Reasoning

Reasoning is the process of drawing conclusions from information taken as given, the premises. It is commonly divided into deduction, where the conclusion follows necessarily from the premises; induction, where it is generalised from cases; and abduction, where it is the best explanation of what is observed (Wikipedia contributors, n.d.b). Four kinds of theory explain how people reason and why they err: mental logic, which posits formal rules of inference in the mind; content-specific rules; mental models, in which people imagine the possibilities the premises allow; and probabilistic accounts, which treat reasoning as estimating how likely a conclusion is (Wikipedia contributors, n.d.b). Dual-process theory adds a fifth framing that cuts across them. Underneath the competition lies a more basic dispute: which standard, classical logic, non-monotonic logic or probability theory, defines correct human reasoning (Wikipedia contributors, n.d.b). Most of the central questions are contested, including how to measure the best-known error, belief bias (Dube et al., 2010; Klauer & Kellen, 2011).

History

Early research assumed that people reason with a mental logic, a set of formal inference rules like those of a logic textbook. Most cognitive psychologists abandoned that hypothesis after experiments showed poor performance on logical tasks, and turned to heuristics, schemas or mental models instead (Cosmides & Tooby, 1994). Mental-model theory became a main alternative, contrasted with both logic-based and probabilistic accounts (Johnson-Laird & Ragni, 2024). Probabilistic accounts came later; Johnson-Laird and Ragni credit them as the first attempt to explain human reasoning without relying on logic at all (Johnson-Laird & Ragni, 2024).

Core concepts

Mental models

Mental-model theorists hold that people reason by imagining models of the situations the premises describe, each model representing one possibility, rather than by applying formal rules of inference (Johnson-Laird & Ragni, 2024). Johnson-Laird, Khemlani and Goodwin argue that the theory solves problems that both classical logic and probability logic leave open (Johnson-Laird et al., 2015), and Johnson-Laird and Ragni argue that probabilistic accounts cannot handle modal assertions, statements about what is possible or necessary (Johnson-Laird & Ragni, 2024).

The theory predicts illusory inferences: conclusions that seem compelling but are wrong, because the reasoner’s models leave out a possibility. Three experiments published in 2024 found that people draw illusory inferences from conditional statements, with the pattern depending on the wording, “if … then”, “only if” or “if and only if” (Espino et al., 2024).

Probabilistic accounts

In the Bayesian approach, an everyday conditional, “if p then q”, is understood as the conditional probability of q given p, written P(q|p). This differs from the material conditional of classical logic, which counts “if p then q” as true whenever p is false or q is true (Cruz et al., 2015). On this view, many answers counted as errors under logic are reasonable under probability, which is why the choice of standard matters. Probabilistic models in general face the criticisms covered in Bayesian models of cognition.

Dual-process accounts of reasoning

Dual-process theories of reasoning assume two qualitatively different kinds of processes. Type 1 processes are autonomous, fast and make little demand on capacity; Type 2 processes need working memory and are slower. Biases are attributed to the outputs of Type 1 processes, which Type 2 processing may or may not correct (Thompson et al., 2018). On this view, people with higher cognitive capacity reason better because they engage Type 2 processing more (Thompson et al., 2018).

Questions, controversies, and new developments

Which standard of correctness?

A central controversy in the psychology of reasoning is which competence model, classical logic, non-monotonic logic (in which new information can withdraw a conclusion) or Bayesian probability, should be the standard against which human reasoning is judged (Wikipedia contributors, n.d.b). The answer decides what counts as an error, so it shapes the evidence for every theory.

Is belief bias a reasoning effect?

Belief bias is the tendency to accept arguments whose conclusions are believable and reject those whose conclusions are not, regardless of whether the argument is valid. Dube, Rotello and Heit analysed syllogistic reasoning with receiver operating characteristic (ROC) curves, a method from signal-detection theory that separates how well people discriminate valid from invalid arguments from how willing they are to say “valid”. They found curved ROCs and concluded that belief bias is a response-bias effect, a shift in willingness to accept, not an effect on accuracy. That conclusion challenged the existing theories of belief bias (Dube et al., 2010). Klauer and Kellen reanalysed the data and concluded that belief bias is more than a response bias (Klauer & Kellen, 2011). The measurement dispute is unsettled.

Does logic come first, or last?

The standard dual-process account says the correct logical answer requires slow Type 2 thinking. Several findings challenge that sequence:

• Correct logical responses are often given as the first, immediate response, before any time for deliberation (Bago & De Neys, 2017).
• High-capacity reasoners favoured logic or statistics over belief when the two conflicted, even under a deadline. Thompson and colleagues conclude that part of the link between capacity and accuracy reflects intuitive Type 1 processes, not Type 2 engagement (Thompson et al., 2018).
• Dual-process accounts disagree on whether logical responses can be generated automatically (“intuitive logic”) or only by reflection. A 2025 study found that conflict arises between two heuristic responses rather than between logic and a heuristic (Kosourikhina & Handley, 2025).
• Whether biased reasoners detect that their answer conflicts with logic is disputed. Mata and Ferreira argue that conflict detection should not be expected in people who miss the critical premises, while leaving open that they detect conflict implicitly (Mata & Ferreira, 2018).

The broader dispute over whether two types of process exist at all is covered in Dual-process theory.

Replication record

Mental-model illusory inferences reappeared in new experiments with new wordings (Espino et al., 2024). Belief bias is reliably observed, but whether it reflects reasoning or response bias depends on the analysis (Dube et al., 2010; Klauer & Kellen, 2011).

Broader connections

Reasoning shares its theories with judgment and decision-making: heuristics, dual processes and probability norms appear in both, and the dispute over narrow norms in judgment mirrors the dispute over the standard of correct reasoning here (Gigerenzer, 1996). Reasoning that searches for a way to reach a goal is treated in Problem solving.

Further reading

• Johnson-Laird et al. (2015): the mental-model case against both logic and probability as accounts of reasoning.
• Cruz et al. (2015): the probabilistic reading of conditionals.
• Thompson et al. (2018): evidence that intuitions, not only deliberation, explain why capable reasoners do better.