Judgment and decision-making

Judgment is the estimation of quantities and likelihoods, such as how probable an event is; decision-making is the choice between options whose outcomes are uncertain. Three accounts dominate. The heuristics-and-biases programme of Tversky and Kahneman holds that people rely on a few mental shortcuts that usually work but produce systematic, predictable errors (Tversky & Kahneman, 1974). Gigerenzer’s fast-and-frugal programme holds that simple heuristics, which deliberately ignore part of the information, are often efficient and accurate in the environments where people use them (Gigerenzer & Gaissmaier, 2011). Dual-process theory explains biases as the default output of fast intuitive processes that slower deliberation fails to correct (Evans & Stanovich, 2013). The two heuristic programmes clashed directly in 1996 over whether the “biases” are errors at all (Gigerenzer, 1996; Kahneman & Tversky, 1996). Large replication projects since 2014 show that most core findings hold in direction but often with smaller effects, and some fail (Klein et al., 2014; Klein et al., 2018). Loss aversion, the claim that losses weigh more than equal gains, is robust in aggregate but disputed in interpretation (Brown et al., 2024; Gal & Rucker, 2018).

History

Heuristics and biases

Tversky and Kahneman’s 1974 paper proposed three heuristics. Representativeness judges probability by resemblance to a typical case; availability judges frequency by how easily examples come to mind; and adjustment from an anchor starts from an initial value and adjusts it insufficiently. Each is economical and usually effective but leads to systematic and predictable errors (Tversky & Kahneman, 1974). Kahneman and Tversky later described the programme’s aim as understanding the processes behind both valid and invalid judgments, with heuristics often useful and only sometimes leading to characteristic errors (Kahneman & Tversky, 1996).

The 1996 exchange

Gigerenzer argued that the heuristics-and-biases programme fails on two counts (Gigerenzer, 1996). Its heuristics are too vague: they explain both too little and too much, because one of them can be fitted after the fact to almost any result. And its norms are too narrow: the conjunction rule of probability, for example, is applied to the Linda problem without regard to its content and to single events, where Gigerenzer holds it does not straightforwardly apply. Gigerenzer also argued that frequency formats, which present information as counts rather than probabilities, remove many biases (Collins, 2019). Kahneman and Tversky replied that Gigerenzer misrepresented their position, and that judgments of frequency are also subject to large systematic biases (Kahneman & Tversky, 1996). They pointed out that their first demonstration of the conjunction effect used frequency judgments, contradicting Gigerenzer’s claim that conjunction violations are confined to subjective probabilities (Kahneman & Tversky, 1996). Gigerenzer in turn said they had misrepresented Gigerenzer’s position, and that the real issue was building process models that predict when the illusions appear and disappear (Gigerenzer, 1996).

Core concepts

Fast-and-frugal heuristics

Gigerenzer and Gaissmaier define heuristics as efficient processes that ignore part of the available information (Gigerenzer & Gaissmaier, 2011). They describe the cognitive-illusions programme as having arisen in contrast to theories of optimisation, and argue that representativeness and availability are not testable models, since they specify neither a process nor the conditions under which a heuristic succeeds or fails (Gigerenzer & Gaissmaier, n.d.).

Anchoring

In anchoring, an arbitrary number shown before an estimate pulls the estimate toward it. Many Labs 1, a replication of 13 effects across many samples, found anchoring effects very large but varying across samples, possibly because samples knew different amounts about the quantities to be estimated (Klein et al., 2014).

Framing

A framing effect is a change in choice caused by how equivalent options are described. In risky-choice framing, the classic case being the “Asian disease” problem, the same outcomes are described either as gains or as losses. It is a replicated effect of about d = 0.5 to 0.6 (d expresses a difference in standard deviations), with size depending on the design (Kühberger, 2023).

Conjunction fallacy and base-rate neglect

The conjunction fallacy is judging a combination of two events more likely than one of them alone, as in Tversky and Kahneman’s Linda problem. Base-rate neglect is ignoring how common something is in general when judging a particular case. Presenting problems as natural frequencies rather than probabilities improves Bayesian reasoning, the correct use of base rates, but how much depends on numeracy, so the improvement is less general than often claimed (Chapman & Liu, 2009). A meta-analysis found that natural frequencies raise correct answers to about 24 percent, from about 4 percent with probabilities, leaving about three quarters of people still wrong (Weber et al., 2018).

Prospect theory and loss aversion

Prospect theory, Kahneman and Tversky’s 1979 account of choice under risk, includes loss aversion: losses loom larger than gains of the same size (Gal & Rucker, 2018). A meta-analysis of 607 estimates from 150 articles published between 1992 and 2017 gives a mean loss-aversion coefficient of 1.955 (95% interval 1.820 to 2.102): losses weigh about twice as much as equal gains (Brown et al., 2024).

Questions, controversies, and new developments

Do the classic findings replicate?

The classic findings mostly replicate in direction, often with smaller effects, and some fail.

In a probability sample of US adults, six of seven classic social-psychology and heuristics effects replicated, but with notably smaller effects than in meta-analyses of student samples (Yeager et al., 2019). The reflection effect, the reversal of risk preferences between gains and losses, was the weakest part of the prospect-theory replication: one of its contrasts replicated in only 16 percent of countries, and the combined reflection effect in 73 percent, while the other effects replicated in all 19 (Ruggeri et al., 2020).

Is loss aversion real?

Loss aversion is robust in aggregate but contested in interpretation. Gal and Rucker conclude that current evidence does not support losses being more impactful than gains on balance (Gal & Rucker, 2018). Yechiam argues that early utility studies show that very large losses are overweighted but smaller losses often are not, and that some findings were misrepresented as loss aversion (Yechiam, 2018).

Mrkva and colleagues address these critiques, together with Ert and Erev’s and Gal’s claim that loss aversion is a fallacy (Mrkva et al., 2020). Across five samples totalling 17,720 people, they found everyone loss averse, though less so with more knowledge of the domain, which they read as support for an account in which preferences are constructed at the moment of choice (Mrkva et al., 2020). They doubt that loss aversion is a fallacy or is fully explained by status quo bias, risk aversion or reliance on educated laboratory samples, and report that at least ten articles separated loss aversion from status quo bias by using choices between two lotteries (Mrkva et al., 2020). The question remains contested.

Heuristics: errors or adaptations?

Whether the biases show flawed reasoning or an inappropriate standard is the core of the Gigerenzer–Kahneman and Tversky dispute and is not settled. Natural frequencies reduce base-rate neglect, as Gigerenzer predicted, but most people still fail with them (Weber et al., 2018), and the benefit depends on numeracy (Chapman & Liu, 2009). Small incentives improve conjunction judgments only slightly (Yechiam & Zeif, 2023).

Dual-process explanations

Evans and Stanovich defend a “default-interventionist” account of judgment: fast, autonomous Type 1 processes give a default response, which Type 2 processes, defined by hypothetical thinking and working-memory load, may override (Evans & Stanovich, 2013). Melnikoff and Bargh call the typology of two process types an unsupported myth (Melnikoff & Bargh, 2018b). The dispute is covered in Dual-process theory.

Broader connections

The argument over narrow norms parallels the dispute over the standard of correct reasoning, and probability-based accounts of both draw on Bayesian models of cognition. The replication projects behind the table above are discussed in Replicability and generality.

Further reading

• Tversky and Kahneman (1974): the founding paper of the heuristics-and-biases programme.
• Gigerenzer (1996) and Kahneman and Tversky (1996): the two sides of the 1996 exchange.
• Gigerenzer and Gaissmaier (2011): a review of fast-and-frugal heuristics.
• Mrkva et al. (2020): the loss-aversion dispute and a large test of it.