2024 | Arran Zeyu Wang, David Borland, and David Gotz
This paper explores the impact of counterfactual visualizations on users' understanding of causal relationships in datasets. Counterfactuals, which express what might have been true under different circumstances, have been increasingly used in statistics and machine learning to aid causal inference. The study proposes a preliminary model of causality comprehension, connecting causal inference theory with visual data communication. An empirical study was conducted to evaluate how counterfactuals affect users' ability to recognize, understand, analyze, and recall causal relationships in static visualizations. The results indicate that counterfactual visualizations significantly improve users' interpretations of causal relationships, enhance recall, and do not negatively impact the ability to recognize features. However, they also require longer response times. The study provides design heuristics for integrating counterfactuals into data visualizations, offering insights for researchers and designers to enhance users' comprehension of complex data. The findings support the effectiveness of counterfactuals in visual analytics and provide a framework for future research.This paper explores the impact of counterfactual visualizations on users' understanding of causal relationships in datasets. Counterfactuals, which express what might have been true under different circumstances, have been increasingly used in statistics and machine learning to aid causal inference. The study proposes a preliminary model of causality comprehension, connecting causal inference theory with visual data communication. An empirical study was conducted to evaluate how counterfactuals affect users' ability to recognize, understand, analyze, and recall causal relationships in static visualizations. The results indicate that counterfactual visualizations significantly improve users' interpretations of causal relationships, enhance recall, and do not negatively impact the ability to recognize features. However, they also require longer response times. The study provides design heuristics for integrating counterfactuals into data visualizations, offering insights for researchers and designers to enhance users' comprehension of complex data. The findings support the effectiveness of counterfactuals in visual analytics and provide a framework for future research.