Key takeaways
- An inductive argument supports its conclusion to some degree rather than guaranteeing it, so it is judged on strength along a scale instead of being valid or invalid.
- Induction is defeasible: new evidence can weaken an argument that was previously strong, which never happens to a valid deduction.
- Four shapes cover most of it — generalising from a sample, applying a general rate to an individual case, inferring a cause, and arguing from an analogy.
- Strength rises with sample size, representativeness and variety of cases, and falls as the conclusion claims more than the evidence covers.
- The same finite evidence always fits incompatible generalisations, which is why a rule that holds for five terms of a sequence can still fail at the sixth.
- Every number sequence puzzle is an inductive inference, and the intended answer is a convention about simplicity rather than a fact the terms establish.
An inductive argument does not promise its conclusion. It supports it to some degree, and that degree is the whole subject. Where a deduction is either valid or not, an induction sits somewhere on a scale between useless and overwhelming, and the same argument can slide down that scale when new evidence arrives. The Stanford Encyclopedia of Philosophy's entry on inductive logic draws the line precisely: in a good deductive argument the truth of the premises guarantees the truth of the conclusion, while in a good inductive argument the premises support the conclusion to some appropriate degree.
That gap never closes, no matter how much evidence accumulates. Understanding why is more useful than any list of examples, so this article works up to it rather than starting there.
Strength, not validity
Three features distinguish induction from the deductive reasoning covered in our companion article on deductive arguments.
It is graded. "Every swan I have seen was white, so the next one will be white" is stronger after a thousand sightings than after three, and neither version is ever complete.
It is defeasible. A strong inductive argument can be weakened by information that leaves its original premises untouched. Learning that all thousand sightings happened in one county does not make any of them false; it makes the generalisation worse.
It adds content. A deduction only unpacks what the premises already contained. An induction says something the premises do not, which is exactly why it can be wrong — and exactly why it is worth making.
The four shapes most inductions take
Almost every everyday induction is one of these.
Generalising from a sample. You inspect a subset and conclude something about the whole. Suppose you check forty items from a delivery of nine hundred and find two faulty. Concluding that roughly five per cent of the delivery is faulty is an induction, and its strength depends on how the forty were chosen.
Applying a rate to an individual. The reverse move: most flights on this route arrive on time, so this flight will probably arrive on time. The conclusion is about one case, and the support is statistical.
Inferring a cause. The heating comes on and the pipes knock; it happens every morning; you conclude the heating causes the knocking. Repetition is the evidence, and a common third cause is the standing risk.
Arguing from an analogy. Two situations share several relevant features, so a further feature is projected from one to the other. Analogical arguments have their own evaluation criteria, which we treat separately in analogy questions.
What makes one strong
Four dials, in rough order of usefulness:
- Sample size. More cases, more support. The relationship is not linear — the gain from ten to a hundred cases is larger than from a hundred to a thousand — but the direction is reliable.
- Representativeness. A sample that resembles the population beats a bigger sample that does not. This is why official statistics agencies spend more effort on sampling design than on sample size. The United States Census Bureau describes its own work in exactly these terms: survey sampling exists to produce timely and cost-efficient estimates of population characteristics, using design-based and model-based estimation to draw reliable inferences about a wider population from sampled data.
- Variety. Cases that differ from one another in irrelevant ways make a hidden common factor less likely. Forty faulty checks from forty different pallets say more than forty from one.
- Scope of the conclusion. The cheapest way to strengthen any induction is to claim less. "Most of this delivery is sound" is far better supported by the same evidence than "every remaining item is sound".
The fourth dial deserves emphasis because it is the one people forget. A conclusion that reaches past its evidence is not a bolder version of the same argument; it is a weaker argument.
The leap, stated plainly
Here is the part that cannot be engineered away.
David Hume's argument, set out in the Stanford Encyclopedia's entry on the problem of induction, is that inferences from observed cases to unobserved ones have no non-circular justification: a demonstrative argument cannot establish a contingent conclusion, and a probable argument would assume the very principle it is supposed to justify.
Nelson Goodman then sharpened the problem into something more immediately practical. His example asks you to consider a predicate that means green if observed before a certain time and blue otherwise. Every green emerald observed so far supports both "all emeralds are green" and the rival generalisation built on that odd predicate — and the two make opposite predictions about emeralds examined later. The same evidence is compatible with incompatible generalisations. Something other than the observations is doing the work of choosing between them.
You do not need to settle the philosophy to use the lesson. Evidence never picks its own generalisation. A preference for the simpler rule is a convention we bring to the evidence, and it is usually a good convention — but it is not a proof.
Every sequence puzzle is an induction
This is where the abstraction becomes concrete, and it is why this article sits in a puzzles hub.
A sequence puzzle gives you finitely many terms and asks for the next. That is an inductive inference in miniature: the terms are the evidence, the rule is the generalisation, and no finite list of terms picks out exactly one rule.
The classic demonstration is geometric. Mark points around a circle, join every pair of points with a straight line, arrange the points so that no three of those lines meet at a single interior point, and count the regions the circle is divided into. For one point through five, the counts run 1, 2, 4, 8, 16. Doubling is the obvious rule and it has held for five terms. The sixth count is 31, not 32. The On-Line Encyclopedia of Integer Sequences catalogues this as sequence A000127, continuing 57, 99, 163 — nothing like powers of two.
Five confirmations were not enough, and no number of confirmations would have been, because the doubling rule was never the rule. That is the honest basis for every claim made in this hub about finding patterns: the procedure in our guide to number sequence puzzles tells you how to find the simplest rule consistent with the terms, and "simplest consistent rule" is a convention that makes puzzles answerable, not a guarantee that the setter's rule is the true one.
Practising the habit rather than the theory
The transferable skill here is small and specific: before committing to a rule, ask what else would fit, and ask what you would expect to see if the rival rule were right.
Our own app, IQ Test & Brain Training Games, has logic games built on analogies, sequences, conditional reasoning, deduction and patterns, and an IQ-style test that covers logic and numerical reasoning; its listing describes the test's results as estimates based on app activities, not clinical or professional assessments. It also ships a Number Pattern Challenge mode that generates fresh rule-finding items at three difficulty settings, which is useful here for one reason: because the items are generated rather than drawn from a fixed list, you cannot recognise them, so you have to run the inference each time. The app records accuracy and response time on that mode, which are game statistics about your play rather than measurements of anything about you. What it tracks is set out on the IQ Test & Brain Training Games app page, and our honest account of what practice apps in this category can support is in do brain training apps work.
Two related skills are worth reading next. Judging whether a number is roughly the right size before trusting it is covered in estimation techniques, and the specific traps in "if, then" statements are in conditional reasoning. The full set sits in the logic and pattern puzzles hub and the wider education and brain training section.
Good to know
Frequently asked questions
What is the difference between inductive and deductive reasoning?
What makes an inductive argument strong?
Can a pattern that holds several times still be wrong?
Where does induction show up in daily life?
Sources
- Inductive Logic (opens in a new tab)Stanford Encyclopedia of Philosophy — accessed
- The Problem of Induction (opens in a new tab)Stanford Encyclopedia of Philosophy — accessed
- Sampling Estimation and Survey Inference (opens in a new tab)United States Census Bureau — accessed
- A000127: Maximal number of regions obtained by joining n points around a circle by straight lines (opens in a new tab)The On-Line Encyclopedia of Integer Sequences, OEIS Foundation — accessed
About the author
Reign Creative LLC — The independent app studio that builds the apps covered on this site and publishes every article here. On Google Play it publishes as Reign Collective Apps.
We use AI tools to help research, draft and edit our articles. Each article lists the sources its factual claims rely on, and details about our apps come from their live Google Play listings. No article on this site has been reviewed by an independent expert unless the reviewer is named on it.
Spotted an error? Tell us — how corrections are handled is set out in our editorial policy.