Faculty Talk: Sports Analytics Since Moneyball | 22 Jul 2026 | Columbia Business School

1. Executive Briefing (TL;DR)

• Ubiquitous Human Overconfidence in Quantitative Estimation: [00:00:52] Live estimation experiments reveal that ~99% of people exhibit systematic overconfidence, providing far too narrow confidence intervals (e.g., only 20% to 43% accuracy on 90% confidence intervals) when estimating unknown quantities.
• Overconfidence Risk in AI & Quantitative Models: [00:03:47] Data scientists and automated AI systems consistently demonstrate overconfidence in model output accuracy and recommendations, creating significant risk when quantitative outputs are accepted without accounting for psychological bias.