Illusion of Causality in Visualized Data.

Area of Science:

• Cognitive Psychology
• Data Visualization
• Human-Computer Interaction

Background:

• A common reasoning error is mistaking correlation for causation, exemplified by the observed link between student breakfast frequency and higher grade point averages.
• The way data is presented can influence the prevalence of this correlation-causation fallacy.
• Understanding how visualization design affects causal inference is crucial for accurate data interpretation.

Purpose of the Study:

• To investigate whether different data visualization methods can mitigate the reasoning error of inferring causation from correlation.
• To examine how data aggregation and visual encoding types influence perceived causality.

Main Methods:

• Conducted three crowdsourced experiments presenting simple data relations, such as breakfast frequency and GPA.
• Varied visualization types (text, bar graphs, scatter plots) and data aggregation levels.
• Assessed participants' ratings of correlation strength and causal interpretation.

Main Results:

• Participants readily perceived correlation but also erroneously inferred causation, especially from text descriptions and bar graphs.
• Scatter plots led to weaker causal interpretations compared to bar graphs.
• Higher data aggregation in visualizations was associated with increased perceived causality; line and dot encodings were seen as more causal than bar encodings.

Conclusions:

• Visualization design choices significantly influence the perception of causality from data.
• Specific designs, like scatter plots and less aggregated data, can help mitigate unwarranted causal inferences.
• Awareness of these effects is essential for developing more effective and less misleading data visualizations.