Underestimating Correlation from Scatterplots

Abstract

Eighty subjects estimated the correlation coeffi cient, r, for each of 13 computer-printed scatter plots. Making judgments were 46 students in a graduate-level statistics course and 34 faculty and graduate students in a department of psychology. The actual correlation values ranged from .010 to .995, with 200 observations in each scatterplot and with the order of scatterplot presentation ran domized. As predicted, subjects underestimated the degree of actual correlation. Also as predicted, but with substantial moderation by a method-of-presen tation factor, this underestimation was most pro nounced in the middle of the correlational range—between the 0 and 1 extremes. Though per ception of correlation was shown not to be veridical (i.e., in terms of r), little support was given one al ternative view—its being in terms of r ².