Judgement under uncertainty: heuristics, biases and the argument over rationality

A settings screen arrives with an option already selected. A form arrives with a field already filled. A price list offers the expensive plan first. None of those is an argument, and each shapes what a person chooses. That literature is usually taught as a catalogue of biases, and the catalogue is the wrong shape for design. The units before it described the limits of the mind. This one asks how a person decides when the facts run out. The heuristics that shape choice are usually adaptive, and only sometimes produce the errors the field names. The strongest counter is that the laboratory effects are robust features of intuitive judgement rather than artefacts of presentation, and that they persist across framings (Kahneman and Tversky, 1996). A default or an anchor is therefore a decision a designer must defend, and not a neutral setting of the form.

A shortcut is an economy before it is an error

Tversky and Kahneman (1974) described three heuristics used under uncertainty. Representativeness judges the probability that A belongs to B by their resemblance. Availability judges how common a class is by the ease of recalling examples. Adjustment from an anchor moves too little from a starting value, so different starting points yield different estimates.

These heuristics are highly economical and usually effective, but they lead to systematic and predictable errors (Tversky and Kahneman, 1974).

Kahneman (2003) later set the same material inside a division between intuitive judgement and deliberate reasoning. He labels the two System 1 and System 2, crediting the terms to Stanovich and West (2000), and argues that a description of behaviour needs both halves. The division helps a designer because an interface can act on either: it can shape the first impression, or buy the time a second thought needs.

Gigerenzer and Goldstein (1996) built a one-reason rule that looks up cues in order and stops at the first that discriminates. In simulation it matched or beat procedures such as multiple regression on speed and accuracy. Gigerenzer and Brighton (2009) generalised the result into less-is-more effects: under uncertainty, a biased mind that ignores part of the information can be more accurate and more robust than one that uses all of it. A heuristic, on this reading, is how a decision gets made when the calculation's inputs are unavailable.

Work the rule on a comparison. Asked which of two cities is larger, a take-the-best reasoner checks one cue at a time and stops at the first cue that separates the pair. In simulation the one-reason rule matched or beat a weighted procedure on accuracy while using less information (Gigerenzer and Goldstein, 1996). Extra cues add noise faster than signal. The case shows why ignoring information can be a strategy. It does not show that the rule wins where every cue is equally valid.

The errors move with the format, and the format is a design choice

The reframing does not abolish the errors. It relocates part of them. Gigerenzer (1991) showed that overconfidence, the conjunction fallacy and base-rate neglect weaken, reverse or vanish when a problem is posed as frequencies rather than as a single-case probability.

Work the format on one medical result. Stated as a probability, a test result requires the reader to combine sensitivity, specificity and a low base rate. Stated as frequencies, so that of a thousand people a given number have the condition and of those a given number test positive, the answer moves towards the true rate. The case shows that part of the bias lives in the presentation a designer controls, and that the frequency form is the one to choose for a reader with low numeracy. It does not show that the frequency form removes every error.

Tversky and Kahneman (1981) demonstrated predictable reversals of preference when a logically equivalent problem is described in terms of gains or of losses. Kahneman and Tversky (1979) had supplied the formal alternative to expected utility. Prospect theory assigns value to gains and losses rather than to final wealth, with a value function steeper for losses than for gains, which accounts for risk aversion before a sure gain and risk seeking before a sure loss.

Kahneman and Tversky (1996) insisted that the effects are not artefacts of wording, and that they survive changes of presentation. That leaves the designer in an awkward position: the presentation is part of the decision, so it carries responsibility for the choice it produces and is not a neutral container.

A default is a decision, and its size is contested

Defaults are where the argument bites, because a default is an anchor that the designer has chosen. Johnson and Goldstein (2003) compared organ-donation consent across European countries by default policy. Their opt-in countries ranged from 4.25 to 27.5 per cent consent against 85.9 to 99.98 per cent under opt-out, nearly 60 percentage points apart. Madrian and Shea (2001) found the same inertia in retirement saving, and Thaler and Benartzi (2004) carried it over time. 78 per cent of employees offered a plan committing future pay rises joined. 80 per cent of those remained through a fourth raise. Average saving rates rose from 3.5 to 13.6 per cent over 40 months.

Those field effects are the strongest case for treating a default as a decision, and they are now contested. Maier and colleagues (2022) reanalysed the pooled nudge literature with corrections for publication bias and found it severe. The unadjusted pooled effect of 0.43 fell to 0.04 once small-study effects were modelled, with evidence against an effect in the information and assistance categories. The dispute is not about whether the organ-donation figures exist. It is about whether the average published nudge is as large as it appears.

Hu and colleagues (2025) widened the correction. Their second-order meta-analysis pooled 14 meta-analyses covering 1,638 primary studies and about 30 million participants. It found a small unadjusted effect of 0.27, which fell to 0.004 once publication bias was modelled. The authors also rated most of the pooled meta-analyses as low or critically low in quality, which limits what the correction itself can bear.

• If the corrected estimates are right, a default still changes the answer, and must be chosen on principle rather than on a promise of large gains.
• If the original estimates are right, the same rule holds with more confidence in the size.

The design rule survives either correction: choose the default you would defend if the user read it as your recommendation.

Contested question: are the biases real, and how large in the field?

The disagreement is genuine, and part of it is about method. Kahneman and Tversky (1996) hold that the laboratory effects are robust features of intuitive judgement across varied presentations. Gigerenzer (1991) and Gigerenzer and Brighton (2009) hold that many of them are properties of the task and of unrepresentative experiments rather than of the mind. A heuristic can be rational in the environment it suits. Dhami, Hertwig and Hoffrage (2004) turn the ecological side into a requirement of method. Representative design samples stimuli from the environment where the judgement is made, and much of the classic literature used stimuli chosen by the experimenter instead. The two camps differ in what they accept as evidence: the bias literature rests on controlled laboratory demonstrations, the ecological critique on representative sampling.

Stanovich and West (2000) examined individual differences across the classic tasks. They found that performance errors explain little of the gap between normative and descriptive answers. Computational limits explain non-normative answers on some tasks. People vary enough that a single bias rate describes no one in particular: a bias is a distribution, not a constant.

A settling study would need a standard that both camps accept. The dispute would narrow if the same participants met the same problem in probability and in frequency form, with stimuli drawn from a real task rather than chosen for convenience, and if results were reported with their variation between people rather than as one percentage. An effect that survived representative design and frequency framing would support the heuristics-and-biases reading, and one that disappeared under both would support the ecological reading. Until such comparisons are routine, the honest expectation is an effect that is real but smaller and more variable in the field than in the laboratory.

While the dispute is open, a designer can act on what both readings allow. Treat a default as a recommendation you would defend if the user read it as one, and present proportionate information beside it. Test the effect in your own setting rather than importing a figure from the literature, and prefer the presentation that reduces the error without hiding the information.

Where the argument is heading

The field is moving from the existence of biases to their size and their setting. The applied literature is being re-examined with the tools that humbled other famous effects, publication-bias correction first among them (Maier et al., 2022; Hu et al., 2025). The surviving claim is that a default or an anchor is a choice with a consequence (Johnson and Goldstein, 2003), and the open problem is how to measure that consequence honestly in a working interface.

The sources this unit rests on

• Dhami, M. K., Hertwig, R. and Hoffrage, U. (2004). The role of representative design in an ecological approach to cognition. Psychological Bulletin, 130(6), 959-988. doi:10.1037/0033-2909.130.6.959 Reviews representative design, in which stimuli are sampled from the environment or built to preserve its properties, and shows that the design changes research findings in judgement and decision making. It converts the ecological critique into a sampling requirement.
• Gigerenzer, G. (1991). How to make cognitive illusions disappear: beyond heuristics and biases. European Review of Social Psychology, 2(1), 83-115. doi:10.1080/14792779143000033 Shows that overconfidence, the conjunction fallacy and base-rate neglect weaken, reverse or vanish when problems are posed as frequencies rather than single-case probabilities, so part of the bias lives in the presentation an interface controls.
• Gigerenzer, G. and Brighton, H. (2009). Homo heuristicus: why biased minds make better inferences. Topics in Cognitive Science, 1(1), 107-143. doi:10.1111/j.1756-8765.2008.01006.x Reviews the study of heuristics and argues that less information, computation and time can improve accuracy, with a biased mind as the model. It states the adaptive account in its strongest form, set against the bias reading.
• Gigerenzer, G. and Goldstein, D. G. (1996). Reasoning the fast and frugal way: models of bounded rationality. Psychological Review, 103(4), 650-669. doi:10.1037/0033-295x.103.4.650 Entered a one-reason rule against procedures such as multiple regression in simulation and found it matched or beat them on speed and accuracy, an existence proof that ignoring information can be a strategy rather than a defect.
• Hu, B., Xia, Z., Guo, Q., Lu, C., Constantino, S. M. and Ju, X. (2025). Assessing nudge impact: a comprehensive second-order meta-analysis. Journal of Behavioural Decision Making, 38(5), e70053. doi:10.1002/bdm.70053 Pooled 14 meta-analyses covering 1,638 primary studies and about 30 million participants, and found a small unadjusted effect of 0.27 that fell to 0.004 once publication bias was modelled.
• Johnson, E. J. and Goldstein, D. (2003). Do defaults save lives? Science, 302(5649), 1338-1339. doi:10.1126/science.1091721 Compared effective organ-donation consent across European countries by default policy and found 4.25 to 27.5 per cent under opt-in against 85.9 to 99.98 per cent under opt-out, plus an online experiment in which consent roughly doubled when the default was to donate.
• Kahneman, D. (2003). Maps of bounded rationality: psychology for behavioural economics. American Economic Review, 93(5), 1449-1475. doi:10.1257/000282803322655392 Reviews bounded rationality and separates intuitive judgement from deliberate reasoning, arguing that behaviour needs both. It is the unit's synthesis, framing heuristics as adaptive as well as fallible.
• Kahneman, D. and Tversky, A. (1979). Prospect theory: an analysis of decision under risk. Econometrica, 47(2), 263-291. doi:10.2307/1914185 Sets out prospect theory as a descriptive alternative to expected utility, with value assigned to gains and losses and a value function steeper for losses than gains. It is the formal account behind loss framing as one of a designer's variables.
• Kahneman, D. and Tversky, A. (1996). On the reality of cognitive illusions. Psychological Review, 103(3), 582-591. doi:10.1037/0033-295x.103.3.582 Defends the biases against the ecological critique, arguing that they persist under varied presentation and are not artefacts of wording. It is the opposing position the unit presents and keeps live.
• Madrian, B. C. and Shea, D. F. (2001). The power of suggestion: inertia in 401(k) participation and savings behaviour. Quarterly Journal of Economics, 116(4), 1149-1187. doi:10.1162/003355301753265543 Found that automatic enrolment in a savings plan raised participation sharply and that the effect persisted, showing inertia in the face of a default.
• Maier, M., Bartos, F., Stanley, T. D., Shanks, D. R., Harris, A. J. L. and Wagenmakers, E. J. (2022). No evidence for nudging after adjusting for publication bias. Proceedings of the National Academy of Sciences, 119(31), e2200300119. doi:10.1073/pnas.2200300119 Reanalysed the pooled nudge literature with bias corrections and found the unadjusted effect of 0.43 falling to 0.04, with evidence against an effect in several categories. It is the reason the unit reports a disputed size.
• Stanovich, K. E. and West, R. F. (2000). Individual differences in reasoning: implications for the rationality debate? Behavioural and Brain Sciences, 23(5), 645-665. doi:10.1017/s0140525x00003435 Examined the classic heuristic and bias tasks and found performance errors a minor explanation of the normative gap, computational limits explaining some non-normative answers, and substantial variation between people.
• Thaler, R. H. and Benartzi, S. (2004). Save More Tomorrow: using behavioural economics to increase employee saving. Journal of Political Economy, 112(S1), S164-S187. doi:10.1086/380085 Tested a plan that committed future pay rises to saving: 78 per cent of those offered it joined, 80 per cent remained through a fourth raise, and average saving rates rose from 3.5 to 13.6 per cent over 40 months.
• Tversky, A. and Kahneman, D. (1974). Judgement under uncertainty: heuristics and biases. Science, 185(4157), 1124-1131. doi:10.1126/science.185.4157.1124 Described representativeness, availability and adjustment from an anchor as the heuristics used under uncertainty, calling them highly economical and usually effective while leading to systematic errors. It is the founding statement the unit reads carefully.
• Tversky, A. and Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211(4481), 453-458. doi:10.1126/science.7455683 Showed that logically equivalent descriptions of a problem produce different preferences, in choices about money and about the loss of human lives. It is the evidence that framing, which interfaces supply, changes decisions.