[Paper Review] Why scatter plots suggest causality, and what we can do about it

Carl T. Bergstrom, Jevin D. West|arXiv (Cornell University)|Sep 25, 2018

Data Visualization and Analytics6 references4 citations

TL;DR

The paper proposes diamond plots—45°-rotated scatter plots with a 1:1 aspect ratio—to reduce the implicit causality bias in conventional scatter plots, where the y-axis is conventionally seen as dependent on the x-axis. By symmetrically treating both variables and visually disrupting the causal assumption, diamond plots promote recognition of purely correlational relationships, with interactive versions allowing users to toggle between standard scatter plots to reinforce symmetry.

ABSTRACT

Scatter plots carry an implicit if subtle message about causality. Whether we look at functions of one variable in pure mathematics, plots of experimental measurements as a function of the experimental conditions, or scatter plots of predictor and response variables, the value plotted on the vertical axis is by convention assumed to be determined or influenced by the value on the horizontal axis. This is a problem for the public understanding of scientific results and perhaps also for professional scientists' interpretations of scatter plots. To avoid suggesting a causal relationship between the x and y values in a scatter plot, we propose a new type of data visualization, the diamond plot. Diamond plots are essentially 45 degree rotations of ordinary scatter plots; by visually jarring the viewer they clearly indicate that she should not draw the usual distinction between independent/predictor variable and dependent/response variable. Instead, she should see the relationship as purely correlative.

Motivation & Objective

• To address the widespread misinterpretation of scatter plots as implying causation, especially in scientific communication and media reporting.
• To reduce the cognitive bias that leads viewers to assume the y-axis variable is causally influenced by the x-axis variable.
• To develop a new visualization format that treats both variables symmetrically, avoiding privileging one as independent or dependent.
• To evaluate whether diamond plots can mitigate the tendency to infer causality from correlation in data visualization.
• To explore the feasibility and usability of interactive diamond plots that allow users to switch between standard scatter plot orientations.

Proposed method

• Propose a 45° rotation of standard scatter plots to create diamond plots, visually disrupting the conventional causal framing of x (independent) and y (dependent) axes.
• Use a 1:1 aspect ratio instead of the typical 1.61:1 golden ratio to maintain symmetry and prevent visual hierarchy between axes.
• Position axis labels horizontally to improve readability and provide visual grounding despite the rotated orientation.
• Incorporate grid lines to aid interpretation due to the non-standard axis rotation.
• Design an interactive version allowing users to rotate the diamond plot clockwise or counterclockwise to view data as either standard scatter plot orientation.
• Implement a flip in the vertical axis during counterclockwise rotation to maintain consistent coordinate system alignment.

Experimental results

Research questions

• RQ1 Do traditional scatter plots unintentionally suggest causality by conventionally positioning one variable as dependent on another?
• RQ2 Can a 45°-rotated visualization (diamond plot) reduce the tendency to infer causation from correlation?
• RQ3 Does the diamond plot design effectively communicate symmetry between variables without privileging one as independent or dependent?
• RQ4 What is the cognitive cost of interpreting data in diamond plot form compared to standard scatter plots?
• RQ5 Can interactive diamond plots help users understand that no causal direction is implied in the data?

Key findings

• Traditional scatter plots implicitly suggest causality by conventionally placing the dependent variable on the y-axis and the independent variable on the x-axis, leading viewers to assume a causal relationship even when none exists.
• The diamond plot design, through 45° rotation and symmetric axis treatment, visually disrupts the causal framing and encourages viewers to interpret the relationship as purely correlational.
• Interactive diamond plots allow users to toggle between standard scatter plot orientations, reinforcing that neither variable is inherently independent or dependent.
• Initial feedback from researchers and social media users indicates positive reception of the diamond plot concept, with some users independently proposing the same 45° rotation.
• The authors acknowledge that a formal user study is needed to confirm whether diamond plots reduce causal misinterpretation and to assess any increase in cognitive load for standard data interpretation tasks.