Heuristics and Biases in Data-Based Decision Making: Effects of Experience, Training, and Graphical Data Displays

Abstract

Managers use numerical data as the basis for many decisions. This research investigates how data on prior advertising expenditures and sales outcomes are used in budget allocation decisions and attempts to answer three important questions about data-based inferences. First, do biases exist that are strong enough to lead to seriously suboptimal decisions? Second, do graphical data displays, real-world experience, or explicit training reduce any observed biases? Third, are the observed biases well explained by a relatively small set of natural heuristics that managers use when making data-based allocation decisions? The results suggest answers of yes, no, and yes, respectively. The authors identify three broad classes of heuristics: difference-based (which assess causation by comparing adjacent changes in expenditures to changes in sales), trend-based (which assess causation by comparing overall trends in expenditures and sales), and exemplar-based (which emulate the allocation pattern of the observations with the highest sales). All three heuristics create biases in some situations. Overall, exemplar-based heuristics were used most frequently and induced the greatest biasing of the three (sometimes allocating the most to an advertising medium that was uncorrelated with sales). Difference-based heuristics were used less frequently but generated the most extreme allocations. Trend-based heuristics were used the least.