Start free trial
Trending New Genres Authors
EnglishEnglish
EspañolSpanish
简体中文Chinese
繁體中文Chinese (Traditional)
FrançaisFrench
DeutschGerman
日本語Japanese
PortuguêsPortuguese
ItalianoItalian
한국어Korean
РусскийRussian
NederlandsDutch
العربيةArabic
PolskiPolish
हिन्दीHindi
Tiếng ViệtVietnamese
SvenskaSwedish
ΕλληνικάGreek
TürkçeTurkish
ไทยThai
ČeštinaCzech
RomânăRomanian
MagyarHungarian
УкраїнськаUkrainian
IndonesiaIndonesian
DanskDanish
SuomiFinnish
БългарскиBulgarian
עבריתHebrew
NorskNorwegian
HrvatskiCroatian
CatalàCatalan
SlovenčinaSlovak
LietuviųLithuanian
SlovenščinaSlovenian
СрпскиSerbian
EestiEstonian
LatviešuLatvian
فارسیPersian
മലയാളംMalayalam
தமிழ்Tamil
اردوUrdu
Searching...
SoBrief
Simple Heuristics That Make Us Smart

Simple Heuristics That Make Us Smart

Why a quick rule can outperform elaborate analysis, and how context determines when it will.
by Gerd Gigerenzer 1999 416 pages
3.73
178 ratings
Listen
Immersive
V2.3
Amazon Kindle Audible
Try Full Access for 3 Days
Unlock listening & more!
Continue

Key Takeaways

Judge a heuristic by its fit with the world, not its elegance

Rationality has a setting. The book challenges the idea that sound thinking must calculate every probability, consequence, and cost. Real decision makers have limited time, information, and mental capacity. They can still decide well by using a fast and frugal heuristic: a simple rule that searches selectively and makes a decision with little computation. Ecological rationality describes how well that rule fits the information structure of the environment where it is used.

An emergency-room example. A heart-attack decision tree classified patients using at most three cues, rather than the 19 measurements available. In the cited study, it was more accurate than some complex statistical methods. The lesson is not that fewer cues always win. It is that a rule should be evaluated against the actual task, including speed and accuracy, rather than an imaginary mind with unlimited resources.

Analysis

What's striking is the shift from asking whether a decision process obeys formal rules to asking whether it works in its setting. That is a useful design principle for people as well as machines: first specify the decision, the available evidence, and the cost of delay. The qualification matters. A compact rule can be excellent in a stable environment and brittle after the environment changes. Simplicity earns trust through tested fit, not through simplicity alone.

Use recognition when familiarity tracks the thing you need to infer

Recognition can carry evidence. The recognition heuristic applies when choosing between two objects and only one is recognized: infer that the recognized object has the higher value on the criterion. It works only where recognition tends to track that criterion. The book calls the proportion of correct inferences in such cases recognition validity.

Ignorance can help. American students asked which city was larger, San Diego or San Antonio, were correct 62% of the time. German students, many of whom did not recognize San Antonio, were correct every time in the reported sample. In another comparison, Chicago students averaged 71% on familiar U.S. city pairs and 71.4% on less familiar German city pairs. The less-is-more effect occurs when partial knowledge lets recognition guide choices more accurately than additional knowledge does.

Analysis

The heuristic turns a feeling of familiarity into a measurable clue, rather than treating familiarity as an explanation by itself. Its strength depends on a hidden causal chain: the environment influences what is publicized, exposure shapes recognition, and exposure must correlate with the criterion. If that chain reverses, recognition can mislead. The practical move is to ask what makes an object recognizable in this domain before trusting the signal.

Treat recognition-based investing as a test, not a promise

A striking short-run result. Researchers asked Germans and Americans which companies they recognized, then built portfolios from the most recognized names. From December 1996 to June 1997, a portfolio based on German pedestrians' recognition of German companies gained 47%, while the Dax index rose 34%. International recognition-based portfolios beat their market indices in all four reported comparisons. The rule needed no company analysis beyond recognition.

The evidence is narrow. The study covered six months during a strong bull market, and the authors explicitly raised the possibility that recognized companies were simply large firms that did well in rising markets. It does not establish a reliable investment strategy across market conditions. Its broader point is that public attention can contain information about company performance, even when the investor lacks specialist knowledge.

Analysis

This result is memorable precisely because it tempts readers to overgeneralize. A strong historical return is evidence that recognition may encode useful environmental information, not proof that it predicts future returns after costs or across market cycles. The experiment is best read as a demonstration of a mechanism. In financial settings, a plausible signal still needs out-of-sample testing, comparison with simple alternatives, and attention to the period that produced it.

Let the first strong cue decide when cues are ordered

One cue can be enough. Take The Best first uses recognition when possible. If both objects are recognized, it checks cues in order of validity, stops at the first cue that distinguishes the objects, and chooses the object favored by that cue. For example, when comparing two cities, a major-league soccer team might distinguish them before the decision maker needs to consider any other information. The rule does not add up every cue or allow later evidence to overturn the first discriminating cue.

A broad comparison. Across 20 real-world data sets, Take The Best averaged 75% correct when fitting the data, compared with 77% for multiple regression. When generalizing to new cases, it averaged 71%, compared with regression's 68%. Take The Best examined an average of 2.4 cues, while the linear methods used 7.7.

Analysis

The result challenges the intuition that every cue deserves a vote. When cue quality is uneven, a strong first cue can dominate weaker, noisier evidence, much as a well-chosen diagnostic test can simplify a larger decision. But the rule depends on having a useful cue order. If the ranking is learned from a tiny or shifting sample, the supposedly best cue may not remain best, so the simplicity of the decision rule does not remove the need to validate its inputs.

Match your decision rule to the environment's information structure

There is no universal winner. Take The Best is especially suited to noncompensatory information, where the best cue outweighs combinations of weaker cues, and to settings with scarce information. In those conditions, combining every cue can add noise rather than insight. When information is abundant, however, linear rules that combine cues can perform perfectly in some environments, while Take The Best may leave accuracy on the table.

Robustness matters. In the book's comparisons, more complex regression models often fit training data well but lost accuracy on new cases, a pattern called overfitting. Simple rules can avoid some of that fragility by ignoring details that may be accidental. The practical task is to identify whether the environment rewards a single strong signal, a combination of evidence, or another structure before selecting a method.

Analysis

The deeper contribution is a conditional theory of good judgment. Instead of ranking methods once and for all, it asks which patterns in a domain make a method succeed. This resembles a scientific model's scope conditions: every useful claim should say where it applies. The risk is that environmental structure can be described after seeing the result. Strong tests therefore specify the domain and the expected pattern in advance, then check performance on fresh cases.

Measure search as well as choice when testing a decision model

Answers do not reveal the whole process. In an experiment about choosing the most profitable company, participants searched a computerized information board. Under 50 seconds of decision time, they looked up about 33 information boxes on average; under 20 seconds, about 19. The information search became more focused on high-validity cues. Outcome analysis found that a compensatory strategy called Weighted Pros described more choices under low time pressure, while the noncompensatory LEX strategy described more under high pressure.

Identification is difficult. Several strategies often predict the same answer, especially when alternatives are chosen randomly. The researchers therefore built a selected set of problems designed to make the strategies disagree. This improved identification, but made the task less representative. Choice data and search records together offer stronger evidence than either alone.

Analysis

This is a useful methodological warning: matching a person's answer is not the same as explaining how it was reached. Process data add leverage, but they can also change the task by requiring people to reveal or record their search. The best evidence comes from convergence across methods, with tasks that preserve realistic choices while allowing competing models to make different predictions. A neat fit should be treated as a clue, not a window into the mind.

Use Bayesian calculation as a benchmark, not a picture of the mind

A demanding comparison. Bayesian models combine evidence according to probability theory and can serve as benchmarks for judging simpler methods. Across 20 data sets, the Bayesian network averaged 79% correct when fitting known cases and 74% when generalizing. Take The Best averaged 75% and 71%, respectively. The differences were small relative to the computational cost of learning and applying the Bayesian model.

The benchmark is not omniscient. The network used a search procedure to identify one promising structure rather than averaging over every possible network. It was itself an approximation. That makes it useful for comparison, but not a final, perfect standard. In practice, the right question is whether the extra accuracy of a more elaborate method justifies its greater computational and informational demands.

Analysis

The strongest role for a complex model may be diagnostic: it establishes what additional computation can buy, rather than dictating how a person must think. This preserves the value of Bayesian methods without assuming that people carry out their calculations mentally. The comparison also resists a false choice between rigor and simplicity. A simple method can be preferred when the accuracy gain is slight, but high-stakes decisions may justify greater effort.

Expect memory to revise the past when new facts arrive

Recall can be reconstruction. The RAFT model explains hindsight bias as a consequence of updating knowledge. If people cannot retrieve an earlier judgment directly, they may rebuild it using their current understanding. New information about the correct answer can change related cue beliefs, so the reconstructed judgment differs from the original one.

A food-choice test. Participants learned how saturated fat, calories, and protein predicted cholesterol, made choices, received feedback, and then recalled both their answers and cue knowledge. With feedback, confidence rose by an average of 3.4 percentage points. The model, which reapplied Take The Best to recalled cue knowledge, predicted 76.3% of individual hindsight or reversed-hindsight outcomes, compared with 67.9% for a chance model. The effect may be the cost of keeping knowledge current.

Analysis

This account makes hindsight bias more than a simple failure of storage: an updated memory can be useful for present decisions while distorting an account of the past. That distinction matters in settings where testimony or self-report is treated as a record. The evidence supports one reconstruction mechanism, not every form of hindsight error. Direct recall, motivation, and changes in strategy can also affect what people report.

Exploit skewed distributions to estimate common cases cheaply

Most cases are not average. QuickEst is an estimation heuristic designed for J-shaped distributions, where many objects have low values and a few have very high values. It orders cues by the average size of objects lacking each feature, checks them in that order, and stops at the first feature an object lacks. A catchall category handles the few objects that keep passing the checks.

German city populations. QuickEst examined 2.3 cues per estimate on average, compared with all eight cues for multiple regression. With only 10% of cities in its training sample, its mean absolute error was about 132,000 people, versus 303,000 for regression and 153,000 for an estimation tree. With complete knowledge, the tree was better. QuickEst gains its advantage from the distribution's shape, not from being universally accurate.

Analysis

The heuristic makes an important modeling choice: spend effort on the common cases and accept coarse estimates for rare extremes. That is sensible when typical decisions concern ordinary cases, but less suitable when outliers carry disproportionate consequences. Its catchall category makes the tradeoff visible rather than hiding it inside an average. Before adopting a shortcut, decision makers should ask whether errors on rare cases are cheap, or precisely the errors they cannot afford.

Categorize by eliminating options, stopping once one remains

Narrow the possibilities. Categorization by Elimination orders cues by how accurately each cue works alone. For a new object, it uses the first cue to retain compatible categories, then intersects that set with the possibilities left by the next cue. It stops when one category remains, or guesses among the remaining options if the cues run out.

Intent from motion. In an experiment, people categorized moving bugs as pursuing, evading, courting, being courted, fighting, or playing. They correctly identified 49% of two-bug trajectories without practice or feedback. CBE reached 65% on the training examples while using an average of 3.6 of seven cues. When tested on a separate set with all seven cues, its accuracy was 57%. The method offers a simple way to make a decision before every possible feature is processed.

Analysis

Elimination is especially compelling when delayed categorization carries a cost, as in rapidly interpreting another agent's behavior. It also makes errors inspectable: a wrong early cue can remove the correct category before later evidence appears. That is the central price of noncompensation. The motion experiment is a controlled simplification, not a complete model of social understanding, which also uses posture, context, facial expression, and other signals.

In mate search, learn your threshold from others' responses

Search changes when choice is mutual. A rule that works when one chooser selects passive options can fail when potential mates also choose. In the secretary problem, the 37% rule samples the first 37% of candidates, then selects the next one better than the best already seen. It maximizes the chance of selecting the single best option under that problem's assumptions. In simulations of mutual mate choice, similar rules often left most people unmatched because their standards were too high.

Feedback helps calibrate. Search rules that adjusted an individual's aspiration level using offers and rejections from others paired more people and produced better-matched pairs. Some versions also used the other person's estimated mate value to decide how much to adjust. The lesson is to account for the other side's preferences, not just rank the options one encounters.

Analysis

Mate search shows why decision rules cannot be judged apart from the behavior of other agents. A threshold is not merely a private preference when the other person can decline. The simulations use simplified populations, common rankings, and known mate values, so their results do not prescribe a human dating rule. They do make a sound conceptual point: feedback from mutual choice can reveal whether an aspiration level is realistic.

Change parental investment rules as food becomes scarce or abundant

One rule does not fit every nest. A simulation modeled western bluebird parents feeding four chicks that hatched a day apart. It compared simple feeding rules under different levels of food availability, measuring the total weight of chicks that fledged. With food availability at 29% or lower, feeding the largest chick first did best. From 30% to 69%, feeding the smallest first was most successful. At 70% to 79%, feeding the hungriest first did best. Above 80%, all tested rules performed about equally well.

The pattern is plausible, not universal proof. The authors relate these results to field observations of birds using similar strategies and to human inheritance patterns, such as equal division when land is plentiful and primogeniture when it is scarce. The simulation shows how simple rules can match changing conditions without parents calculating every future outcome.

Analysis

The striking point is environmental contingency: scarcity can make favoritism toward the most viable offspring useful, while abundance removes the need to discriminate. This cautions against interpreting birth-order patterns as fixed features of parenting. The simulation simplifies real family life and measures fledging weight, not every later outcome. It offers a mechanism-based hypothesis whose value depends on testing the relevant environmental conditions in actual populations.

Analysis

The book's central move is to redefine rationality as an achievement under constraints, not a resemblance to an ideal calculator. Its strongest evidence comes from putting explicit algorithms against real-world data, then measuring accuracy, information use, speed, and generalization separately. That matters because a model that fits observed cases may simply memorize noise, while a simple rule may preserve the strongest regularities. The repeated comparison with regression and Bayesian methods gives the argument force: simplicity is not praised as a virtue in itself, but tested as a potentially effective design.

The framework is also vulnerable to a serious methodological challenge. Ecological rationality depends on specifying which environmental structure matters, and that structure can be easier to identify after a heuristic succeeds. The book's strongest cases partly answer this by defining algorithms precisely and testing them across multiple data sets, but results remain domain-sensitive. The stock portfolio study, the bluebird simulation, and the motion-categorization task differ greatly in evidential strength and real-world reach. They should not be treated as interchangeable demonstrations.

The most useful synthesis is therefore pluralistic rather than anti-statistical. Fast rules, linear models, and Bayesian networks are competing tools with distinct costs, assumptions, and failure modes. A decision maker should ask what information is genuinely available, whether cue relations are stable, how costly delay is, and which errors matter most. A field should then test the proposed rule on new cases and record the process, not merely the answer. The adaptive toolbox is persuasive when it remains a research program for discovering and validating mechanisms, not a license to call every intuition smart after the fact.

Summary in 30 Seconds
Good decisions do not require unlimited time, information, or calculation. Simple rules can outperform complex models when they match the structure of the task: across 20 datasets, one rule averaged 71% accuracy on new cases, compared with 68% for regression, while using fewer cues. But no method wins everywhere. Test rules on fresh cases, account for the cost of delay, and track how people search for information as well as what they choose.
Contains spoilers

Last updated:

Report Issue

Review Summary

3.73 out of 5
Average of 178 ratings from Goodreads and Amazon.

The reviews for Simple Heuristics That Make Us Smart are mixed. Some readers find it insightful, praising its exploration of decision-making shortcuts and their effectiveness. Others criticize it for being one-sided, lacking depth, and misrepresenting competing theories. Critics argue it oversimplifies complex ideas and cherry-picks evidence. Some readers find it dry and unengaging, while others see it as a valuable contribution to the field of judgment and decision-making. The book's approach to heuristics and ecological rationality is both praised and contested by reviewers.

Your rating:
4.63
34 ratings
Want to read the full book?

Glossary

Ecological rationality

Fit between rule and setting

A heuristic is ecologically rational to the extent that its structure fits the information structure of the environment where it is used. The concept evaluates reasoning by its performance in real situations, including accuracy, speed, and information use, rather than by logical consistency alone.

Adaptive toolbox

Collection of specialized heuristics

The adaptive toolbox is the collection of cognitive mechanisms available for different kinds of inference and decision. Rather than relying on one universal calculating procedure, a mind can use specialized heuristics built from processes such as recognition, cue search, stopping rules, and simple decision rules.

Fast and frugal heuristics

Rules using limited search

Fast and frugal heuristics make decisions using limited time, information, and computation. They specify how to search for information, when to stop, and how to decide from what has been found. Their simplicity can support accuracy when their design fits the environment.

Recognition heuristic

Choose the recognized object

When comparing two objects on a criterion, the recognition heuristic chooses the object that is recognized if the other is not. It is useful only when recognition is correlated with the criterion in the relevant environment. Its accuracy depends on that relationship, not on recognition alone.

Less-is-more effect

Less knowledge can improve accuracy

The less-is-more effect occurs when a person or group with less knowledge makes more accurate inferences than one with more knowledge. In the book's account, partial recognition can help when recognition is a strong predictor, while additional knowledge about recognized objects is less reliable.

Take The Best

Decide from the first distinguishing cue

Take The Best compares two objects by checking cues in order of their validity. It stops at the first cue that distinguishes the objects and chooses the one favored by that cue. It does not combine all cue values, so its performance depends on the cue order and the structure of the environment.

RAFT

Reconstruct judgments after feedback

RAFT, or Reconstruction After Feedback with Take The Best, models hindsight bias when an original judgment cannot be retrieved directly. New outcome information can update related cue knowledge. Reapplying a simple inference rule to that updated knowledge can produce a reconstructed judgment different from the original one.

QuickEst

Estimate from ordered negative cues

QuickEst is a heuristic for estimating quantities in environments with J-shaped distributions, where small values are common and large values rare. It orders cues by the average criterion value of objects lacking each cue, searches until it finds a negative cue, and uses a coarse estimate associated with that cue.

Categorization by Elimination

Narrow categories cue by cue

Categorization by Elimination orders cues by their individual categorization success. It uses each cue to remove incompatible categories, retaining only possibilities consistent with the cues seen so far. Search stops when one category remains, or when no cues remain and a choice among surviving categories is required.

Satisficing

Stop at an acceptable option

Satisficing is a search method for choosing among alternatives encountered sequentially. The decision maker sets an aspiration level and stops searching when an option meets or exceeds it. This avoids needing to know every possible alternative or compute the globally optimal stopping point.

About the Author

Gerd Gigerenzer is a German psychologist renowned for his work on bounded rationality and heuristics in decision-making, particularly in medicine. He challenges the notion that heuristics lead to irrational cognitive biases, instead viewing rationality as an adaptive tool. Gigerenzer directs the "Adaptive Behavior and Cognition" department and the Harding Center for Risk Literacy at the Max Planck Institute for Human Development in Berlin. His research focuses on making rational decisions with limited time, information, and certainty. Gigerenzer's book "Gut Feelings" has been translated into 17 languages, bringing his ideas to a wider audience. He is married to Lorraine Daston and is known for his critical stance on the work of Daniel Kahneman and Amos Tversky.

Download PDF

To save this Simple Heuristics That Make Us Smart summary for later, download the free PDF. You can print it out, or read offline at your convenience.
Download PDF
File size: 0.36 MB     Pages: 19

Download EPUB

To read this Simple Heuristics That Make Us Smart summary on your e-reader device or app, download the free EPUB. The .epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.
Download EPUB
File size: 1.43 MB     Pages: 21
Want to read the full book?
Follow
Listen
Home
Swipe
Library
Get App
Try Full Access for 3 Days
Listen, bookmark, and more
Compare Features Free Pro
📖 Read Summaries
Read unlimited summaries. Free users get 3 per month
🎧 Listen to Summaries
Listen to unlimited summaries in 40 languages
—
❤️ Unlimited Bookmarks
Free users are limited to 4
—
📜 Unlimited History
Free users are limited to 4
—
📥 Unlimited Downloads
Free users are limited to 1
—
Risk-Free Timeline
Today: Get Instant Access
Listen to full summaries of 26,000+ books. That's 12,000+ hours of audio!
Day 2: Trial Reminder
We'll send you a notification that your trial is ending soon.
Day 3: Your subscription begins
You'll be charged on Nov 25,
cancel anytime before.
Consume 2.8× More Books
2.8× more books Listening Reading
Our users love us
1M+ readers
Trustpilot Rating
TrustPilot
4.6 Excellent
This site is a total game-changer. I've been flying through book summaries like never before. Highly, highly recommend.
— Dave G
Worth my money and time, and really well made. I've never seen this quality of summaries on other websites. Very helpful!
— Em
Highly recommended!! Fantastic service. Perfect for those that want a little more than a teaser but not all the intricate details of a full audio book.
— Greg M
Save 62%
Yearly
$119.88 $44.99/year/yr
$3.75/mo
Monthly
$9.99/mo
Start a 3-Day Free Trial
3 days free, then $44.99/year. Cancel anytime.
Unlock a world of fiction & nonfiction books
26,000+ books for the price of 2 books
Read any book in 10 minutes
Discover new books like Tinder
Request any book if it's not summarized
Read more books than anyone you know
#1 app for book lovers
Lifelike & immersive summaries
30-day money-back guarantee
Download summaries in EPUBs or PDFs
Cancel anytime in a few clicks

Why one summary?

Scanner
Find a barcode to scan

⭐
⭐
⭐
We have a special gift for you
Open
38% off
DISCOUNT FOR YOU
⭐
⭐
⭐
$79.99
$49.99/year
Only $4.16 per month
Continue
2 taps to start, super easy to cancel
Settings
General
Widget
Loading...