BOSTON - Science is less a single procedure than a habit of doubt, practiced patiently and in public. It begins, almost always, with an observation that will not sit still: a pattern noticed, an anomaly that resists the standing explanation, a question that ordinary experience cannot answer on its own. The habit consists of what comes next, a disciplined refusal to let a hunch stand in for an answer until it has been tested against the world.
From Hunch to Hypothesis
The first formal step is the hypothesis, a proposed explanation specific enough to generate a prediction that could, in principle, turn out to be wrong. This last clause matters more than any other in the scientific vocabulary. A hypothesis earns the name only if some observation could contradict it. An explanation compatible with every possible outcome explains nothing at all. Researchers describe this property as falsifiability, and they treat it as the boundary line between a scientific claim and a merely appealing one.
A good hypothesis leads to an experiment, a deliberately constructed situation designed to expose the hypothesis to the risk of failure. Here the humble control group does much of the intellectual labor. By holding every plausible variable steady except the one under study, and by comparing a treated group against an untreated or differently treated one, researchers try to isolate a single cause from the tangle of coincidence. Randomization, blinding, and careful sample selection exist for the same reason: to keep the experimenter's own expectations from quietly shaping the result.
Measurement then supplies the raw material of judgment, and here science asks for a kind of honesty that ordinary conversation rarely demands. Every instrument has limits, and every measurement carries some margin of uncertainty. Reputable results are reported with that margin attached, whether as a range, a margin of error, or a stated confidence level, rather than as a bare, falsely tidy number. A measurement without a stated uncertainty is not a more confident claim. It is an incomplete one.
Why One Result Is Never Enough
A single striking experiment, however carefully run, settles very little on its own. Chance alone guarantees that some studies will produce impressive results that later prove impossible to reproduce. Replication, the repetition of a study by the same or by different researchers to see whether the result holds, is the mechanism by which science separates durable findings from statistical noise or simple error. A result that cannot be reproduced elsewhere is treated, provisionally, as not yet established, however compelling it seemed the first time.
Before most findings reach a wider audience, they generally pass through peer review, in which other specialists examine the methods and reasoning for evident flaws. Peer review is a filter, not a guarantee. It catches many errors and misses others, and it settles no argument by itself. Its purpose is more modest: to ensure that a claim has at least survived informed scrutiny before it enters the broader, slower conversation that will continue to test it.
That broader conversation never really closes. Scientific knowledge is provisional by design, held with a confidence proportional to the evidence behind it and revised, sometimes substantially, when better evidence arrives. This is not a weakness of the method but very nearly its definition. A body of knowledge that could never be revised would not be responding to the world at all.
Correlation, Causation, and Honest Doubt
Among the most persistent errors in reasoning from data is the confusion of correlation with causation. Two things may rise and fall together for any number of reasons: one may cause the other, the second may cause the first, some third factor may drive both, or the pattern may simply be coincidence. Establishing which of these is at work generally requires more than an observed association. It requires the kind of controlled comparison, or the accumulation of converging evidence across many independent studies, that allows rival explanations to be set aside one by one.
None of this makes science a source of certainty, and it is not meant to be one. Its distinctive contribution runs the other way: a disciplined, publicly checkable account of how confident one ought to be, and why. A stated uncertainty, an acknowledged limitation, a result described as provisional rather than final: these are not admissions of failure. They are the ordinary vocabulary of a method built to correct itself, and their presence is generally a better sign of rigor than their absence.