Perceiving Correlations from Scatterplots

Abstract

This study investigated the effects of statistical training and various perceptual characteristics of scatterplot displays on intuitive estimates of correlations. University professors and first year undergraduate students estimated correlations from scatterplots (1) with three different levels of correlation in the data, (2) with or without a regression line, and (3) with three different types of dispersion of the data point clouds. The faculty generally used higher and a wider range of values, but both groups perceived a higher correlation when a regression line was present and both groups were equally influenced by the different types of dispersion of the point cloud. These findings indicate that the estimation of correlations from scatterplots is a perceptually based process, which is largely independent from formal statistical training.