Induction and the Weight of Evidence

Almost everything you know about the world beyond this moment rests on induction. You expect the sun to rise, bread to nourish you, and a dropped cup to fall, not because logic forces these conclusions but because they have held so far. Science, medicine, and daily planning all run on the same move: from what has been observed to what is likely true in general.

Because induction is so useful, it is easy to forget that it is a gamble. A conclusion drawn from evidence can be well supported and still turn out false. Learning to weigh that evidence — and to see where the weighing can fail — is the difference between reasoning carefully and merely trusting the past.

From the particular to the general

Inductive reasoning generalises. You observe many instances — swans, sunrises, patients — and infer something broader: a rule, a probability, a prediction about the next case. Unlike deduction, induction adds content. The conclusion says more than the premises strictly contain, which is exactly what makes it powerful and exactly what makes it fallible.

Strength versus validity

Deductive arguments are valid or invalid: if the premises are true, a valid conclusion must be true. Inductive arguments are never valid in that sense; they are strong or weak. A strong inductive argument makes its conclusion probable, not certain. Ten thousand white swans make "all swans are white" likely, not guaranteed — and one black swan overturns it. Strength is a matter of degree, and more evidence can raise or lower it.

Sample size and representativeness

Two things govern an inductive argument's strength. Sample size: a handful of cases supports far less than thousands. And representativeness: the sample must resemble the whole it stands for. A survey of a million people still misleads if they were all recruited from one city. A large but skewed sample can be worse than a small fair one, because its size lends false confidence.

A hospital reviews 4,000 patients given a new drug and finds recovery times shorter than in a comparable untreated group. The inductive conclusion — the drug tends to speed recovery — is reasonably strong: the sample is large, and if patients were assigned fairly, it is representative. But the argument is not airtight. The next 4,000 patients might differ, or an unnoticed factor might explain the result. The evidence raises the probability; it does not close the case.

Suppose a fund manager beats the market for seven straight years. Inductively, one might conclude she has genuine skill and will keep winning. Yet with thousands of managers, some streaks are expected by chance alone, and past returns are a notoriously weak guide to future ones. Here the pattern is real but the inductive leap is fragile: the sample of years is small, and the mechanism that produced the streak may not persist. Strong-looking evidence can still support a weak inference.

In The Problems of Philosophy (1912), Bertrand Russell offered a memorable image. A chicken fed every morning comes, quite reasonably, to expect food whenever the farmer appears — until the day the farmer wrings its neck instead. Russell's point was not that induction is worthless but that a run of confirming instances, however long, does not guarantee the next case. "More refined views as to the uniformity of nature," he wrote, would have served the chicken better.

Russell was sharpening a challenge David Hume had posed in the eighteenth century, in A Treatise of Human Nature (1739) and An Enquiry Concerning Human Understanding (1748). Hume asked what justifies inferring the future from the past. The only support seems to be that such inferences have worked before — which is itself an inductive argument, and so assumes the very thing in question. This is the problem of induction: we cannot prove that nature will stay uniform without already relying on the assumption that it will. Induction remains indispensable; Hume simply showed it rests on habit and expectation, not on proof.

• "A strong inductive argument proves its conclusion." It makes the conclusion probable, never certain; strength is degree, not proof.
• "A bigger sample is always better." Size helps only if the sample is representative; a large biased sample can mislead more confidently than a small fair one.
• "If it has always happened, it must keep happening." That is precisely the assumption Hume questioned; past regularity is evidence, not a guarantee.
• "Induction and deduction are just two words for reasoning." Deduction preserves truth; induction extends it beyond the evidence, trading certainty for reach.

Take three claims you hold about how the world usually works, and for each rate the sample size and representativeness of the evidence behind it — then name one thing that could make the pattern break.

Think Like a Maester: Count your evidence and ask who it leaves out, because a pattern that has always held is a promise nature never signed.

1. How does inductive reasoning differ from deductive reasoning?
2. Why are inductive arguments described as strong or weak rather than valid or invalid?
3. What two features most affect the strength of a generalisation from a sample?
4. What was the point of Russell's inductivist chicken?
5. Why does Hume say we cannot justify induction without arguing in a circle?

Induction carries us from specific observations to general and probable conclusions, and nearly all practical knowledge depends on it. Its arguments are graded by strength rather than validity: more, and more representative, evidence makes a conclusion likelier without ever making it certain. Sample size and representativeness are the levers to check first. Russell's chicken and Hume's problem of induction remind us that no run of past confirmations guarantees the next case — so reason from evidence with confidence, but hold that confidence open to revision.

• Deductive vs Inductive Reasoning
• Causal Reasoning