Nudge Wiki

Concept

Heuristics Under Uncertainty (Gigerenzer's Fast-and-Frugal Rules)

heuristics decision-making expertise

Gerd Gigerenzer distinguishes two kinds of decision environment. Under risk, every possible outcome and its probability is knowable in advance (a casino roulette wheel: numbers 0-36, nothing else can happen) — here, probability theory and optimization genuinely produce the best answer. Under uncertainty, the real condition of most business and life decisions, the full set of possible outcomes isn't knowable in advance ("37s happen"), so trying to compute a mathematically optimal answer is often impossible or actively counterproductive. In genuinely uncertain conditions, a good heuristic — a simple, robust rule of thumb that doesn't pretend to find the theoretically optimal answer — frequently outperforms complex models, including machine-learning algorithms, not despite its simplicity but because of it: fewer parameters means less overfitting to noise in past data that won't repeat.

The speed-accuracy tradeoff reverses for genuine experts. For novices, speed usually costs accuracy — the standard finding in psychology labs. But for people with real domain experience, more time to deliberate often makes decisions worse, not better, because reflection disrupts well-trained intuition. Experienced firefighters given only 5 seconds to say what they'd do next in an unfolding fire scenario gave better answers than when given 1-3 minutes to think it over (Gary Klein, 1999); a 2020 study of 122 senior managers found heuristic-based strategic decisions were as accurate as lengthy analytical processes, just far faster; and a separate study titled "Strategic Decision Speed and Firm Performance" found companies that decided faster showed both greater profits and faster growth than slower-deciding competitors (see Gigerenzer Expert Fast-Decision Studies).

The fluency heuristic — go with the first option that comes to mind — explains why. Studying professional handball players shown 10 seconds of match footage, then asked what the ball-carrier should do next, researchers Johnson and Raab found the first option a player named was, on average, the best available option; the second-named option was the second-best; and the third-named option was typically twice as bad as the first. Giving players more time to study the frozen frame didn't improve their answer — it just let them talk themselves into a worse one, because the fluency heuristic only functions reliably for people with genuine domain experience, not beginners (see Gigerenzer Expert Fast-Decision Studies). A famous real-world instance of deliberately disrupting an opponent's fluency: German goalkeeper Jens Lehmann's blank cheat-sheet during Germany vs. Argentina's 2006 World Cup penalty shootout (see Lehmann Penalty-Shootout Overthinking Case). A controlled golf-putting study found the same effect directly: expert golfers' accuracy dropped 10% when given more time or asked to consciously attend to their own movements, while beginners improved under the identical conditions.

More data doesn't help, and can actively hurt, once you're in genuinely uncertain territory. Google Flu Trends, built on 50 million search terms and repeatedly made more complex (45 variables growing to 160) after early failures, still missed the 2009 swine flu outbreak and was shut down by 2015 — more data and more model complexity didn't fix a fundamentally uncertain prediction problem (see Google Flu Trends Overfitting Failure). Nobel laureate Harry Markowitz's own mean-variance portfolio optimization is the textbook-optimal way to invest across multiple assets — yet Markowitz himself invested his own retirement savings using the far simpler "1/N" heuristic (split money equally across however many assets you're considering) rather than his own Nobel-winning formula. A 2009 study testing 14 sophisticated portfolio models against plain 1/N found a 25-asset portfolio would need 250 years of historical data, and a 50-asset portfolio 500 years, before a complex model reliably beat the simple equal-split rule — far more history than any real market actually offers (see Markowitz 1-Over-N Portfolio Heuristic).

The recognition heuristic: when judging between two options, if you recognize one name and not the other, bet on the one you recognize. It works specifically because name recognition correlates with real-world prevalence or quality, and specifically works better the more "semi-ignorant" (moderately, not fully, uninformed) the judge is — genuine experts have heard of everyone, so recognition carries no useful signal for them. In a real Wimbledon tournament, semi-informed participants using pure name recognition (no data at all) out-predicted both official ATP ranking systems and expert seedings (see Serwe Frings Wimbledon Recognition-Heuristic Study).

Even Nobel-level breakthroughs lean on intuition, not pure analysis. A study of 17 Nobel laureates across physics, chemistry, medicine, and economics found most described their key discoveries as emerging from switching back and forth between rigorous analysis and simple intuition, not from analysis alone (see Nobel Laureates Intuition-Analysis Study). Einstein: "I believe in intuition. I sometimes feel right even without knowing the reason why."

Hiring under uncertainty: a single "clever cue" beats weighing many factors (Episode 231). Elon Musk hired early Tesla staff by asking one question and looking for one cue — exceptional ability at anything — reasoning that exceptional ability in any domain correlates with traits (perseverance, focus) useful in any role, and is hard to fake under follow-up questioning. Jeff Bezos used a three-step "fast-and-frugal tree" instead (exceptional ability → do I admire them → will they raise the team's average), trading more missed candidates for fewer bad hires. A 2019 field study of 236 real airline hires found a single-cue heuristic outperformed full logistic regression at predicting actual job performance, and found more experienced managers spontaneously converge on single-cue hiring more than less experienced ones (see Gigerenzer Musk-Bezos Hiring Heuristics). A related 2014 study found adding a second interviewer to a strong interviewer's hiring decision reduces accuracy rather than improving it, through simple vote-dilution math (see Gigerenzer Two-Interviewers Dilution Study).

Discussed in

  • Episode 229 — 229-should-you-trust-your-gut
  • Episode 231 — 231-elon-musk-s-controversial-interview-question (Musk/Bezos hiring heuristics; two-interviewers dilution study)

Related