Same Data, Diverging Perspectives: The Power of Visualizations to Elicit Competing Interpretations

Cindy Xiong Bearfield, Lisanne van Weelden, Adam Waytz, Steven Franconeri

Abstract

People routinely rely on data to make decisions, but the process can be riddled with biases. We show that patterns in data might be noticed first or more strongly, depending on how the data is visually represented or what the viewer finds salient. We also demonstrate that viewer interpretation of data is similar to that of 'ambiguous figures' such that two people looking at the same data can come to different decisions. In our studies, participants read visualizations depicting competitions between two entities, where one has a historical lead (A) but the other has been gaining momentum (B) and predicted a winner, across two chart types and three annotation approaches. They either saw the historical lead as salient and predicted that A would win, or saw the increasing momentum as salient and predicted B to win. These results suggest that decisions can be influenced by both how data are presented and what patterns people find visually salient.

Introduction

In the fields of data science and analytics, visualizations are often treated as objective, neutral windows into truth. However, this research challenges that assumption, arguing that a single dataset contains multiple perceivable patterns. Which pattern a viewer notices—and subsequently acts upon—depends heavily on visual saliency. By treating visualizations as "ambiguous figures" (similar to the famous duck-rabbit illusion), we can see how design choices like chart type and annotations act as rhetorical devices that sway decision-making without altering a single underlying data point.

Experimental Design and Methodology

The researchers conducted four experiments to investigate how visualization formats influence predictive decision-making. Participants were presented with a neutral "competition" dataset involving two entities (e.g., political parties or companies) over several time points.

The neutral bar chart stimulus used in the study, showing a competition between 'Blue' and 'Green' over three years.

The data was iteratively balanced so that neither the "historical leader" (Entity A) nor the "rising underdog" (Entity B) was mathematically dominant. Entity A maintained a higher overall value but a flat or declining trend, while Entity B started lower but exhibited sharp upward momentum. The study compared different formats—tables, bar charts, and line charts—and tested storytelling techniques, specifically contrasting explicit annotations (such as trend lines or boxes) with simple highlighting (changing colors).

Key Insights: Visualizations as Multi-stable Figures

A primary finding is that visualizations are "multi-stable." In a bar chart, one viewer might focus on the absolute height of the bars (favoring the historical leader), while another focuses on the slope of the tops of the bars (favoring the rising underdog).

Examples of how annotations can nudge viewers toward different interpretations: (A) focusing on a recent win versus (B) focusing on increasing momentum.

The research identified several critical patterns in how people process these charts:

  • The Table Bias: Tables significantly biased viewers toward the historical leader. This suggests that without visual cues for trends, people default to simple value comparisons which require less cognitive effort.
  • Individual Consistency: Viewers often exhibit stable "interpretation styles," consistently prioritizing either position or direction across different chart types (e.g., Bar vs. Line charts).
  • The Power of Annotations: Explicitly drawing a trend line or a box was found to be far more effective at swaying decisions than simple color highlighting. This suggests that color alone is often insufficient to override a viewer's natural interpretation style.

Results and Conclusion

The experiments revealed a distinct bimodality in bar charts, where participants were nearly evenly split between predicting a win for the leader or the underdog. However, the introduction of annotations shifted these distributions significantly. The Earth Mover’s Distance (EMD)—a metric for quantifying shifts in prediction distributions—was nearly seven times higher for annotations (13.72) than for color highlighting (1.88).

These findings underscore the ethical responsibility of data designers. Since designers have the power to "nudge" decisions through saliency, the visualization community must develop ethical guidelines for storytelling with data. While the study used simple datasets, the implications for real-world, noisy data are significant: design choices are never truly neutral.

Summary updated

BibTeX
@article{bearfield2024same,
  author = {Cindy Xiong Bearfield and Lisanne van Weelden and Adam Waytz and Steven Franconeri},
  title = {{Same Data, Diverging Perspectives: The Power of Visualizations to Elicit Competing Interpretations}},
  journal = {IEEE TVCG},
  year = {2024},
  doi = {10.1109/tvcg.2024.3388515},
}
Figure