Cognitive Errors in Estimation: Does Anchoring Cause Overconfidence?
Date
2008
Authors
Heywood-Smith, A.
Welsh, M.
Begg, S.
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Conference paper
Citation
Proceedings of 84th SPE Annual Technical Conference and Exhibition, 21-24 September, 2008:www1-www10
Statement of Responsibility
A.B. Heywood-Smith, M.B. Welsh and S.H. Begg
Conference Name
SPE Annual Technical Conference and Exhibition (84th : 2008 : Denver, Colorado, USA)
Abstract
Tversky and Kahneman [1] described the biases known as anchoring and overconfidence, respectively, as: the tendency of people to base estimates on any number just seen; and the tendency of people to provide too narrow ranges when estimating the range that an uncertain value might fall within. They also suggested that anchoring might be causing overconfidence if people start by working out their best guess and then adjust from there to get the end points of their range – such that the best guess acts as an anchor. This has passed into the management literature as fact [2, 3] but research on this relationship is actually mixed [4, 5]. This previous research has, however, focused almost exclusively on decision makers making decisions significantly different from those that they normally would. We present results of a study of 56 oil and gas personnel, designed to demonstrate whether or not anchoring and, specifically, best guesses, cause overconfidence in people when responding to technical scenarios of the sort they might encounter in their real-life roles. These results indicate that having the expert participants give a best guess prior to estimating ranges they were 80% confident would contain the true value (sourced from open access field data) did increase overconfidence as Tversky and Kahneman predicted. Externally provided anchors, however, are shown to have potentially equally strong effects on overconfidence due to their ability to shift the entire range upward or downward. Expertise in regards to the specific parameters being assessed was not found to be related to bias susceptibility. Finally, we discuss these results, with particular focus on the differences between this study and previous research and how this might account for our different findings, and its implications for confidence interval elicitation and uncertainty representation within the industry.
School/Discipline
Dissertation Note
Provenance
Description
Copyright © 2008 Society of Petroleum Engineers