Variable Biases: A Study of Scientists’ Interpretation of Plot Types Commonly Used in Scientific Communication
In scientific communication, there are visualization conventions that are widely used to convey uncertainty, such as representing the variability of a dataset with error bars. Yet prior research indicates that scientists frequently misinterpret error bars. In this study, we compared bar charts with error bars to four alternative visualizations: dot, box, violin, and density plots. Our goal was to determine whether these other plot types would produce fewer biases in interpretation relative to bar plots. Scientists who have experience generating and interpreting statistical graphs used plots to assess whether the difference between two datasets was statistically significant. Our results replicated the patterns of biases that have been observed in prior studies of error bar interpretation. However, we found that our participants still had the best overall performance for bar plots with error bars, because they were most familiar with this type of plot.