding, not immunity from cognitive and display limits.

The coordinate transformation determines what distance means. Equal spacing expresses equal differences on a linear scale and equal ratios on a logarithmic scale; time and probability transformations carry their own semantics. Sharing an inappropriate or undisclosed transform produces consistent positions that may still be misunderstood. A precise point can also imply more certainty than the data support, so intervals, distributions, or quality need to accompany the estimate.

Studying it

Cleveland and McGill's position–length experiment asked participants to estimate the smaller quantity as a percentage of the larger and compared encodings with log absolute error. Common-scale position performed well among the tested judgment types. Heer and Bostock later replicated spatial-encoding experiments through web crowdsourcing, recovering broadly similar relative patterns while showing that exact errors depend on stimulus and implementation. A replication of a new design should retain data, task, marked pairs, scale, and error definition, and report individual variation and uncertainty rather than only a rank.

Product evaluation should add the intended task. Exact reading, ordering, difference or ratio comparison, trend detection, and anomaly discovery use different outcomes. Test at deployed size, zoom, input mode, and with intended readers. For screen-reader users or others who cannot rely on visual position, evaluate whether a structured table, summary, and navigable data marks support the same conclusion instead of extrapolating a visual experiment to a nonvisual channel.

Where it stops holding

In maps and spatial tasks, position normally represents place, route, or topology and cannot be freely reassigned just because another variable needs precision. Projection changes distance, area, or direction, so a shared screen coordinate does not make every geographic quantity directly comparable. Category lookup, networks, overall shape, trend, and clustering may favor hue, connection, facets, or spatial layout over precise magnitude position.

Position is not an accessibility alternative by itself. Redundant color, shape, or labels can distinguish visible series, but nonvisual access generally needs programmatic names, values, units, intervals, and relationships plus a table or textual summary. A precise visual channel also cannot improve the evidential quality of uncertain measurements.

Applying it

• State whether the task is difference, ratio, ordering, trend, or spatial relation. Prioritize a shared x or y scale for a variable only when precise magnitude comparison is central.
• Align unit, domain, transform, direction, and pixel range for comparable marks, and disclose linear, logarithmic, or other transforms. Start bars at a meaningful zero. If a point or line scale is truncated, show the domain and break semantics without treating it as length.
• Encode appropriate intervals, distributions, or data quality alongside estimates. Expose values, units, and uncertainty programmatically, with a table or prose summary for reading that does not depend on visual position.
• Compare candidates on the real task, reporting accuracy, time, error distribution, and reader variation. When geographic position or an established spatial model cannot move, use another channel or an adjacent view for magnitude rather than sacrificing task semantics to a ranking.

Related

• Same group: U1.02.2 Losing the shared baseline sharply degrades position judgment · U1.02.3 Position carries both quantitative and categorical data and is the scarcest channel · U1.02.4 A 2D plane has only two position channels; once allocated, they are spent · U1.02.5 Position judgment depends on alignment; axis misalignment cancels the advantage
• Adjacent: U1.01.2 The ranking comes from magnitude-judgment experiments, not design intuition · U1.01.5 The ranking is an accuracy ceiling; alignment and mark size erode it
• Search terms: position along a common scale · Cleveland McGill · graphical perception