Avoid full-range context when values near 50% must be read without bias

Avoid integrating a target mark with a visible 0–100% whole when readers must read a value near 50% without bias. For example, use a stand-alone bar instead of a stacked bar, or keep a dot separate from the axis rather than placing it directly on the axis, when the important values sit around 50%.

Visible whole-range context creates an implicit halfway category. Readers then remember values as farther from that category boundary than they were, which shifts values below 50% downward and values above 50% upward.

Mechanism: A visible 0–100% frame helps people estimate magnitude, but it also creates a salient 50% boundary that repels remembered values away from the midpoint.

Evidence: Across the paper’s bar and dot experiments, integrated context reduced overall unsigned error but produced a consistent signed-error pattern around 50%: values from 25–49% were underestimated and values from 51–75% were overestimated. The paper explicitly recommends avoiding stacked bars for nearly equal values around 50/50 when precision is critical (McColeman et al., 2021).

Notes: The midpoint bias was strongest in the integrated conditions.

• User Goal: Read or remember one proportion without systematic bias.
• Task: Exact value reading or reproduction, especially for values that sit on opposite sides of 50%.
• Data: Single part-to-whole values on a 0–100% scale.
• Chart Setting: The mark would otherwise be shown with a visible whole, such as a stacked bar, a reference bar, or a dot placed directly on an axis.
• Success Criterion: Minimize systematic underestimation below 50% and overestimation above 50%.

Break it when: Overall unsigned error across the full range matters more than unbiased reading around 50%, or the important values are very high proportions near 100%. Why: The added whole-range context lowered overall error, and the paper notes that it can help with very high values.

Sacrifice: You give up some of the accuracy benefit that comes from a visible 0% and 100% frame. Risk: Stand-alone values can have more overall error, especially at large magnitudes. Mitigation: Apply reduced context mainly when the key values cluster around 50% or cross that midpoint.

Mistake: Keeping a stacked bar or axis-integrated mark for a near-even comparison such as 48% versus 52%. Why it fails: The visible whole can push the lower value down and the higher value up, making the difference look larger than it is.

Failure Sign: Critical values fall in the 25–49% or 51–75% range and are shown with a visible whole. Quick Check: Make a stand-alone version of the same values and compare it with the integrated version before finalizing the chart. Stronger Test: Show sample values just below and above 50% briefly and ask people to redraw them; watch for underestimation below 50% and overestimation above 50%.

• Unstack the focal value so it is shown as a stand-alone bar rather than as a segment of a stacked bar.
• Remove full-range integration from the focal mark, such as drawing the dot away from the axis instead of directly on it.
• Keep the reduced-context version for values whose interpretation depends on a neutral read around 50%.