Judgments of correlation from scatterplots with contaminated distributions

This study examines the sensitivity of intuitive judgments of correlation to the characteristics of outlying data points in the scatterplot. Consistent with previous research, the results of two experiments demonstrate that people tend to underestimate the attenuating effect of outliers. The results demonstrate, however, that these judgments are extremely sensitive to the presence and characteristics of attenuating data points in the distribution. Finally, the results also demonstrate that the density of the data cloud moderates the attenuating effect of outliers. Thus, the study provides partial support for the psychophysical model of Mayer et al. (1997). It is suggested that the task demands of intuitive judgments of correlation from scatterplots with and without outliers are distinct, and that uncontaminated scatterplots call for holistic processing, whereas, contaminated scatterplots call for analytic processing of the display. Implications for future research are discussed.