An Empirical Study of Counterfactual Visualization to Support Visual Causal Inference

TL;DR

This study investigates whether counterfactual visualizations can improve visual causal inference in static charts. By proposing a four-level visual causality comprehension model that maps association, intervention, and counterfactual reasoning to Recognize, Understand, Analyze, and Recall, the authors design an empirical study using line, bar, and scatter visualizations across IN, EX, CF, and REM data subsets. Results show that counterfactuals enhance understanding of interventions and causal relationships and improve recall, albeit with longer response times, while not harming basic recognition. The work yields design heuristics for incorporating counterfactuals into visualizations and identifies future opportunities to extend evaluation to uncertainty, interactivity, and broader populations. Overall, counterfactual visualizations emerge as a promising approach to advance causal reasoning in data communication.

Abstract

Counterfactuals -- expressing what might have been true under different circumstances -- have been widely applied in statistics and machine learning to help understand causal relationships. More recently, counterfactuals have begun to emerge as a technique being applied within visualization research. However, it remains unclear to what extent counterfactuals can aid with visual data communication. In this paper, we primarily focus on assessing the quality of users' understanding of data when provided with counterfactual visualizations. We propose a preliminary model of causality comprehension by connecting theories from causal inference and visual data communication. Leveraging this model, we conducted an empirical study to explore how counterfactuals can improve users' understanding of data in static visualizations. Our results indicate that visualizing counterfactuals had a positive impact on participants' interpretations of causal relations within datasets. These results motivate a discussion of how to more effectively incorporate counterfactuals into data visualizations.

Figures (9)

• Figure 1: The four types of data subsets used in our study, illustrated with the student test scores example from the introduction. (a-b) are all data points in the dataset and corresponding traditional bar chart visualization of the average test score for all students. When the students who ate launch after the test are selected as IN, (c-d) shows the subset relations and visualizations between the IN and EX subsets, and (e-f) shows the relations and visualizations between the IN, CF, and REM subsets.
• Figure 2: Framework of the proposed causality comprehension model. The left dashed box shows the causal inference theory pearl2009causalpearl2009causality, connected to cognitive objectives in visual data communication on the right.
• Figure 3: Three examples of data subset visualizations seen by participants in the user study: (a) IN subset scatterplot visualization of the Health Insurance dataset, showing only data in the IN subset. (b) EX subset line charts visualization of the Census Income dataset, showing data in both the IN and EX subsets. (c) CF subset bar charts visualization of the UCI Credit Card dataset, showing data in the IN, CF, and REM subsets.
• Figure 4: The box plots show correctness rates for each visualized subset group for tasks T1 (a) and T2 (b).
• Figure 5: The box plots show correctness rates (a) and relative impact ratios (b) of each visualized subset group for task T3.