Guidelines
Suggest edit

Use scatterplots instead of pie charts for anomaly detection

For anomaly detection in small static two-dimensional displays, use a scatterplot instead of a pie chart on tabular data to improve insight and accuracy and mitigate missed outliers for readers scanning for abnormal values.

  • purpose:select
  • basis:empirical
  • chart:scatter:use
  • chart:pie-donut:avoid
  • data:tabular
  • quality:insight:use
  • lever:chart-family
  • reading-mode:overview

advice

Choose scatter over pie

Use a scatterplot when the main question is whether any displayed value looks abnormal. For example, replace a pie chart with a scatterplot when readers must spot an outlying point rather than compare slices.

reason

Why scatter works better here

A scatterplot exposes individual points and makes unusual positions easier to notice. A pie chart organizes the display around slices, which makes abnormal observations harder to isolate.

Mechanism: Point positions in a scatterplot let readers scan for values that break the visible pattern, while a pie chart centers attention on wedge sizes instead of outlying cases.

Evidence: In the experiment, scatterplots ranked above pie charts for anomaly detection in accuracy and user preference, and the later review summarizes this study as recommending scatterplots for find-anomalies tasks. (Saket et al., 2019; Zeng & Battle, 2023)

Notes: The paper’s own summary recommendation for this task was to use scatterplots.

context

Use when the task is spotting abnormal values

  • User Goal: Identify whether one or more displayed values are abnormal.
  • Task: Find anomalies.
  • Data: Tabular data shown in a small display with 5-34 marks.
  • Chart Setting: Static two-dimensional chart.
  • Success Criterion: Higher anomaly-detection accuracy with a chart people prefer to use.

exceptions

Do not use when the task changes to part comparison or exact lookup

Break it when: the task changes from spotting abnormal points to reading exact values or making other non-anomaly judgments such as retrieve-value tasks. Why: the study did not show scatterplots as the strongest option for exact lookup, and other chart types performed better there.

costs

Costs of switching from pie to scatter

Sacrifice: You give up the pie chart’s slice-based presentation. Risk: A scatterplot is not automatically the best choice for every task on the same data. Mitigation: Use the scatterplot when finding anomalies is the main question, not just one possible secondary read.

mistakes

Common failure mode

Mistake: Keep a pie chart for anomaly detection because it already separates categories into slices. Why it fails: slice comparison does not make abnormal observations as visible as point positions do.

check

Check the task against the chart choice

Failure Sign: Reviewers must find an abnormal value from a pie chart. Quick Check: Render the same data as both a scatterplot and a pie chart and ask which version makes the abnormal case easier to spot. Stronger Test: Run one anomaly question on both versions and compare which chart produces fewer missed anomalies.

fix

Fix the chart choice

  • Replace the pie chart with a scatterplot when the goal is to find abnormal values.
  • Redraw the same observations as individual points so unusual positions can stand out.
  • If the pie chart must stay for another purpose, add a separate scatterplot for anomaly finding.

References

Saket, B., Endert, A., & Demiralp, Ç. (2019). Task-Based Effectiveness of Basic Visualizations. IEEE Transactions on Visualization and Computer Graphics, 25(7), 2505–2512. https://doi.org/10.1109/TVCG.2018.2829750
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