Align visual centroid with the true mean in bar-chart comparisons

Align the occupied bar area so the chart with the larger arithmetic mean also has the more right-shifted visual center of mass. For example, avoid a smaller-mean bar chart whose long bars are pushed toward the high-value side or toward the extrema, because those patterns made it more likely to be chosen as having the larger mean.

Quick mean judgments can follow a center-of-mass shortcut instead of an arithmetic calculation. When the smaller-mean chart visually places more bar area farther along the value axis, that shortcut can overpower the true answer.

Mechanism: Viewers can use the chart’s centroid-like impression as a proxy for mean, so a lower-mean series can look larger if its visual mass is shifted farther toward larger values.

Evidence: In the theory-driven experiment, centroid was the strongest misleading proxy on average for larger-mean judgments in paired bar charts. In the data-driven experiment, optimized deceptive charts for mean often pushed bars toward the extrema in ways consistent with centroid and hull-area motifs (Ondov et al., 2021).

Notes: The paper reports that proxy use varies across people, so centroid alignment reduces a common risk rather than guaranteeing correct judgments for every viewer.

• User Goal: Help viewers pick which of two bar-chart series has the larger mean.
• Task: Rapid visual comparison after a brief glance.
• Data: Two quantitative series shown with the same number of bars.
• Chart Setting: Side-by-side bar charts where bar arrangement is under design control.
• Audience: Viewers making quick perceptual judgments rather than computing values.
• Success Criterion: Viewers choose the chart with the larger mean reliably.

Break it when: The comparison is not a rapid larger-mean judgment between paired bar charts, or the bar arrangement is fixed by the data semantics. Why: The evidence is specific to quick mean comparison in paired bar charts, and the paper notes that proxies depend on task and data arrangement.

Sacrifice: Some bar arrangements or candidate chart pairings become unsuitable for fast mean comparison. Risk: Fixing centroid alignment alone can still leave another misleading cue, such as hull area or max bar, pointing to the wrong chart. Mitigation: Review more than one proxy-like cue before finalizing the comparison.

Mistake: Leaving a smaller-mean chart with its visual mass shifted farther toward larger values and assuming the arithmetic mean will remain obvious. Why it fails: Under brief exposure, viewers can choose the centroid-favored chart instead of the true larger-mean chart.

Failure Sign: The lower-mean chart looks more right-weighted overall than the higher-mean chart. Quick Check: Compare the apparent center of occupied bar area across the two charts; if the lower-mean chart appears farther along the value axis, flag it. Stronger Test: Run a brief forced-choice A/B review and verify that viewers still pick the chart with the larger mean.

• Reorder the bars, when order is not semantically fixed, so the lower-mean series does not place more visual mass toward larger values.
• Remove or replace candidate arrangements in which one very long bar or an extrema-heavy pattern shifts the smaller-mean chart’s centroid past the larger-mean chart.
• Compare alternative bar arrangements and keep the version in which centroid-like cues and arithmetic mean point to the same chart.