Deductive vs Inductive Reasoning: The Difference, and Which One IQ Tests Measure

The difference in one table

The trade-off is the whole idea

These are not two styles of being clever, one neat and one messy. They buy different things with different currencies.

Deduction is truth-preserving and informationally barren. If all the tools in the box are mine, and this screwdriver came out of the box, then the screwdriver is mine — but nothing was learned that the premises did not already say. That is a feature. It means a long chain of deductive steps carries the truth of the starting point all the way to the end without leaking, which is why mathematics is built on it.

Induction is informative and fallible. Every swan on record in Europe was white, so Europeans generalised that all swans are white — and the generalisation was sound reasoning and also wrong, as anyone who has seen an Australian black swan knows. Induction is the only way to learn a rule you were not given, and it can never be made safe. No number of confirming cases turns a probable rule into a certain one.

A third term belongs here, because the most famous fictional “deduction” in English is not one. Abduction — inference to the best explanation — starts from an observation and reaches for the hypothesis that would best account for it. Sherlock Holmes looking at a tan line and a bearing and concluding that a man has been in Afghanistan is not deducing; the evidence is consistent with other stories and he picks the most likely one. It is the reasoning of a diagnosis, not of a proof.

Which one do IQ tests measure? Almost entirely induction

This is the part that no page-one result for the query answers, and it is the part an intelligence site is best placed to say.

The most widely used non-verbal test of reasoning is the Raven Progressive Matrices: a grid of figures changing according to some rule, with one cell missing, and the task of working out which candidate belongs there. Nobody tells you the rule. You have to infer it from the cases in front of you and then apply it. That is induction, in the textbook sense.

In 1990 Carpenter, Just and Shell took the test apart, using verbal protocols, eye-fixation patterns and error analysis, and built computer models that performed like the median and the best college students in their sample. Their conclusion about what separates high scorers from low scorers is worth quoting in its own terms: the distinguishing processes are “primarily the ability to induce abstract relations and the ability to dynamically manage a large set of problem-solving goals in working memory”. Not deduction. Induction, plus the working memory to hold several candidate rules at once.

Formal deduction does appear in cognitive testing — syllogisms and conditional-logic items turn up in some batteries, and in graduate and employment admissions tests — but it is nowhere near the centre of gravity. If you have taken a reasoning test and wondered why so much of it was patterns and sequences rather than logic puzzles, this is why. Our guide to how IQ tests work covers what else goes into a battery, and IQ test questions shows the item types.

People are strikingly bad at deduction — and it is not about brains

If deduction is the kind that comes with a guarantee, you would expect people to be good at it. They are not, and the way they fail is specific.

Peter Wason’s 1968 work looked at conditional sentences of the form if P then Q and the inference that not-P follows from not-Q. That step turns out to be very hard. Wason argued the difficulty came from a mental set for expecting sentences to match states of affairs, and reported that two kinds of “therapy” designed to break that set failed to produce the inference.

Three years later the result that matters most arrived. Wason and Shapiro gave the same logical problem to two groups: ten of sixteen people solved it when it was dressed in familiar, “thematic” content, against two of sixteen when it was abstract. Same logic, same structure, different subject matter, and performance went from one in eight to five in eight. Later work in this tradition pushed the point further: reasoning about permissions and obligations — situations with a social shape — is far easier than reasoning about arbitrary letters and numbers.

The implication is uncomfortable for the idea of logic as a general mental faculty you either have or lack. What people mostly have is reasoning that works well on content they understand.

When belief and logic disagree, belief usually wins

The other well-documented failure is belief bias: people accept a conclusion that fits what they already think, and reject one that does not, largely regardless of whether the argument leading to it is valid. It was demonstrated in syllogistic reasoning by Evans, Barston and Pollard in 1983 in a paper whose title names the problem exactly — the conflict between logic and belief.

Watch the structure rather than the content and it is obvious: “all A are B; this is an A; therefore this is a B” is valid whether the letters stand for something you find plausible or not. Reading the same argument with familiar words attached, most people judge validity by whether the ending sounds right. Our guide to cognitive bias sets this alongside its relatives.

Does a higher IQ mean better reasoning? Only in places

Here the honest answer is genuinely two-sided, and the two sides come from the same body of work.

Across seven studies reported in 2008, Stanovich and West found that a large number of classic thinking biases are uncorrelated with cognitive ability: the conjunction effect, framing effects, anchoring, outcome bias, base-rate neglect, “less is more” effects, omission bias, myside bias, the sunk-cost effect and certainty effects that violate the axioms of expected utility. Being measurably smarter did not protect people from them.

In the same programme, however, cognitive ability did correlate with avoiding a shorter list: denominator neglect, probability matching rather than maximising, belief bias, and matching bias on the four-card selection task — which is to say, on exactly the deduction problems described above.

So the defensible summary is narrow and useful. A higher reasoning score does predict handling the classic deductive traps better. It predicts very little about the judgement errors people make when deciding, estimating and arguing. Our page on the Dunning-Kruger effect deals with a neighbouring question — how well people judge their own performance — and critical thinking covers the evaluation of arguments as a skill in its own right.

Can you get better at either one?

The pessimistic view for most of the twentieth century was that people hold only domain-specific rules, so training in one area would not generalise, or that abstract rules are acquired by self-discovery and cannot be taught at all. Research from the late 1980s onward is more optimistic: Nisbett and colleagues reported in Science in 1987 that even brief formal training in inferential rules can improve how people reason about everyday events, and suggested the earlier pessimism came partly from misidentifying which rules people use naturally.

The most interesting result is that the two halves are trained by different subjects. In a 1990 longitudinal study of undergraduates, Lehman and Nisbett found that social-science training produced large effects on statistical and methodological reasoning — the inductive side, generalising properly from evidence — while natural-science and humanities training produced large effects on reasoning about problems in conditional logic, with the gain among natural-science students appearing to come in large part from mathematics courses. Social-science training did not produce the conditional-logic gain; natural-science and humanities training produced only smaller effects on statistical reasoning.

If you want to argue better from evidence, that points at methods and statistics. If you want to stop being caught by conditional logic, it points at mathematics. Neither points at generic brain training — whether IQ itself can be raised is a separate and much weaker story.

What to take from this

Deduction and induction are not a ranking. They answer different questions and fail in different ways, and the one your reasoning score mostly reflects is induction: spotting the rule behind the cases. The deductive side is harder than its reputation suggests, improves sharply when the material is familiar, loses to prior belief more often than people expect, and is one of the few places where a higher measured ability reliably shows up.

Trust and scope notes

This page explains published research; it is not a diagnostic tool. Our IQ test measures reasoning of the inductive kind described above — finding the rule that governs a set of items — along with other reasoning domains. It does not produce a separate deductive-logic score, and it cannot tell you how you would perform on a formal logic examination.

IQ Revealed is not affiliated with, endorsed by or connected to the publishers of the Raven Progressive Matrices, nor with any of the journals, universities or researchers named on this page. Those names are used only to identify the work being described.