If satisficing were only “settle for less,” it would be a counsel of defeat — a way to lose gracefully to the optimiser you can’t be. Gerd Gigerenzer’s research delivers the opposite and far more radical message: simple rules don’t just save effort, they can be more accurate than complex optimisation — not despite their simplicity but because of it. His program, ecological rationality, picks up Simon’s second scissor blade: a rule is smart or stupid only relative to the structure of the environment it runs in. This lesson meets the fast-and-frugal heuristics, sees why “less is more” is a real statistical phenomenon, and marks exactly where these rules shine and where they break.
Before you read — take a guess
A simple rule uses just ONE good reason to decide, ignoring all other information. A complex model weighs every available cue optimally. When predicting NEW, unseen cases, which tends to do better?
What makes a heuristic ‘fast and frugal’
A fast-and-frugal heuristic is a decision rule that is deliberately simple in three ways: it examines few cues (frugal), it decides quickly with little computation (fast), and it’s transparent — you can write it on an index card. These aren’t sloppy approximations to some ideal calculation. Gigerenzer’s claim is that they are adaptive tools, each matched to a kind of environment, that a mind reaches for because they work there.
The key idea binding them is ecological rationality: a heuristic has no fixed IQ. It is smart in environments whose structure it exploits and dumb in environments that violate its assumptions. The question is never “is this rule rational?” but “is this rule rational for this environment?” — Simon’s two blades again, the mind and the world cutting together.
Take-the-best: one good reason
Take-the-best decides between two options by the single most valid cue that distinguishes them, and ignores the rest entirely. Which German city is bigger, A or B? Check cues in order of validity — has a soccer team? is a state capital? has a cathedral? — and the moment one cue points one way, decide. Stop. Don’t integrate the others.
It sounds reckless. Yet across many real-world prediction tasks, take-the-best matches or beats multiple regression — a model that optimally weighs all the cues — when predicting new cases. Why? Because estimating the optimal weights for many cues from limited, noisy data means fitting the noise; the simple rule has fewer knobs to mis-set, so it generalises better. One good reason, robustly applied, outperforms many reasons optimally combined.
The recognition heuristic: ignorance as information
The recognition heuristic: if you recognise one option and not the other, infer the recognised one scores higher on the criterion. Which city is bigger, one you’ve heard of or one you haven’t? Bet on the familiar one. Astonishingly, this can produce a less-is-more effect: people who recognise fewer cities sometimes answer more geography questions correctly than people who recognise them all — because those who’ve heard of everything can’t use recognition to discriminate, while the partly-ignorant have a valid signal precisely because their ignorance is systematic (you hear of big cities more often).
Less-is-more, stated carefully
“Less-is-more” doesn’t mean information is bad. It means there are conditions under which using less information yields better decisions — because recognition tracks the criterion, or because ignoring cues avoids overfitting. It’s a claim about robustness: simple rules trade a little bias for a lot less variance, and in noisy, changing worlds that trade wins. It is not a licence to be ignorant on purpose about everything.
1/N: the equal-split rule
Faced with allocating money across N investment options, the optimal approach is Markowitz mean-variance optimisation — estimate every asset’s return, variance, and correlation, then compute the ideal weights. The fast-and-frugal alternative is 1/N: split your money equally across the options. It ignores all the fancy estimation.
In a well-known study, Gigerenzer and colleagues found 1/N matched or beat sophisticated optimised portfolios out of sample across most datasets. The optimiser has to estimate a mountain of parameters from limited history, and those estimates are so noisy that the “optimal” weights are often worse than just splitting evenly. With enough data and stable correlations the optimiser eventually wins — but in the messy, small-sample world investors actually inhabit, the dumb rule frequently comes out ahead.
Why simple rules win: bias, variance, and a changing world
The deep reason “less is more” happens is the bias–variance trade-off. Any prediction error has two parts:
- Bias — error from a rule being too simple to capture the true pattern.
- Variance — error from a rule being so flexible it fits the noise in the sample and swings wildly when the data changes.
Complex optimising models have low bias but high variance: with many free parameters estimated from limited data, they latch onto accidents of the sample and then fail on new cases. Fast-and-frugal heuristics accept a little more bias to buy a large cut in variance — and when data is scarce, noisy, or the world is shifting, that trade is a net win. This is the same reason a simpler model often out-predicts a complicated one in statistics and machine learning: the fancy model overfits.
| Complex optimising model | Fast-and-frugal heuristic | |
|---|---|---|
| Cues used | all of them, optimally weighted | one or a few good ones |
| Free parameters to estimate | many | very few |
| Fit to the training data | excellent | rougher |
| Prediction on new cases | can be worse (overfits noise) | often better (robust) |
| Data needed to shine | lots, stable | little, works in flux |
| Transparency | opaque | write it on a card |
What is the core statistical reason a fast-and-frugal heuristic can out-predict a complex optimised model on new data?
Ecological rationality: rules fit environments
The unifying lesson is that a heuristic is only as good as its fit to the environment. Each fast-and-frugal rule exploits a specific structure:
- Take-the-best shines when cues vary a lot in validity and are redundant (so one good cue carries most of the signal). It’s poor when many weak cues each add independent information that ought to be combined.
- The recognition heuristic works when recognition genuinely correlates with the criterion (bigger things are more talked about). It fails when what you’ve heard of is uncorrelated with — or inversely related to — what you’re judging.
- 1/N wins when returns are noisy and correlations unstable (small samples). It loses when you have abundant, stable data that makes good estimates possible.
So the same rule is brilliant here and foolish there. That’s not a bug — it’s the whole point. Ecological rationality means matching the tool to the terrain. A fast-and-frugal rule carried into an environment it doesn’t fit will misfire, sometimes badly. Which is exactly the pitfall the next lesson (and the model’s honest limits) will press on.
The limit that matters most
A heuristic that’s ecologically rational in one world misfires when the world shifts. The recognition heuristic betrays you in a domain where fame is inversely related to the criterion; take-the-best fails when cues genuinely need combining; 1/N loses when correlations are stable and estimable. Simple rules aren’t magic — they’re adaptations, and adaptations can be stranded when the environment changes. Always ask: does this rule fit the terrain I’m actually on?
Match each fast-and-frugal heuristic (or idea) to what it does.
Pick a term, then click its definition.
Sort each situation by whether a fast-and-frugal heuristic is likely to WORK WELL or MISFIRE, given the environment.
Place each item in the right group.
- Take-the-best where many weak cues each add independent information
- Recognition heuristic for 'which city is bigger?' — fame tracks size
- 1/N when you have decades of stable data and reliable correlations
- Take-the-best where one dominant cue carries most of the signal
- Recognition heuristic for 'which stock will rise?' — fame doesn't track returns
- 1/N allocation with noisy returns and only a few years of data
The one thing to remember
Fast-and-frugal heuristics — take-the-best (one good reason), the recognition heuristic (bet on the familiar), 1/N (split evenly) — are simple, transparent rules that often match or beat complex optimisation on new data, because complex models overfit noise while frugal rules stay robust (the bias–variance trade-off). But this is ecological rationality: a rule is smart only relative to the environment it exploits, and it misfires when the world shifts. Less can genuinely be more — but only when the tool fits the terrain. Next: how to set aspiration levels, choose satisfice-vs-optimise, and where the whole model lies.