Depict means with points instead of bars

Replace the bar mark with a point mark when the chart encodes a mean. For example, show the mean as a point rather than as the endpoint of a rising or falling bar, and do not rely on added error bars to make the bar safe to interpret.

A mean is symmetric around values above and below it, but a bar is not. The filled bar behaves like a visual object that appears to contain values on one side of the mean, so readers give those within-bar values more weight than equally distant values on the other side.

Mechanism: A point mark locates the mean without creating a one-sided enclosed region that readers may treat as containing the underlying data.

Evidence: Across six experiments, viewers judged values inside a mean bar as more likely than equally distant values outside it; the effect held with and without error bars, with bars rising from below or falling from above, from memory and during free viewing, and it also changed downstream decisions (Newman & Scholl, 2012).

• User Goal: Report an average and let readers interpret what values around that average are plausible.
• Task: Read a mean correctly without biasing one side of the mean.
• Data: Quantitative values summarized by a mean.
• Chart Setting: A static chart would otherwise depict the mean as a bar from a lower or upper axis, with or without error bars.
• Audience: Readers may be students, adults, or general viewers rather than trained statistical specialists.
• Success Criterion: Readers treat equally distant values above and below the mean as equally plausible.

Break it when: The value being depicted is inherently asymmetric, such as a count, range, or measure of extremity. Why: The source recommends bar graphs for quantities whose meaning is genuinely one-sided.

Sacrifice: You depart from a ubiquitous convention for reporting averages. Risk: Applying this rule to inherently asymmetric quantities removes a form that matches the data’s one-sided meaning. Mitigation: Keep bars for counts, ranges, or measures of extremity.

Mistake: Add error bars to a mean bar and keep the bar as the main mark. Why it fails: The within-the-bar bias persisted both with and without error bars.

Failure Sign: One equally distant candidate value would fall inside the filled bar and the matching value on the other side would fall outside it. Quick Check: Ask whether flipping the same mean from a rising bar to a falling bar would change which side of the mean feels more likely. Stronger Test: Show two values equally distant above and below the mean while the chart is visible; if the within-bar value seems more likely, the encoding is misleading.

• Replace the mean bar with a point mark located at the mean.
• Remove the filled bar body when it serves only to connect the mean to an axis.
• If readers need to reason about likely values around the mean, show the distribution instead of a one-sided mean bar.