Heuristics and biases in the "eyeballing" of data: the effects of context on intuitive correlation assessment.

Abstract

For a wide variety of real-world decisions, people must examine numerical tables and intuitively assess the correlations that exist among meaningful variables. The normative properties of correlation coefficients suggest that such decisions should be unaffected by perceptual factors (e.g., changes in row and column locations), semantic factors (e.g., the referents of the numbers), or certain transformations of the variables (e.g., adding a constant or multiplying by a constant). Four experiments demonstrated that judgments based on perceived correlations violate these normative properties. A general model of intuitive covariation assessment was proposed to explain the observed biases. Estimation of this model at the aggregate and individual levels suggested that no single heuristic is consistent with all of the results. Instead, the existence of several qualitatively different types of heuristics was supported. The distribution of individual-level decision rules across types of heuristics was systematically related to contextual factors.