More-or-less elicitation (MOLE): reducing bias in range estimation and forecasting

Abstract

Biases like overconfidence and anchoring affect values elicited from people in predictable ways—due to people’s inherent cognitive processes. The more-or-less elicitation (MOLE) process takes insights from how biases affect people’s decisions to design an elicitation process to mitigate or eliminate bias. MOLE relies on four, key insights: (1) uncertainty regarding the location of estimates means people can be unwilling to exclude values they would not specifically include; (2) repeated estimates can be averaged to produce a better, final estimate; (3) people are better at relative than absolute judgements; and, (4) consideration of multiple values prevents anchoring on a particular number. MOLE achieves these by having people repeatedly choose between options presented to them by the computerized tool rather than making estimates directly, and constructing a range logically consistent with (i.e., not ruled out by) the person’s choices in the background. Herein, MOLE is compared, across four experiments, with eight elicitation processes—all requiring direct estimation of values—and is shown to greatly reduce overconfidence in estimated ranges and to generate best guesses that are more accurate than directly estimated equivalents. This is demonstrated across three domains—in perceptual and epistemic uncertainty and in a forecasting task.