Conflict simulation, peacebuilding, and development

Daily Archives: 20/10/2018

Beware the confidence heuristic

This quick tweet today by political psychologist Philip Tetlock caught my eye, since it has important implications for serious policy gaming.

As I have noted elsewhere, research on political forecasting (including Tetlock’s seminal book Expert Political Judgment (2005), as well as the work of he and his colleagues with the Good Judgment Project) has highlighted the greater efficacy of cognitive “foxes” (those not overly attached to a single paradigm) and Bayesian updaters in correctly anticipating future outcomes. By their very nature, such individuals are willing to accept new information and change their views accordingly.

By contrast, groups (including teams within wargames or other serious games) may be heavily swayed by persuasive, overly-confident rhetoric—the “confidence heuristic” referenced in the linked Bloomberg article. In many settings—especially with military participants—this dynamic may be further aggravated by the effects of hierarchy and rank. As a result, confident pronouncements by senior leaders may obscure uncertainty and drive out differing views, even if the uncertainty is important and the differing views might be correct.


Much depends on the mix of individuals and group dynamics at work during the game, then, as well as the analysis and aggregation methods used to assess game findings.

For more insight into individuals, groups, and forecasting, I strongly recommend Superforecasting: The Art and Science of Prediction (2015), a highly readable book by Tetlock and Dan Gardener. Nate Silver (of FiveThirtyEight fame) stresses the importance of Bayesian updating in The Signal and the Noise: Why So Many Predictions Fail—But Some Don’t (2015).

For a few brief thoughts of my own, see my presentations earlier this year on Wargaming and Forecasting (Dstl) and In the Eye of the Beholder? Cognitive Challenges in Wargame Analysis (Connections UK, audio available here).

%d bloggers like this: