Guidelines
Suggest edit

Keep a standard scatterplot preset for correlation estimation

For relate tasks, avoid switching among tested scatterplot presets on quantitative scatterplots to prevent unsupported expectations of fidelity gains and address unnecessary preset churn in preset-selection workflows.

  • purpose:refine
  • basis:empirical
  • task:relate
  • chart:scatter
  • data:quantitative
  • quality:fidelity:use
  • lever:encoding

advice

Scatterplot preset for correlation reading

Keep a standard scatterplot preset when the job is to judge correlation. For example, the evaluated task-tuned, package-default, and prior-study scatterplots showed no significant differences in correlation accuracy or completion time.

reason

Why preset switching does not help here

When tested presets perform equivalently on the same correlation-reading task, switching among them does not buy measurable performance. In that situation, extra preset tuning adds process without improving correlation judgments.

Mechanism: If accuracy and time stay unchanged across tested scatterplot presets, a standard preset is sufficient for the correlation-reading task being performed.

Evidence: In the collated extraction, the optimizer-generated, MATLAB, R, and prior-study scatterplots tied for both correlation accuracy and correlation time, with no significant pairwise differences (Micallef et al., 2017; Zeng & Battle, 2023).

Notes: This guideline is bounded to the tested scatterplot presets and the correlation-estimation task.

context

Use when choosing among existing scatterplot presets

  • User Goal: Judge the correlation between two quantitative variables.
  • Task: Correlation estimation.
  • Data: Quantitative point data shown as a scatterplot.
  • Chart Setting: You are choosing among already available scatterplot presets rather than changing chart family.
  • Success Criterion: Correlation-reading accuracy and completion time are the deciding measures.

exceptions

Do not use when the task changes

Break it when: The task changes from correlation estimation to anomaly detection. Why: In this study, preset choice did matter for anomaly detection.

costs

Costs of keeping the standard preset

Sacrifice: You give up the possibility of an untested custom improvement. Risk: You may overgeneralize this tie and assume preset choice never matters for other scatterplot tasks. Mitigation: Apply this rule only when the decision is among similar tested presets for correlation estimation.

mistakes

Common over-tuning failure

Mistake: Retune or swap among tested scatterplot presets expecting measurable gains for correlation estimation. Why it fails: The tested presets did not differ in either correlation accuracy or completion time.

check

Check whether preset switching is worth it

Failure Sign: Teams keep changing scatterplot presets for correlation reading without measured performance gains. Quick Check: Run an A/B comparison between the current preset and a standard preset on data with known correlation answers, and record both accuracy and completion time. Stronger Test: If the presets tie on both measures, keep the standard preset.

fix

Fix unnecessary preset churn

  • Revert to a standard scatterplot preset when correlation estimation is the only task.
  • Stop switching among tested presets unless you can show a gain in correlation accuracy or completion time.
  • Reserve extra preset tuning for tasks where the comparison shows a measurable difference.

References

Micallef, L., Palmas, G., Oulasvirta, A., & Weinkauf, T. (2017). Towards Perceptual Optimization of the Visual Design of Scatterplots. IEEE Transactions on Visualization and Computer Graphics, 23(6), 1588–1599. https://doi.org/10.1109/TVCG.2017.2674978
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