Adam Malitek¹, Minsuk Chang², Cindy Xiong Bearfield², Keisuke Fukuda^1,3; ¹University of Toronto, ²Georgia Institute of Technology, ³University of Toronto Mississauga

The accurate interpretation of statistical data is critical in the modern information landscape, yet evidence suggests that objective information evaluation is compromised by pre-existing beliefs. For instance, pre-existing belief biases individuals’ correlation estimates of a scatterplot to be more consistent with that belief (Xiong et al., 2022). Recently, we showed that gaze pattern predicted the magnitude of this belief-based bias only early on (i.e., the first 1.5 seconds) during scatterplot viewing. This suggests that there exists a limited time window (i.e., critical period) during which the visual information can interact with the pre-existing belief to form a mental representation of the scatterplot (Malitek et al., 2025). Therefore, we tested the hypothesis that belief-based bias can be mitigated by increasing the number of critical periods that individuals are exposed to. In Experiment 1 (N = 64), participants viewed eight belief-laden scatterplots (e.g., X: commute duration, Y: stress level of commuters) for five seconds each. Importantly, four scatterplots were presented three times across the experiment (repeated viewing condition) while the remaining four were shown only once (single-viewing condition). Here, we found that the belief-based bias was significantly reduced as the number of encoding opportunities increased, especially for those whose belief was comparatively weak. Critically, by keeping the total viewing time constant between the repeated and single-viewing conditions, Experiment 2 (N = 72) confirmed that this bias reduction was due to the increase in the number of critical periods rather than the total viewing time. Taken together, our results highlight the importance of the number of encoding opportunities in mitigating belief-based bias and provide insight into how we can promote bias-free interpretation of visualized data.

Acknowledgements: This research was supported by the Natural Sciences and Engineering Research Council (5009170).