Confirmation Bias: The Double-Edged Sword of Data Facts in Visual Data Communication

Incorporating data facts, which are natural language descriptions of data patterns, alongside visualizations can guide readers and enhance the visibility of data patterns. However, data facts might also induce confirmation bias in visual analysis. We conducted a series of crowdsourced experiments to explore the biasing effects of data facts. Our findings show that the presentation style, strength, and alignment of data facts with pre-existing beliefs significantly impact confirmation bias. Data facts that support prior beliefs can exacerbate confirmation bias, whereas those that refute an individual’s beliefs can mitigate it. This effect is amplified when data facts are used in combination with visual annotations. Data facts describing variable correlations are perceived to be more compelling than ones describing average values and are associated with higher levels of confirmation bias. We underscore the persuasive influence of data facts in visualizations and caution against their indiscriminate use in efforts to mitigate bias.