Cognitive Biases as Sampling Limitations

Understanding Cognitive Biases Through a Sampling Limitations Framework:

A Comprehensive Review and Theoretical Integration

Abstract

This paper presents a comprehensive theoretical framework that reconceptualizes cognitive biases as systematic consequences of sampling limitations in human information processing. Building on foundational work by Kahneman and Tversky (1974), Simon's bounded rationality theory (1955), and recent advances in cognitive-ecological approaches (Fiedler, 2000), we demonstrate that virtually all known cognitive biases can be understood as arising from constraints in how humans sample, store, and process information from their environment. This sampling-based framework provides a parsimonious explanation for diverse biases including confirmation bias, availability heuristic, base-rate neglect, and anchoring effects. We review empirical evidence from experimental psychology, behavioral economics, and neuroscience that supports this unifying perspective, and discuss its implications for understanding human rationality, decision-making, and the design of debiasing interventions.

1. Introduction

The study of cognitive biases has been central to understanding human judgment and decision-making since Tversky and Kahneman's seminal work in the 1970s (Tversky & Kahneman, 1974). These systematic deviations from normative standards of reasoning have been documented across diverse domains, from medical diagnosis to financial investment, raising fundamental questions about human rationality. While numerous individual biases have been identified and studied, the field has lacked a comprehensive theoretical framework that explains their common origins.

This paper proposes that cognitive biases can be understood through a unified framework based on sampling limitations. As Fiedler (2000) notes, "Latent properties of the environment are not amenable to direct assessment but have to be inferred from empirical samples that provide the interface between cognition and the environment." The sampling process may draw on the external world or on internal memories. For systematic reasons including proximity, salience, and focus of attention, the resulting samples tend to be biased (selective, skewed, or conditional on information search strategies).

Specifically, we argue that biases emerge from systematic constraints in how humans:

• Sample information from their environment (external sampling)

• Access information from memory (internal sampling)

• Process sampled information given computational limitations

• Update beliefs based on limited samples

2. Theoretical Foundation

2.1 Bounded Rationality and Information Processing Constraints

Herbert Simon's concept of bounded rationality provides the foundational understanding that human decision-making is constrained by limited information, cognitive capacity, and time (Simon, 1955). As Simon argued, humans are not perfectly rational beings but rather operate under constraints that limit their ability to analyze all available information. These limitations necessitate the use of heuristics—mental shortcuts that simplify complex problems but can lead to systematic biases (Tversky & Kahneman, 1974).

The human brain faces fundamental computational constraints. As noted in recent neural network models of cognitive bias, "the number of computations required for exact logical or probabilistic reasoning grows exponentially with the number of facts and variables to be considered" (Hilbert, 2012). This computational complexity forces the brain to rely on sampling strategies that prioritize efficiency over accuracy.

2.2 The Sampling Framework

The sampling framework conceptualizes human cognition as a process of drawing samples from distributions of information. This perspective has several key components:

External Sampling: When making judgments about the world, humans cannot access all relevant information but must rely on available samples. These samples are systematically biased by factors such as attention allocation (what we choose to observe), environmental structure (what information is available), temporal constraints (recent events are more accessible), and social networks (who we interact with).

Internal Sampling: When retrieving information from memory, we sample from our stored experiences. This process is biased by memory accessibility (recent and vivid memories are oversampled), encoding strength (emotional events are better encoded), retrieval cues (context-dependent sampling), and forgetting curves (older information is undersampled).

Processing Limitations: Even with perfect sampling, humans face constraints in processing information. Working memory capacity limits simultaneous consideration of variables, computational resources limit complex calculations, and time pressure forces premature termination of sampling.

3. Cognitive Biases as Sampling Phenomena

3.1 Availability Heuristic and Memory Sampling

The availability heuristic, identified by Tversky and Kahneman (1973), occurs when people estimate the frequency or probability of events based on how easily examples come to mind. From a sampling perspective, this bias reflects the use of ease of retrieval as a proxy for frequency.

When individuals "sample in the mind" to estimate frequencies, they are conducting a biased sampling process. As research has shown, "sampling in the mind, however, is hardly random. It is subject to many effects that determine the ease with which examples come to mind" including salience, recency, imaginability, and actual frequency (Kahneman & Tversky, 1973). Vivid instances are more readily recalled than pallid ones, and instances that are easily imaginable are more readily brought to mind than those difficult to imagine.

3.2 Confirmation Bias and Selective Sampling

Confirmation bias—the tendency to search for, interpret, and recall information that confirms pre-existing beliefs—represents a fundamental sampling bias in information acquisition. Research on selective exposure shows that "people prefer decision-consistent to decision-inconsistent information" and this preference affects both information search and evaluation (Fischer et al., 2010).

From a sampling perspective, confirmation bias emerges through multiple mechanisms:

• Biased Search: People systematically sample information sources that are likely to confirm their beliefs

• Biased Attention: Confirmatory information receives more attention and processing resources

• Biased Memory: Confirmatory information is better encoded and more easily retrieved

• Biased Interpretation: Ambiguous information is interpreted to fit existing beliefs

3.3 Base-Rate Neglect and Sample Size Insensitivity

Base-rate neglect—the tendency to underweight prior probabilities when making judgments—can be understood as a failure to properly account for sample sizes and population frequencies. Tversky and Kahneman's famous "hospital problem" demonstrated that people fail to recognize that smaller samples show more variability than larger samples (Tversky & Kahneman, 1974).

Recent Bayesian models of base-rate neglect suggest it arises from "noisy sampling" in mental representations. As shown in recent studies, "base-rate neglect results from noise in the internal representation of variables such as the prior" (Chen et al., 2022). When the brain samples from its representation of base rates, noise in this process leads to systematic underweighting of prior information.

4. Summary of Major Cognitive Biases as Sampling Limitations

The following table summarizes how major cognitive biases can be understood through the sampling limitations framework:

5. Implications for Debiasing

Understanding biases as sampling limitations suggests specific debiasing strategies:

• Increase Sample Size: Encourage people to consider more examples before making judgments

• Diversify Sampling: Prompt consideration of information from varied sources

• Statistical Education: Teach people about sampling variability and its implications

• Environmental Design: Structure environments to make relevant information more available

• Decision Aids: Provide tools that augment human sampling capacity

6. Conclusion

The reconceptualization of cognitive biases as consequences of sampling limitations provides a unifying framework for understanding human judgment and decision-making. Rather than viewing biases as a collection of unrelated errors, we can understand them as systematic consequences of how bounded rational agents must sample from complex environments with limited resources.

This framework has several advantages: (1) Parsimony: A single mechanism (biased sampling) explains diverse phenomena; (2) Predictive Power: The framework generates testable predictions about when biases will occur; (3) Practical Applications: It suggests specific interventions to improve decision-making; (4) Theoretical Integration: It connects cognitive psychology, behavioral economics, and neuroscience.

As Simon noted, the goal is "to replace the global rationality of economic man with a kind of rational behavior that is compatible with the access to information and the computational capacities that are actually possessed by organisms" (Simon, 1955). The sampling framework provides a concrete instantiation of this bounded rationality, showing how intelligent agents can be simultaneously rational (given their constraints) and biased (relative to normative standards).

Understanding the mind as a sampling system—brilliant in its efficiency but limited in its capacity—provides a foundation for both appreciating human intelligence and improving human decision-making. In a world of ever-increasing information, recognizing and compensating for our sampling limitations becomes ever more critical.

References

Chen, J., et al. (2022). The effects of base rate neglect on sequential belief updating and real-world beliefs. PLOS Computational Biology, 18(12), e1010796.

Fiedler, K. (2000). Beware of samples! A cognitive-ecological sampling approach to judgment biases. Psychological Review, 107(4), 659-676.

Fischer, P., et al. (2010). The process of selective exposure: Why confirmatory information search weakens over time. Organizational Behavior and Human Decision Processes, 114(1), 37-48.

Hilbert, M. (2012). Toward a synthesis of cognitive biases: How noisy information processing can bias human decision making. Psychological Bulletin, 138(2), 211-237.

Hunt, L. T., et al. (2016). Approach-induced biases in human information sampling. PLOS Biology, 14(11), e2000638.

Kahneman, D., & Tversky, A. (1973). On the psychology of prediction. Psychological Review, 80(4), 237-251.

Lieder, F., et al. (2018). The anchoring bias reflects rational use of cognitive resources. Psychonomic Bulletin & Review, 25(1), 322-349.

Simon, H. A. (1955). A behavioral model of rational choice. The Quarterly Journal of Economics, 69(1), 99-118.

Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207-232.

Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131.

Westerwick, A., Johnson, B. K., & Knobloch-Westerwick, S. (2017). Confirmation biases in selective exposure to political online information: Source bias vs. content bias. Communication Monographs, 84(3), 343-364.

Human judgment systematically departs from the prescriptions of classical rationality, leading to a large catalog of cognitive biases. Traditional taxonomies organize these biases by task (e.g., memory vs. social judgment) or by heuristic (e.g., availability, representativeness, anchoring). I argue that most, and plausibly almost all, of these biases can be re-framed as consequences of sampling limitations in how the brain selects, represents, and aggregates information from the environment and from memory. This view synthesizes work on bounded rationality, resource-rational analysis, sampling-based Bayesian inference, and cognitive-ecological approaches to judgment.

I first review evidence that many cognitive computations are implemented by sampling algorithms operating under strict resource constraints. I then propose a taxonomy of cognitive biases grounded in five types of sampling limitation: (1) coverage limits (too few samples), (2) selective sampling (non-random inclusion of instances), (3) temporal sampling limits (short or distorted time windows), (4) noisy memory sampling (distortions in retrieval), and (5) social and ecological sampling distortions (biased input from one’s environment). For each class I show how canonical biases—such as availability, representativeness, base-rate neglect, confirmation bias, present bias, planning fallacy, survivorship bias, hindsight bias, false consensus, and group polarization—arise naturally as expected behavior of bounded sampling systems. Finally, I outline empirical predictions and implications for debiasing: if biases are often properties of samples rather than “flaws” in the mind, then changing what and how we sample may be more effective than trying to change abstract rules of reasoning.

1.1 Cognitive biases and existing taxonomies

Since the seminal work of Tversky and Kahneman, cognitive biases have been described as systematic deviations from normative rationality produced by simple heuristics used under uncertainty.(sites.socsci.uci.edu) Large catalogs now list more than 150 named biases, spanning judgment, decision making, memory, social cognition, and perception.(Wikipedia) Popular resources such as The Decision Lab and curated online indices further disseminate these lists.(The Decision Lab)

Existing taxonomies tend to group biases by:

• Task domain (e.g., “memory biases,” “social biases”).(Wikipedia)

• Heuristic family (e.g., availability, representativeness, anchoring).(Science)

• Functional outcome, such as loss aversion within prospect theory or attributional biases in social psychology.(Wikipedia)

These schemes are descriptively useful but often mix levels of explanation (task vs. mechanism vs. function) and do not clearly connect biases to implementation constraints of the human brain.

1.2 From “biased minds” to “biased samples”

An alternative perspective has emerged in which biases are not necessarily the result of irrational rules but often reflect the structure and sampling properties of the environment together with limited cognitive resources.

• Bounded rationality emphasises that agents with limited time, information, and computation must satisfice rather than optimise.(Cooperative Individualism)

• Fast-and-frugal heuristics can be seen as simple sampling and stopping rules tuned to ecological regularities.(MPG.PuRe)

• Cognitive-ecological sampling accounts explicitly argue that many classic “biases” arise because people draw inferences from small, skewed, or truncated samples of events and memories, not from faulty logic per se.(PubMed)

• Resource-rational analysis formalises cognition as the optimal use of limited computational resources, often via approximate, sampling-based inference.(PubMed)

On this view, cognitive systems frequently must rely on sampling rather than exhaustive enumeration of possibilities. When the samples are few, selective, or distorted, the resulting judgments will systematically diverge from classical norms.

The central thesis of this report is:

Most cognitive biases can be understood as consequences of sampling limitations—how the brain selects, weights, and aggregates a small subset of available information in space, time, and memory to guide action under resource constraints.

I develop this thesis in three steps. Section 2 reviews evidence that the brain uses sampling algorithms and that ecological sampling explains many biases. Section 3 proposes a general sampling-limitation framework. Section 4 re-organises well-known biases into new categories reflecting different sampling limitations and illustrates each category with examples. Section 5 discusses implications and limitations.

2. Sampling and bounded rationality in cognition

2.1 Bounded rationality and satisficing

Herbert Simon introduced bounded rationality to characterise real decision makers with limited computational capacity, incomplete information, and time pressure.(Cooperative Individualism) Rather than computing optimal choices over all possibilities, agents search for options and satisfice when something “good enough” is found.

This view already implies a sampling perspective: decision makers rarely see the full distribution of possible states, outcomes, or actions. Instead, they sample a subset of alternatives, cues, and episodes that are accessible given their constraints.

2.2 Sampling-based approximate inference

Modern computational cognitive science makes this link explicit. A large body of work models perception, categorisation, and decision making as approximate Bayesian inference implemented via sampling:

• Monte-Carlo and particle-filter models of categorisation show that human inferences can be captured by algorithms that maintain only a small number of samples (sometimes just one) from a posterior distribution.(PubMed)

• Sampling in human cognition (e.g., Vul, 2010) argues that humans often behave as if they draw a small set of samples from internal probability distributions and base judgments on these samples.(DSpace)

• Sequential sampling models such as the drift–diffusion model (DDM) describe speeded choices as noisy accumulation of evidence over time until a decision boundary is reached.(PMC)

In these models, biases naturally arise when the number of samples is small, when sampling is biased toward certain hypotheses or time windows, or when stopping rules are asymmetric.

2.3 Cognitive-ecological sampling accounts

Fiedler’s “cognitive-ecological” approach argues that many judgment biases are better explained by properties of the external and internal information samples than by internal flaws.(PubMed) People infer latent properties (e.g., risk, prevalence, trustworthiness) from samples of observed events or remembered episodes. If those samples are skewed—because of media coverage, selective memory, social networks, or experimental design—then their inferences will look biased even if the underlying inferential rule is normatively sensible.

The recent edited volume Sampling in Judgment and Decision Making further systematises how sampling mechanisms, truncation, and stopping rules affect judgments.(Cambridge University Press & Assessment)

2.4 Resource-rational analysis

Resource-rational analysis integrates rational principles with realistic cognitive constraints, interpreting heuristics as the best algorithms achievable under limitations of time, memory, or attention.(Cognitive Science Lab) Sampling-based algorithms frequently emerge as resource-rational approximations to full Bayesian inference.(OSF) From this perspective, biases often reflect optimal trade-offs between sample quality, sample size, and computational cost—rather than mere irrationality.

3. A sampling-limitation framework for cognitive biases

3.1 What is “sampling”?

Here, sampling refers broadly to any process by which the cognitive system selects a subset of information—from the external environment, internal memory, or hypothesis space—to guide judgments and actions. This includes:

• Attention to a subset of stimuli.

• Retrieval of a subset of memories.

• Consideration of a subset of hypotheses, goals, or actions.

• Observation of a subset of outcomes and feedback.

• Exposure to a subset of social information (opinions, norms, media).

A “sample” is thus a set of instances, cues, or hypotheses that the system entertains at a given moment.

3.2 Dimensions of sampling limitation

I propose five high-level categories of sampling limitations:

• Coverage limits (small-n sampling)

• Too few samples of states, events, or options are gathered to reliably approximate the underlying distribution.

• Leads to over-interpretation of random variation and law-of-small-numbers effects.

• Selective sampling (non-random inclusion)

• Some kinds of instances are sampled with higher probability than others (e.g., salient, confirmatory, emotionally intense).

• Produces skewed samples that systematically over-represent particular features or outcomes.

• Temporal sampling limits

• Only a restricted time window is sampled (recency, peaks), or future outcomes are steeply discounted.

• Generates distortions in how sequences and delays are evaluated.

• Noisy memory sampling

• Retrieval from memory is probabilistic and reconstructive; samples are biased by schemas, emotions, and inference at recall.

• Yields systematic memory distortions and mis-estimation of frequencies.

• Social and ecological sampling distortions

• The social and informational environment itself is unrepresentative (e.g., media, networks, institutions), so the brain samples from a biased ecology.

• Produces false consensus, stereotype learning, survivorship illusions, and norm misperception.

Each limitation class can be connected to specific computational properties: sample size, sampling distribution, truncation, and stopping rules. The next section re-organises common cognitive biases accordingly.

4. Re-organizing cognitive biases by sampling limitation

The following mapping draws on standard lists of cognitive biases(Wikipedia) and on the sampling literature reviewed above.(PubMed) It is not exhaustive, but illustrates how diverse biases fall under the proposed categories.

4.1 Coverage limits: too few samples

Core idea. When only a small number of instances are sampled, random fluctuations look like real patterns. Judgments then systematically deviate from the population statistics.

Representative biases

• Insensitivity to sample size / law of small numbers

People expect small samples to be as representative as large ones, leading to overinterpretation of streaks or extreme proportions.(Tufts Computer Science)

• Sampling explanation: With only a few samples, variance is high; people behave as if this variance is ignored or underestimated when generalising.

• Gambler’s fallacy and hot-hand fallacy

• Gambler’s fallacy: After a run of heads in coin flips, people expect tails to be “due.”

• Hot-hand fallacy: In sports, a streak of successes is taken as evidence for a “hot hand.”(Wikipedia)

• Sampling explanation: In small samples, streaks are common. If one implicitly expects short sequences to match long-run frequencies, deviations look meaningful; sequential sampling models show how boundary conditions can produce these illusions.(PMC)

• Overconfidence in small-sample experience

People are more confident than justified when generalising from a narrow personal history (e.g., “this investment is safe because I’ve never seen it fail”).

• Sampling explanation: The posterior distribution is based on very few samples; if confidence is calibrated to subjective rather than objective variance, overconfidence follows naturally.

• Extension neglect (neglect of scope)

Biases such as compassion fade and insensitivity to the number of victims (e.g., equal willingness to donate for 1 vs. 100 lives)(Wikipedia) can be seen as treating a single vivid instance as the effective sample.

Example. In a classic hospital problem, participants judge which hospital (large vs. small) is more likely to have days with >60% boy births and often answer “both equally likely,” neglecting the higher variance in small samples.(Tufts Computer Science) Under a sampling view, people act as if sample size does not alter expected variability.

4.2 Selective sampling: biased inclusion of instances

Core idea. Samples are not random. Attention, motivation, and environmental structure increase the probability of including certain instances (e.g., salient, confirmatory, extreme) in the sample.

• Availability heuristic and availability cascade

Judgments of frequency and risk are guided by how easily instances come to mind, which themselves depend on salience and media coverage.(UMass People)

• Terrorist attacks and plane crashes, though rare, are overestimated because they are heavily covered and emotionally vivid; mundane causes of death are under-sampled.

• Salience bias and focusing illusion

People over-weight highly salient features of a situation (e.g., climate when thinking about life satisfaction in another city) and under-sample less salient but important factors.(Wikipedia)

• Confirmation bias and congruence bias

Individuals preferentially seek and attend to evidence consistent with their current hypotheses, effectively sampling from a conditional distribution where disconfirming instances are rare.(Wikipedia)

• Framing effects

Different descriptions of objectively equivalent outcomes (e.g., “90% survival” vs. “10% mortality”) highlight different subsets of states, effectively changing which outcomes are sampled mentally.(Science)

• Survivorship bias and outcome bias

Observers see successful firms, projects, or strategies far more often than failed ones (which disappear), leading to overestimation of their effectiveness.(PubMed)

• In medicine or investing, people may judge decisions solely by the observed outcomes, ignoring the unseen counterfactuals—a form of sampled outcome truncation.

Example. In medical risk perception, rare but dramatic side effects of vaccines may be heavily reported, while the benefits (diseases that do not occur) are effectively invisible. The brain’s sample of “what happens after vaccination” is therefore enriched in vivid harms and impoverished in silent successes, exaggerating perceived risk despite statistically favourable base rates.

4.3 Temporal sampling limits: short or distorted time windows

Core idea. Cognitive systems sample experiences over limited time horizons and with non-uniform weighting over time. This yields distortions in evaluating sequences, delays, and trends.

• Recency bias and primacy effects

More recent events are weighted more heavily in judgments (e.g., last impressions in performance reviews, final arguments in jury decisions).(Wikipedia)

• Sampling explanation: The effective sample is drawn from a recency-weighted window; older events have near-zero inclusion probability.

• Peak–end rule and duration neglect

Retrospective evaluation of an episode is disproportionately influenced by its peak and ending moments, with minimal sensitivity to duration.(Wikipedia)

• The brain samples a sparse subset of moments (peaks, transitions) rather than integrating over the full time series.

• Hyperbolic discounting and present bias

People strongly prefer immediate rewards over delayed ones, even when the delayed option is larger, consistent with hyperbolic rather than exponential discounting.(Wikipedia)

• Under a sampling view, future states are sampled less frequently or with lower precision than near-term states, effectively reducing their weight in decision computations.

• Planning fallacy

Project completion times are systematically underestimated; people focus on idealised scenarios and recent successes, under-sampling delays and failures.(Amazon)

Example. When recalling a painful medical procedure, participants’ overall discomfort is predicted better by the average of peak pain and end pain than by total area under the pain–time curve. The memory system appears to sparsely sample the timeline, producing a biased but resource-efficient summary.

4.4 Noisy memory sampling: distortions in retrieval and reconstruction

Core idea. Memory is reconstructive. Retrieval acts as a stochastic sampling process influenced by schemas, emotions, current beliefs, and inference at recall. The resulting sample may be systematically distorted relative to the original events.

• Hindsight bias (“I knew it all along”)

After learning an outcome, people misremember their prior probabilities as being closer to the truth.(Amazon)

• Sampling explanation: When reconstructing prior beliefs, the memory system samples from traces that are now conditioned on the known outcome; inconsistent traces are less accessible or overwritten.

• Consistency and change biases

People remember their past attitudes as more consistent with their current attitudes than they really were, a form of schema-consistent sampling from memory.(Wikipedia)

• Stereotype-consistent memory biases

Information congruent with existing stereotypes is more easily recalled than incongruent information, effectively oversampling stereotype-confirming instances from memory.(Wikipedia)

• Rosy retrospection and positivity/negativity biases

Positive or negative episodes may be preferentially sampled when reconstructing the past, leading to overly nostalgic or pessimistic narratives depending on context.(Wikipedia)

• Misinformation effect and false memories

Post-event information can be integrated into memory so that later retrieval samples a hybrid trace that never actually occurred, leading to confidently held but inaccurate recollections.(Amazon)

Example. In eyewitness testimony studies, exposure to misleading suggestions about a car accident (e.g., “the cars smashed”) alters later memory reports (e.g., higher estimated speed, recollection of broken glass). The retrieval process samples from a schema that now incorporates the suggestion, producing a systematically biased sample of the original experience.

4.5 Social and ecological sampling distortions

Core idea. Individuals rarely sample directly from the objective distribution of events in the world. Instead, they sample from socially filtered information: peers, media, institutions, and algorithms.

• False consensus effect and pluralistic ignorance

People overestimate how widely their own beliefs are shared and misunderstand the private attitudes of others.(Wikipedia)

• Social networks often connect like-minded individuals; thus, the observed sample of opinions is skewed.

• Conformity, bandwagon effects, and group polarization

Individuals update their beliefs based on others’ expressed positions, effectively re-sampling from an already biased pool. Repeated social sampling can push group opinions toward extremes.(Wikipedia)

• Authority bias and prestige bias

Opinions of authorities or high-status individuals are overweighted. Observers allocate attentional sampling disproportionately to such sources, leading to overrepresentation of their views in the mental sample.(Wikipedia)

• Stereotyping and outgroup homogeneity

Limited and selective contact with outgroup members leads to sparse and unrepresentative samples of their behavior; stereotypes are often based on a few salient instances.(Wikipedia)

• Media-driven availability

News media amplify rare but vivid events (e.g., violent crime, terrorist attacks) and under-report mundane but statistically important phenomena, shaping the ecological sample that feeds availability judgments.(Farnam Street)

Example. In an online social network where individuals predominantly follow those who share their political orientation, the local sample of opinions is highly homogeneous. The brain’s sampling process, operating on this skewed ecology, produces strong false consensus (“everyone thinks this way”) and underestimates the diversity of broader public opinion.

4.6 Mapping classic heuristics into the framework

It is instructive to re-interpret the three classic heuristics of Tversky and Kahneman through the sampling lens:(Science)

• Representativeness

Judging category membership by similarity to a prototype can be seen as sampling a small set of highly diagnostic features and ignoring base-rate information (a selective sampling and coverage limitation).

• Availability

Directly expresses biased sampling from memory: ease of retrieval is a proxy for sampled frequency.

• Anchoring and adjustment

Initial anchors are effectively starting samples; subsequent adjustments correspond to additional samples or partial re-sampling that remains insufficient to fully explore the hypothesis space.

Prospect-theoretic phenomena such as loss aversion and probability weighting can also be re-interpreted via sampling models, where rare severe losses are over-sampled or where attention selectively tracks changes relative to a reference point rather than absolute wealth levels.(Amazon)

5. Discussion

5.1 Explanatory power and testable predictions

The sampling-limitation framework offers several advantages:

• Unification across domains.

Memory biases, social biases, and economic anomalies all emerge from the same underlying constraints on sampling and representation, rather than requiring distinct ad-hoc mechanisms.

• Mechanistic specificity.

Each bias can be linked to specific parameters: sample size, sampling distribution, temporal window, truncation, or stopping rule. These parameters can be manipulated experimentally—for example, by changing the diversity of examples presented, extending the sample size, or altering feedback visibility—to test whether the bias attenuates as predicted.(PubMed)

• Normative grounding.

Resource-rational analysis allows us to ask whether a given sampling strategy is optimal under realistic constraints, reframing many “biases” as rational adaptations to computational scarcity.(PubMed)

• Debiasing implications.

If biases are properties of samples, correcting them may require changing how people sample (e.g., increasing sample size, diversifying sources, de-truncating feedback) rather than simply instructing them in normative rules.

5.2 Relation to other taxonomies

Recent work by Oeberst and Imhoff (summarised by Duke) proposes a functional taxonomy of biases, grouping them into clusters related to information processing, ego protection, and social coordination.(annieduke.substack.com) The sampling framework is compatible with such functional groupings but operates at a more mechanistic level: it specifies how information processing is distorted (via sampling) rather than why (e.g., motivation vs. efficiency).

Similarly, the “adaptive toolbox” program emphasises environment-specific heuristics; the sampling perspective can be viewed as describing the statistical interface through which those heuristics interact with the world.(MPG.PuRe)

5.3 Limitations and open questions

Not all biases are purely sampling-based. Some appear to reflect:

• Motivational factors (e.g., self-serving bias, system justification, moral licensing) where desires change valuation or sampling weights in ways not reducible to resource constraints alone.(Wikipedia)

• Active utility distortions (e.g., inequality aversion, fairness concerns) that embed social preferences into the value function rather than the sampling process.

However, even these “hot” biases often interact with sampling: people may choose information sources that support desired conclusions (motivated selective sampling), or recall self-serving episodes more readily (motivated memory sampling).

Open empirical questions include:

• How many samples does the brain typically use for different tasks, and how does sample size adapt to stakes or time pressure?(PubMed)

• When do people detect that their samples are biased, and what metacognitive mechanisms adjust sampling strategies?(PubMed)

• How do social and algorithmic filtering (e.g., recommender systems) reshape the ecological samples feeding into cognitive processes?

6. Conclusion

Cognitive biases have often been portrayed as evidence of human irrationality. A sampling-based perspective suggests a more nuanced picture: the brain is a resource-limited sampling device operating in a complex, often biased environment. Under such conditions, many allegedly irrational patterns of judgment are predictable by-products of small, selective, temporally constrained, and socially filtered samples.

Re-organising cognitive biases by their underlying sampling limitations—coverage limits, selective sampling, temporal limits, noisy memory sampling, and social/ecological distortions—offers a unifying framework that bridges psychology, neuroscience, and computational modeling. It aligns with modern theories of bounded and resource-rational cognition and points toward debiasing strategies that intervene at the level of sampling, not just reasoning.

References

• Fiedler, K. (2000). Beware of samples! A cognitive-ecological sampling approach to judgment biases. Psychological Review, 107(4), 659–676.(PubMed)

• Fiedler, K., Juslin, P., & Denrell, J. (Eds.). (2023). Sampling in Judgment and Decision Making. Cambridge University Press.(Cambridge University Press & Assessment)

• Forstmann, B. U., Ratcliff, R., & Wagenmakers, E.-J. (2015). Sequential sampling models in cognitive neuroscience: Advantages, applications, and extensions. Annual Review of Psychology, 66, 641–666.(PMC)

• Gigerenzer, G., Todd, P. M., & the ABC Research Group. (1999). Simple Heuristics That Make Us Smart. Oxford University Press.(MPG.PuRe)

• Gigerenzer, G., & Selten, R. (Eds.). (2001). Bounded Rationality: The Adaptive Toolbox. MIT Press.(Wiley Online Library)

• Griffiths, T. L., Lieder, F., & Goodman, N. D. (2015). Rational use of cognitive resources: Levels of analysis between the computational and the algorithmic. Topics in Cognitive Science, 7(2), 217–229.(ResearchGate)

• Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.(Amazon)

• Lieder, F., & Griffiths, T. L. (2020). Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources. Behavioral and Brain Sciences, 43, e1.(PubMed)

• Sanborn, A. N., Griffiths, T. L., & Navarro, D. J. (2010). Rational approximations to rational models: Alternative algorithms for category learning. Psychological Review, 117(4), 1144–1167.(PubMed)

• Sanborn, A. N., & Chater, N. (2016). Types of approximation for probabilistic cognition: Sampling and variational. Cognitive Psychology, 89, 1–38.(sciencedirect.com)

• Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69(1), 99–118.(Cooperative Individualism)

• Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131.(Tufts Computer Science)

• Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207–232.(UMass People)

• Vul, E. (2010). Sampling in human cognition (Doctoral dissertation, MIT).(DSpace)

• Vul, E., & Pashler, H. (2008). Measuring the crowd within: Probabilistic representations within individuals. Psychological Science, 19(7), 645–647.(PubMed)

Online bias indices

• The Decision Lab. (n.d.). Biases index. Retrieved from TheDecisionLab.com.(The Decision Lab)

• de Backer, G. (n.d.). Cognitive biases list. Retrieved from gustdebacker.com.(Wikipedia)

• Wikipedia. (2024). List of cognitive biases. Retrieved from Wikipedia.(Wikipedia)

Google

Cognitive biases can be understood as the systematic errors in thinking that result from using mental shortcuts, or heuristics, in situations where they lead to inaccurate judgments. In the context of "sampling," these biases often stem from how individuals gather and interpret information (or "samples" of data) to make decisions.

Rather than processing the full range of available information, humans rely on efficient, but fallible, sampling heuristics:

Key Concepts

• Heuristics as Mental Shortcuts: Heuristics are simple, efficient rules of thumb that people use to make judgments and decisions quickly. They are a form of "fast and frugal" sampling.

• Biases as Outcomes of Heuristics: Cognitive biases are the predictable, systematic errors that occur when these heuristics are misapplied or used in inappropriate contexts. The shortcut itself isn't a bias, but the resulting skewed judgment is.

• Sampling Implications: These "sampling heuristics" lead to biased conclusions because the data being sampled (from memory or the environment) is not representative or is selectively interpreted.

Examples of Biases as Sampling Heuristics

Several well-known cognitive biases illustrate how sampling heuristics work in practice:

• Availability Heuristic/Bias: This is a direct example of a sampling heuristic in action. People estimate the likelihood of an event based on how easily examples come to mind.

• Heuristic: "If I can think of many examples, it must be common."

• Bias (Sampling Error): This leads to an overestimation of the probability of vivid or recent events (e.g., a plane crash) and an underestimation of less memorable but potentially more frequent ones, because the "sample" of memories is skewed by salience, not actual frequency.

• Confirmation Bias: Individuals tend to search for, interpret, and remember information in a way that confirms their pre-existing beliefs, effectively "sampling" data in a biased manner.

• Heuristic: A tendency to give more weight to evidence that supports current beliefs.

• Bias (Sampling Error): This results in a skewed "sample" of evidence, leading to an overly confident, and potentially incorrect, conclusion because contradictory information is ignored or downplayed.

• Anchoring Bias: People rely too heavily on the first piece of information offered (the "anchor") when making subsequent judgments, even if that information is irrelevant.

• Heuristic: Using a convenient starting point for estimation.

• Bias (Sampling Error): The final judgment is "biased" towards the initial anchor, rather than being based on a comprehensive, unbiased evaluation of all relevant data points.

• False Consensus Effect: People tend to overestimate the extent to which others agree with them.

• Heuristic: Assuming one's own immediate social circle is representative of the general population.

• Bias (Sampling Error): The "sample" of opinion (friends, family, colleagues) is too narrow and unrepresentative, leading to the incorrect assumption of a broader consensus.

In essence, these biases demonstrate how the brain, limited by time and cognitive capacity, employs efficient but potentially unrepresentative "sampling" methods to navigate a complex world.

In addition to the availability and confirmation biases, many other cognitive biases stem from using flawed "sampling" heuristics. These are often rooted in the representativeness heuristic (judging something based on how well it matches a stereotype, rather than statistical data) and actual selection biases in how information is gathered.

Here are several key biases that stem from sampling heuristics:

Biases Stemming from the Representativeness Heuristic

The representativeness heuristic involves judging the probability of an event by how typical or representative it is of a mental model or stereotype, often leading to a failure to consider actual statistical probabilities (base rates).

• Base Rate Neglect (or Base Rate Fallacy): The tendency to ignore general, objective statistical information (the "base rate") in favor of specific, vivid details or anecdotes.

• Sampling Link: People focus on the appealing individual "sample" (e.g., a personality description) and neglect the larger, more representative "sample" of the overall population statistics.

• Insensitivity to Sample Size: The tendency to under-expect variation in small samples, believing that a small sample will perfectly reflect the average of a large population.

• Sampling Link: This bias ignores the statistical principle that small samples are more likely to deviate from the population mean than large ones, essentially treating small and large samples as equally reliable.

• Gambler's Fallacy: The erroneous belief that if a random event has occurred more frequently than normal in the past, it is less likely to happen in the future (or vice versa), because the person expects the outcomes to "balance out" in the short term.

• Sampling Link: This results from a misconception of chance and the belief that a small, local sample of outcomes should be representative of a globally random process.

• Conjunction Fallacy: The tendency to believe that a specific set of conditions (a conjunction) is more probable than a single general condition, especially if the specific conditions are highly representative of a certain stereotype.

Biases Related to Information Selection and Memory

These biases involve flawed methods in the actual process of information gathering and recollection, which create a skewed "sample" of data to draw conclusions from.

• Survivorship Bias: The error of concentrating only on successful outcomes or "survivors" and overlooking failures, which leads to a skewed perspective of the true situation.

• Sampling Link: The "sample" is incomplete because all the failures (e.g., companies that went bankrupt, planes that crashed in certain areas) are excluded from the analysis, leading to incorrect conclusions about the attributes of the entire group.

• Self-Selection Bias (or Voluntary Response Bias): Occurs when individuals select themselves into a group, potentially leading to a sample that does not accurately represent the intended population because people with specific characteristics are more likely to participate.

• Sampling Link: Researchers end up with a non-random, biased sample because participation is voluntary (e.g., an online survey will exclude people without internet access).

• Recall Bias: A systematic error caused by differences in the accuracy or completeness of participants' recollections of past events.

• Sampling Link: The "sample" of memories retrieved is biased because people tend to better remember vivid, recent, or emotionally charged events, while forgetting more mundane ones.

These examples highlight how humans, in the interest of cognitive efficiency, use imperfect sampling strategies that systematically distort their perception of reality.

How does confirmation bias affect the way we sample information?

Confirmation bias is a powerful "sampling heuristic" that systematically skews the information individuals gather, interpret, and recall by pushing them to favor data that aligns with their pre-existing beliefs and devalue or ignore evidence that contradicts them.

This bias manifests in three primary ways regarding information sampling:

• Biased Search for Information

Rather than conducting a neutral and comprehensive search for all relevant information, individuals actively seek out data that supports their current hypothesis or opinion.

• Seeking "Positive Tests": People often phrase questions or search queries to receive an affirmative answer, looking for consequences that would happen if their hypothesis were true, rather than seeking evidence that would prove the view false (falsifying evidence).

• Selective Exposure: Individuals may limit their information sources to those they already agree with, creating "echo chambers" (especially online via social media algorithms), which reinforces a narrow and unrepresentative sample of available viewpoints.

• Biased Interpretation of Information

Even when presented with a balanced set of data, people interpret ambiguous information in a way that aligns with their preconceptions.

• Unequal Standards of Evidence: Information that supports existing beliefs is readily accepted with little scrutiny, while contradictory evidence is treated with high skepticism and is often dismissed as unreliable or flawed.

• Focusing on Confirming Details: When evaluating complex information, people tend to focus on specific details that support their viewpoint while downplaying or ignoring contradictory aspects, effectively cherry-picking data to fit their narrative.

• Biased Recall of Information

Confirmation bias also affects memory, leading to a skewed "sample" of information when recalling past events or data points.

• Selective Memory: Individuals are more likely to remember information that is consistent with their existing beliefs or expectations and forget information that challenges them.

• Reconstructing Memories: People may even reconstruct their memories of past events to be consistent with their current emotional state or beliefs, reinforcing the idea that their current perspective has always been correct.

In essence, confirmation bias prevents a balanced and representative sampling of reality, leading to an overly confident, yet potentially incorrect, fortification of existing beliefs.

1