Correlation & Regression
Figures showing relationships between two continuous variables carry a specific obligation: the distinction between correlation and causation must be disclosed. Every scatter plot with a regression line is a potential causal inference — and every causal inference must be justified by the experimental design, not implied by the figure.
Correlation disclosure is mandatory. Any figure that shows a regression line or reports a correlation coefficient triggers the Correlation vs. Causation Notation guard (3.1.4). The figure legend must specify whether the relationship is described as associative or causal, and why.
How to Choose
- Scatter plot: two continuous variables, correlation coefficient, any n
- Regression plot: scatter + fitted line with confidence band, when a linear model is the primary analysis
- Bubble chart: three-variable encoding (x, y, size), when a third variable adds meaning
- Confidence ellipse: group overlap visualization, when bivariate group distributions are the message
- Bland-Altman: method agreement only — for comparing two measurement methods, not for correlation
Figure Types
Scatter Plot
Pearson / Spearman r with CI, point transparency for overplotting, mandatory correlation disclosure.
Regression Plot
Fitted line + confidence band with R², equation, and assumption diagnostics.
Bubble Chart
Three-variable encoding with size proportional to a continuous third variable.
Confidence Ellipse
95% bivariate confidence ellipse per group for correct group distribution overlap visualization.
Bland-Altman Plot
Method agreement analysis: mean difference vs. average with limits of agreement.