Please use this identifier to cite or link to this item: http://hdl.handle.net/2440/98003
Type: Conference paper
Title: Reducing overconfidence in forecasting with repeated judgement elicitation
Author: Welsh, M.
Begg, S.
Citation: Proceedings of the 37th Annual Meeting of the Cognitive Science Society, 2015 / In Noelle, D., Dale, R., Warlaumont, A., Yoshimi, J., Matlock, T., Jennings, C., Maglio, P. (ed./s), pp.2637-2642
Publisher: Cognitive Science Society
Issue Date: 2015
ISBN: 9780991196722
Conference Name: 37th Annual Meeting of the Cognitive Science Society (23 Jul 2015 - 25 Jul 2015 : Austin, Texas)
Statement of
Responsibility: 
Matthew Brian Welsh, Steve. H. Begg
Abstract: Overconfidence is the tendency for people to underestimate the true range of uncertainty regarding unknown or future values. It results in observed outcomes falling outside people’s estimated ranges more often than their stated confidence would suggest. Previous research has, however, demonstrated various ways of reducing this bias and the More-Or-Less-Elicitation (MOLE) tool has been designed to take these into account while leading people through an elicitation. Previous research showed MOLE’s benefit on a visual estimation task but real world elicitation is more likely to involve forecasting future values. The current study compared forecast ranges, for 7 and 28 day windows, elicited via the MOLE and direct estimation. A significant reduction in overconfidence (the mismatch between stated confidence and the proportion of ranges containing the true value) was observed – from more than 25% to only 7%. We conclude that the MOLE is a useful tool for assisting forecasting.
Keywords: repeated judgement; elicitation; calibration; overconfidence; MOLE
Rights: Copyright status unknown
RMID: 0030061099
Published version: https://mindmodeling.org/cogsci2015/papers/0452/index.html
Appears in Collections:Australian School of Petroleum publications

Files in This Item:
There are no files associated with this item.


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.