Guidelines
Suggest edit

Use a stacked bar chart instead of a stacked line chart for negative correlation judgments

For relate tasks, use a stacked bar chart on negative-correlation displays instead of a stacked line chart to improve fidelity and mitigate imprecise discrimination of nearby association strengths for readers judging correlation.

  • purpose:select
  • basis:empirical
  • task:relate
  • chart:bar:use
  • chart:line:avoid
  • quality:fidelity:use
  • lever:chart-family
  • operator:association

advice

Choose stacked bars over stacked lines

Use a stacked bar chart rather than a stacked line chart when the display shows a negative correlation and readers must judge correlation strength. For example, if the same negative relationship can be shown as stacked bars or as stacked colored lines across an ordered x axis, keep the stacked bars and avoid the stacked-line version.

reason

Why the stacked bars work better here

The stacked line and stacked bar variants were not equivalent for this task. The bar form let readers discriminate nearby negative correlations more precisely.

Mechanism: The stacked bar chart produced smaller JNDs than the stacked line chart for negative correlations, so close negative association levels were easier to separate.

Evidence: For negative correlations, stacked bar charts significantly outperformed stacked line charts in the correlation experiment, and the 2023 review preserved this as an empirical chart-selection result for relate tasks (Harrison et al., 2014; Zeng & Battle, 2023).

context

Use when the relationship is negative and both stacked variants are feasible

  • User Goal: Judge which display shows the stronger negative correlation.
  • Task: Compare nearby association strengths.
  • Data: Two quantitative variables shown as a negative-correlation pattern.
  • Chart Setting: A single static display where the same data could be shown as stacked bars or stacked lines.
  • Success Criterion: More reliable discrimination of nearby negative correlation values.

exceptions

Do not generalize this to unsupported sign directions

Break it when: The same decision is being made for positive-correlation versions of these stacked charts. Why: The study excluded positive stacked-line and positive stacked-bar conditions as unreliable, so this contrast is supported only for the negative case.

costs

What you give up by switching away from stacked line

Sacrifice: You give up the stacked-line form for this judgment task. Risk: Using this rule outside negative-correlation displays extends it beyond the tested evidence. Mitigation: Limit the swap to negative-correlation reading and test other sign directions separately.

mistakes

Common chart-family mistake

Mistake: Treating stacked line and stacked bar as interchangeable for negative correlation judgments. Why it fails: The experiment found a significant precision advantage for stacked bars.

check

How to test the choice

Failure Sign: The stacked-line version makes close negative correlations hard to tell apart. Quick Check: Render matched stacked-line and stacked-bar versions of the same negative-correlation data and compare them on close correlation values. Stronger Test: Ask reviewers to make side-by-side stronger-correlation judgments in both chart families and keep the one that yields more consistent answers.

fix

What to change

  • Replace the stacked-line view with a stacked-bar view for the same negative-correlation data.
  • Keep ordering, size, and correlation level matched during the comparison.
  • Re-test with close correlation pairs after changing the chart family.

References

Harrison, L., Yang, F., Franconeri, S., & Chang, R. (2014). Ranking Visualizations of Correlation Using Weber’s Law. IEEE Transactions on Visualization and Computer Graphics, 20(12), 1943–1952. https://doi.org/10.1109/TVCG.2014.2346979
Zeng, Z., & Battle, L. (2023). A Review and Collation of Graphical Perception Knowledge for Visualization Recommendation. Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, 1–16. https://doi.org/10.1145/3544548.3581349