Inductive vs. Deductive Reasoning: Differences and Examples
Author: Angela Bogdanova · Published: October 12, 2026 · Editorial Policy
Research cluster: Intelligence, Reasoning & Cognitive Abilities · Written by Artificial Sapiens
Inductive and deductive reasoning are two ways of moving from what we know to what we conclude. Inductive reasoning uses observations and evidence to support a generalization, prediction, or other conclusion that could still turn out to be wrong. Deductive reasoning asks whether a conclusion must follow from the premises under a specified logical structure. The distinction is about the relationship between evidence and conclusion, not simply whether someone starts with a broad idea or a specific example.
Consider a workplace situation. After observing that several meetings with clear agendas finish on time, a manager infers that clear agendas generally improve meeting efficiency. That is induction: the observations support a broader hypothesis. Now suppose a company policy says that every meeting classified as confidential must use an approved meeting room, and Tuesday's meeting is classified as confidential. Concluding that Tuesday's meeting must use an approved room is deduction, provided the policy and classification are accepted as premises.
The central comparison is logical strength. In a valid deduction, true premises guarantee the conclusion. In a strong induction, the premises make the conclusion well supported, often probable, while leaving room for exceptions or further evidence. This basic distinction is formalized in the Stanford Encyclopedia of Philosophy's treatment of inductive logic.
Inductive vs. deductive reasoning: key differences at a glance
Inductive reasoning: The conclusion extends beyond what the premises strictly establish. Its quality depends on the evidence, sampling, alternative explanations, relevant background knowledge, and how much confidence is justified. It commonly supports predictions, regularities, categories, and hypotheses. A strong argument can be revised by a new counterexample or a more representative sample.
Deductive reasoning: The conclusion is logically entailed by the premises when the argument is valid. Its quality as a real-world argument also depends on whether its premises are true or justified. It commonly supports rule application, mathematical proof, constraint checking, and conditional predictions. A valid deductive step stays valid when an additional premise is added, although conflicting premises can create separate problems.
The traditional memory aid calls induction bottom-up and deduction top-down. This is useful for introductory examples, but it is incomplete. Induction can produce predictions about a single event, and deduction can establish a general conclusion from general premises. The decisive question is whether the conclusion follows necessarily or receives a degree of support.
What is inductive reasoning?
Inductive reasoning is an inference in which the premises supply evidence for a conclusion without logically guaranteeing that conclusion. It lets people learn from limited experiences and make predictions about cases they have not observed. Induction appears whenever a person recognizes a recurring pattern, expects an event to continue, or extends evidence from known cases to new ones.
Suppose a library records increased attendance on the last eight rainy Saturdays. The librarian predicts that attendance will rise next rainy Saturday. The conclusion may be reasonable, but it could fail: a holiday, closure, or transportation interruption could change the result. Better evidence would include a larger number of Saturdays, comparison with dry days, seasonal trends, and plausible competing explanations.
Induction is often illustrated by moving from particular observations to a general rule: this bird can fly, that bird can fly, so all birds can fly. The conclusion is weakened by the existence of flightless birds. Yet induction also includes a general-to-particular forecast: most trains on this route arrive on schedule, so tomorrow's train will probably arrive on schedule. What unites these examples is their nonconclusive evidential support.
Common kinds of inductive inference
Generalization moves from examined cases to a wider population. Observing frequent delays among a small group of buses may suggest delays are common on the route. The sample's size, selection method, and representativeness matter more than the mere number of observations.
Statistical prediction applies a well-supported frequency or regularity to a new case. If an accurately measured service has a high on-time rate, forecasting that the next delivery will be on time is an inductive prediction, not a guaranteed outcome.
Category-based induction projects a property from some category members to others. Learning that several related plants are resistant to a pest may lead a scientist to test whether other members share that trait. Similarity, typicality, and the diversity of examples influence perceived inductive strength. Experimental work reports meaningful effects of these factors in category-based judgments, including Hampton and Cannon (2004) and Rein, Goldwater, and Markman (2010).
Causal hypotheses often begin with observed regularities, but moving from correlation to causation requires additional tests of mechanisms, temporal ordering, and alternative explanations. Observing that students who sleep longer tend to perform better does not establish that increasing sleep alone causes the entire performance difference. Experimental and longitudinal designs can strengthen causal inference, but an association by itself does not.
Analogical inference draws a tentative conclusion from relevant similarities between cases. It is often inductive, yet analogical reasoning has its own substantial literature and is not reducible to one simple template.
What makes an induction strong?
A strong inductive argument uses evidence appropriate to the claim. Its sample reflects the population of interest, its measurements answer the intended question, its observations are sufficiently independent, and its conclusion does not overreach beyond the data.
Suppose someone observes 99 working devices out of 100 selected at random from production. That is more informative about the factory's failure rate than testing 99 devices selected by a salesperson because they already passed inspection. The two observed counts might look identical while supporting very different claims.
Evidence diversity also matters. A finding repeated across different locations, populations, and conditions may generalize more reliably than the same finding repeated in one narrow setting. Research on category-based induction documents interactions between evidence diversity, typicality, and prior knowledge; for example, López and colleagues (1992) examined how these judgments develop, and Lu and colleagues (2023) studied how typicality modifies diversity effects.
Inductive conclusions should therefore use language that matches the evidence: likely, tends to, is associated with, or supported under these conditions. When evidence is sparse, the appropriate conclusion may be a hypothesis to test rather than a stable generalization.
What is deductive reasoning?
Deductive reasoning evaluates whether a conclusion necessarily follows from given premises. It is often expressed with a rule and a case, but its essential feature is logical entailment: there is no possible situation in which all premises are true and the conclusion is false if the argument is valid.
Consider this argument. Every item in box A has been inspected. This item is in box A. Therefore, this item has been inspected. The inference is deductively valid. Its real-world correctness depends on the truth of the premise that all items in the box were inspected and the accuracy of the item's classification.
Logic distinguishes validity from soundness. A valid argument has the right inferential structure. A sound argument is valid and has true premises. For example, 'All bicycles are electric; this vehicle is a bicycle; therefore this vehicle is electric' has a valid form even though the general premise is false. The conclusion is not established as true in the actual world. The Stanford Encyclopedia of Philosophy entry on argument explains this distinction.
Three familiar deductive structures
Categorical reasoning uses class relationships. All registered participants have identification badges. Maya is a registered participant. Therefore Maya has an identification badge. The conclusion follows from the premises, even if the policy's accuracy should separately be checked.
Modus ponens uses a conditional and its antecedent. If the server loses power, the monitoring system sends an alert. The server lost power. Therefore the system sent an alert. This is a valid form only relative to the assumed conditional. If the system can fail or the policy has exceptions, the major premise requires qualification before the argument describes the real world.
Modus tollens uses a conditional and the denial of its consequent. If a file was successfully encrypted under the specified procedure, it has the expected encryption marker. The file lacks that marker. Therefore the file was not successfully encrypted under the specified procedure. Again, the validity of the inference and reliability of the conditional premise are different questions.
The form 'If P, then Q; Q; therefore P' is invalid. The alert may have sounded for a different reason. Likewise, 'If P, then Q; not P; therefore not Q' is invalid. The monitoring system might send an alert for another event. These mistakes are called affirming the consequent and denying the antecedent.
How the same problem can use both types of reasoning
Imagine that a school counselor notices that students who consistently attend an optional study workshop often submit assignments on time. This pattern generates an inductive hypothesis: participation might help with organization. The counselor then proposes a policy rule for a specific program: any student enrolled in the workshop receives access to a designated planning resource. When Jordan enrolls, the counselor deduces that Jordan qualifies for that resource under the rule.
The inductive hypothesis concerns a possible empirical relationship and remains open to testing. The deductive application follows a stipulated institutional rule. The workshop observation does not make the enrollment rule true; the policy establishes that rule separately.
Scientific inquiry also combines forms of reasoning. Researchers observe patterns and propose explanations, derive predictions from a hypothesis together with background assumptions, test those predictions, and use the resulting evidence to update their confidence. A prediction following deductively from a hypothesis does not mean that observing the prediction proves the hypothesis. Several competing hypotheses may predict the same result. The Stanford Encyclopedia of Philosophy on confirmation examines this challenge in relating evidence to hypotheses.
Worked examples: identify the reasoning step
Example 1: Weather patterns
In a town's records, nine of the last ten afternoons with a certain cloud pattern were followed by rain. A resident predicts rain this afternoon after observing that pattern. This is induction because the historical evidence raises the plausibility of rain without entailing it. Relevant questions include how the cloud pattern was defined, how often it appears without rain, and whether seasons changed.
Example 2: A laboratory safety rule
The laboratory manual says that all samples marked R must be stored below a specified temperature. Sample 27 is marked R. Therefore sample 27 must be stored below that temperature. This is deduction. It establishes what the policy requires, conditional on the stated premises; it does not by itself show the sample was actually stored correctly.
Example 3: Product reliability
A team tests 500 randomly selected units and finds that a component often fails after prolonged heat exposure. The team predicts an increased failure risk for future units used in hot environments. This is induction from measured cases to a wider population, with uncertainty that should be quantified rather than ignored.
Example 4: Mathematics
Every multiple of four is even. Thirty-six is a multiple of four. Therefore 36 is even. This is a valid deduction with true premises. Unlike empirical forecasts, the conclusion can be demonstrated by the mathematical relationship stated in the premises.
Example 5: Reading research
A study finds an association between frequent exercise and improved concentration among its participants. Concluding that exercise causes the improvement in every individual is a causal overreach from an observational pattern. The step begins as induction and fails because the proposed conclusion exceeds the support of the design.
Example 6: Applying an eligibility criterion
A program's published rule says that applicants who submit all required documents by the deadline qualify for an administrative review. Ana submitted all required documents by the deadline. Therefore Ana qualifies for the administrative review. This is deduction from the rule and case, though it does not establish approval of the application.
Example 7: Diagnosing a technical problem
A laptop shuts down unexpectedly; its fan is loud; high temperature often produces this combination. Inferring that overheating is one plausible explanation is an inference to the best explanation, commonly called abduction. It becomes more credible if alternative hypotheses such as battery failure are examined. Abduction is closely related to evidence-based inference, but it has its own explanatory purpose.
Example 8: A counterexample
Someone claims that every animal in a certain pond is a fish. Finding a frog in that pond refutes the universal claim, assuming the classification is accurate. One counterexample is sufficient to show the universal premise false. Discovering it is an empirical observation; recognizing that the universal statement and the counterexample cannot both hold uses deductive logic.
Inductive strength, deductive validity, soundness, and cogency
These terms answer different evaluative questions. Deductive validity asks whether the conclusion follows in every case where the premises are true. Deductive soundness adds the requirement that the premises are in fact true.
Inductive strength asks how well the premises support the conclusion given the evidence and relevant background assumptions. Inductive cogency is commonly used for a strong inductive argument whose premises are true; terminology varies across introductory logic traditions, so the substantive evidential assessment matters more than applying the label.
A strong induction may have a false conclusion, and a weak induction may accidentally reach a true one. Equally, an invalid deductive argument might have a true conclusion by coincidence. Evaluating reasoning means judging how the premises support the conclusion, not simply checking whether the final sentence happens to be correct.
Why a valid argument may still be misleading
A person can offer a perfectly valid inference built on unverified data, an exaggerated universal statement, or a poorly specified conditional. 'If this device is defective, it fails inspection; the device is defective; therefore it fails inspection' is valid, but the first premise could describe an unreliable inspection procedure. Logical form secures the implication conditional on the premises; factual inquiry establishes whether those premises deserve acceptance.
Why additional evidence can change induction
New observations can reverse an inductive assessment. If a restaurant has been excellent on six visits, predicting another good visit may be reasonable. A recent management change, a documented decline in quality, or evidence that earlier visits were atypical could change the forecast. Deductive entailment behaves differently: when an inference is valid, appending another premise cannot create a possible case in which the original premises remain true and the conclusion false. This structural contrast is discussed in the Stanford Encyclopedia of Philosophy's account of induction.
How people actually reason: what psychology finds
Formal logic describes standards for arguments, while cognitive psychology investigates the mental processes people use when evaluating them. People draw on knowledge, memory, language comprehension, representations of possibilities, and judgments about how believable a conclusion seems. Neither induction nor deduction should be equated with one unique neural mechanism or one fixed mental 'system.'
Believability can influence judgments of validity
People are often tempted to accept an argument because the conclusion sounds plausible. Consider: 'All flowers have petals. All roses have petals. Therefore all roses are flowers.' The conclusion is believable, yet the premises are compatible with other petaled plants, so they do not establish the conclusion. This is a standard example of belief bias.
A hierarchical Bayesian meta-analysis by Trippas, Kellen, Singmann, and Pennycook (2018) analyzed 22 syllogistic-reasoning studies with 993 participants. It found that conclusion believability affects people's willingness to endorse arguments; the interpretation of whether beliefs change discrimination between valid and invalid arguments depends on the measurement model and individual differences. This is stronger evidence for a reliable behavioral phenomenon than for any single simple explanation of its mechanism.
Competing models of reasoning remain under study
Psychological theories propose formal rules, mental models, probabilistic assessment, and combinations of heuristic and reflective processing. A meta-analysis by Khemlani and Johnson-Laird (2012) evaluated 12 theories of syllogistic reasoning and found that none adequately accounted for the full range of data considered.
Dual-process accounts distinguish relatively automatic processes from reflective ones, yet the mapping from process characteristics to discrete mental systems is contested. Evans's Annual Review of Psychology (2008) emphasizes that different theories assign different properties to those processes. It would therefore be misleading to equate induction simply with intuition and deduction simply with slow, conscious thought.
Experiments directly comparing inductive and deductive instructions have also challenged the assumption that the two judgments require completely separate processing systems. Rips (2001) reported evidence for distinguishable judgments of correctness and strength, while a modeling study by Stephens and colleagues (2018) found evidence favoring a one-dimensional account with different decision criteria in the tasks analyzed. A later training experiment by Stephens and colleagues (2020) tested further predictions. These findings concern particular laboratory tasks and models, rather than a definitive separation of two human cognitive faculties.
What brain research does and does not establish
Neuroimaging studies find involvement of distributed frontal and parietal regions during reasoning tasks. In a quantitative meta-analysis of 28 deductive-reasoning imaging studies, researchers identified recurrent activation patterns that differed by argument type. An experimental comparison by Goel and Dolan (2004) reported partly overlapping brain activity during induction and deduction, with some task-related differences.
These findings support investigating partially differentiated processes within broader reasoning networks. They do not justify assigning induction to one side of the brain, deduction to another, or treating an activation map as a direct measurement of a person's reasoning capacity. Reviews of reasoning neuroimaging emphasize the interpretive limits of moving from brain activation to an explanation of cognitive operations, as discussed in Brain Imaging, Forward Inference, and Theories of Reasoning.
Are inductive and deductive reasoning forms of intelligence?
Reasoning ability is relevant to individual differences in cognitive performance, but a logical category of inference is not an IQ score. Some cognitive tests contain tasks involving pattern completion, relational inference, or logical relationships. Performance on those tasks depends on the assessment, its instructions, knowledge demands, time limits, and broader cognitive abilities.
The distinction between induction and deduction describes how a conclusion is supported. Intelligence research studies patterns of performance across tasks and people. Working memory, language knowledge, executive control, and processing speed may contribute to success in certain reasoning tasks, but none is interchangeable with a specific inference form. A person can make an invalid argument under distraction without that error defining their overall intelligence, and a successful deduction does not by itself establish exceptional intellectual ability.
Research in cognitive development also finds that the use of sample diversity and category knowledge changes across tasks and ages. Such findings describe developing strategies and background knowledge; they are not grounds for diagnosing a person or judging their global intellectual capacity from a few examples.
Induction, deduction, abduction, and related thinking skills
Induction addresses how evidence supports conclusions beyond what is logically entailed. Deduction addresses whether conclusions necessarily follow. Abduction addresses which available explanation best accounts for observations. Real investigations frequently cycle among all three.
Analogical reasoning uses relational similarities to support a transfer or inference. It may support an inductive projection, but analogies also help represent deductive problems. Statistical inference supplies formal tools for analyzing samples, uncertainty, and hypotheses; its methods involve probabilistic assumptions and can contain deductive mathematical steps.
Critical thinking is a broader practice of evaluating premises, evidence quality, arguments, alternative interpretations, and decision consequences. Metacognition concerns monitoring and regulating one's own thinking. These concepts overlap in application while retaining their own definitions and evidence bases.
How to choose which reasoning approach to use
Use induction when the question is empirical: What pattern appears in the data? Does a tendency generalize? What is likely to happen next? Which hypotheses are worth investigating? Strengthen the answer through representative evidence, competing explanations, replications, and appropriate uncertainty.
Use deduction when the question concerns a consequence of stated rules or assumptions: Does this case meet the eligibility conditions? Does a mathematical result follow? Is a proposed conclusion compatible with the specified premises? Strengthen the answer by checking validity, definitions, exceptions, and the factual accuracy of the premises.
Use both when a project moves from observation to prediction to evaluation. For example, a school may notice a recurring pattern of late assignments, hypothesize a scheduling bottleneck, derive predictions about which classes should improve after a schedule change, test the change, and revise the hypothesis based on the outcome. The empirical effects are assessed inductively; the hypothesis's testable implications depend on deductive steps and auxiliary assumptions.
How to improve reasoning in practice
First identify the conclusion and list its premises explicitly. Ask whether the conclusion is being claimed as necessary or merely likely. This quickly reveals whether the argument should be tested for logical validity or assessed for evidential strength.
For deduction, translate a complicated statement into a simpler conditional or class relationship. Ask whether you can imagine any situation in which the premises hold and the conclusion fails. Search for the classic conditional errors: affirming the consequent and denying the antecedent. Then check whether the premises are actually justified.
For induction, define the population and sampling method, look for missing cases and selection bias, compare alternative explanations, and use confidence language proportional to the evidence. Identify observations that would substantially weaken your current conclusion. A sample drawn from unusually successful cases should not stand in for an entire population.
For both, separate the plausibility of a conclusion from the validity of the argument that produced it. Compare at least one alternative account. Where the stakes are high, use independent verification and domain-relevant expertise rather than a single intuitive judgment.
Practice can help people recognize logical structures and evaluate evidence in the kinds of tasks they practice. Broad claims that a particular reasoning exercise will reliably increase general IQ or produce far transfer to every real-world decision require independent evidence; they do not follow from improved performance on training exercises.
Common misunderstandings
'Induction is always specific to general, and deduction always general to specific.' These patterns are common classroom illustrations but leave out inductive predictions about a particular event and deductions whose conclusions are general. Logical entailment and evidential support provide the more precise distinction.
'Deduction is always true.' Valid deductions preserve truth from their premises, but premises may be false or insufficiently grounded. Validity alone does not establish a true conclusion in the real world.
'Induction is guessing.' Inductive reasoning can be rigorous and supported by systematic data, experimental designs, and quantitative models. Its conclusions remain open to revision because the evidence does not logically force them.
'Deduction is scientific, induction is unscientific.' Scientific inquiry uses observation, hypothesis generation, logical derivation, measurement, probabilistic inference, and replication. Both forms are central to research.
'Inductive reasoning proves a causal relationship.' Observed associations may motivate causal hypotheses; causal evidence requires attention to research design, alternative causes, and mechanisms.
'Deduction requires stronger intelligence than induction.' Neither logical form gives a ranking of people's intelligence. Task difficulty, knowledge, context, instructions, and the criteria for correctness vary substantially.
Frequently asked questions
What is the main difference between inductive and deductive reasoning?
Deductive reasoning seeks a conclusion that follows necessarily from its premises. Inductive reasoning evaluates how strongly evidence supports a conclusion that could still be false. Their characteristic standards are validity and evidential strength.
What is a simple example of induction?
A gardener notices that a particular plant has grown faster during several warm weeks and predicts faster growth during the next comparable warm period. The prediction extends from observed cases to a new case and may be revised when other factors are considered.
What is a simple example of deduction?
All library books labeled reference-only must remain in the building. This book is labeled reference-only. Therefore it must remain in the building. The conclusion follows from the stated rule and classification.
Can inductive reasoning be wrong even when the observations are true?
Yes. A set of observations can be accurate while the inferred generalization fails because the sample is limited, exceptions exist, or the pattern changes. Inductive strength measures justified support, not an absolute guarantee.
Can deductive reasoning be wrong if the premises are false?
A valid deduction can have false premises, and thus it may not establish the truth of its conclusion in the actual world. It can still be logically valid. Soundness requires both validity and true premises.
Is the scientific method inductive or deductive?
Scientific research uses both. Observations may motivate a hypothesis, researchers derive testable implications, and new evidence changes confidence in the hypothesis. Testing a prediction cannot by itself establish that one favored hypothesis is the only possible explanation.
Is Sherlock Holmes using induction or deduction?
Fictional investigations commonly combine observation-driven generalization, inference to the best explanation, and deductions from stated clues or constraints. Inferring the most plausible culprit from incomplete evidence is often more accurately described as abduction than deductive certainty.
Is deductive reasoning the same as critical thinking?
Deduction evaluates entailment. Critical thinking covers a wider set of practices, including assessing evidence credibility, background assumptions, alternatives, relevance, and practical judgment.
Is inductive reasoning always probabilistic?
Its conclusions receive nonconclusive support, often expressed probabilistically. Whether a precise numerical probability can be assigned depends on the model, evidence, and assumptions. An informal induction may be reasonable without a measured probability.
Which is better for problem solving?
Neither is universally superior. Induction is indispensable when learning patterns from uncertain evidence. Deduction is indispensable when deriving implications from rules or constraints. Difficult problems often require a sequence of both, with abduction and other forms of inference playing additional roles.
Related Articles
References
Evans, J. St. B. T. (2008). Dual-processing accounts of reasoning, judgment, and social cognition. Annual Review of Psychology, 59, 255–278. https://doi.org/10.1146/annurev.psych.59.103006.093629
Goel, V., & Dolan, R. J. (2004). Differential involvement of left prefrontal cortex in inductive and deductive reasoning. Cognition, 93(3), B109–B121. https://doi.org/10.1016/j.cognition.2004.03.001
Hampton, J. A., & Cannon, I. (2004). Category-based induction: An effect of conclusion typicality. Memory & Cognition, 32, 235–243. https://doi.org/10.3758/BF03196855
Heit, E. (2015). Brain imaging, forward inference, and theories of reasoning. Frontiers in Human Neuroscience, 8, 1056. https://pmc.ncbi.nlm.nih.gov/articles/PMC4288126/
Khemlani, S., & Johnson-Laird, P. N. (2012). Theories of the syllogism: A meta-analysis. Psychological Bulletin, 138(3), 427–457. https://doi.org/10.1037/a0026841
López, A., Gelman, S. A., Gutheil, G., & Smith, E. E. (1992). The development of category-based induction. Child Development, 63(5), 1070–1090. https://pubmed.ncbi.nlm.nih.gov/1446543/
Lu, L., Yang, J., Shu, R., & Long, C. (2023). The default-interventionist model underlies premise typicality weakening the premise diversity effect during category-based induction. Scandinavian Journal of Psychology, 64(3), 325–338. https://doi.org/10.1111/sjop.12892
Prado, J., Chadha, A., & Booth, J. R. (2011). The brain network for deductive reasoning: A quantitative meta-analysis of 28 neuroimaging studies. Journal of Cognitive Neuroscience, 23(11), 3483–3497. https://pmc.ncbi.nlm.nih.gov/articles/PMC3188687/
Rein, J. R., Goldwater, M. B., & Markman, A. B. (2010). What is typical about the typicality effect in category-based induction? Memory & Cognition, 38, 377–388. https://doi.org/10.3758/MC.38.3.377
Rips, L. J. (2001). Two kinds of reasoning. Psychological Science, 12(2), 129–134. https://doi.org/10.1111/1467-9280.00322
Stanford Encyclopedia of Philosophy. (2021). Argument and argumentation. https://plato.stanford.edu/entries/argument/
Stanford Encyclopedia of Philosophy. (n.d.). Confirmation. https://plato.stanford.edu/entries/confirmation/
Stanford Encyclopedia of Philosophy. (2025). Inductive logic. https://plato.stanford.edu/entries/logic-inductive/
Stephens, R. G., Dunn, J. C., & Hayes, B. K. (2018). Are there two processes in reasoning? The dimensionality of inductive and deductive inferences. Psychological Review, 125(2), 218–244. https://doi.org/10.1037/rev0000088
Stephens, R. G., Dunn, J. C., Hayes, B. K., & Kalish, M. L. (2020). A test of two processes: The effect of training on deductive and inductive reasoning. Cognition, 199, 104223. https://doi.org/10.1016/j.cognition.2020.104223
Trippas, D., Kellen, D., Singmann, H., & Pennycook, G. (2018). Characterizing belief bias in syllogistic reasoning: A hierarchical Bayesian meta-analysis of ROC data. Psychonomic Bulletin & Review, 25, 2141–2174. https://doi.org/10.3758/s13423-018-1460-7