A single number is a claim about precision it usually cannot support

Estimates aren’t measurements

When a study reports that something increased by a certain amount, it is not reporting a fact about the world in the way a ruler reports a length. It is reporting the value that best fits one particular sample, drawn from a population, with all the accidents that sampling involves. Another sample from the same population would have produced a different number.

The interval reported alongside the estimate is an attempt to describe how much that number would move under repetition. It is the difference between "our best guess is this" and "our best guess is this, and here is the range of values the data can’t rule out". The second is a result. The first is a headline.

What the interval actually means

The technical definition is awkward and worth stating properly, because the intuitive reading is wrong. A ninety-five per cent confidence interval is constructed by a procedure which, if repeated across many samples, would produce intervals containing the true value that proportion of the time. It is a statement about the reliability of the method, not a probability that this particular interval contains the answer.

In practice most readers use it as a rough range of plausible values, and that reading is serviceable for everyday purposes provided one thing is kept in mind: the interval only captures uncertainty from sampling. It says nothing about whether the measurement was any good, whether the sample was representative, or whether the analysis was chosen after the fact. Those sources of error are usually larger and never appear in the range.

Wide intervals are the most common finding

A great deal of published research produces intervals so wide that they include effects worth acting on and effects too small to notice. This is not a defect in the reporting; it is an honest reflection of what a study of that size can resolve. The defect appears when the point estimate is quoted alone, which converts an inconclusive result into a definite one.

The habit that helps is to read the ends of the interval rather than the middle. Ask what you would do if the true value were at the lower end, and what you would do if it were at the upper end. If the answer is the same in both cases, the study has settled something. If the answer differs, the study has narrowed the question without answering it, and any confident summary of it’s overstating.

Precision and importance are different axes

A result can be precisely estimated and trivially small, which is what happens when a very large sample measures a real but negligible effect. It can also be large and hopelessly imprecise, which is what happens when a small study finds something dramatic. Reporting conventions tend to reward the second and ignore the first, since a large number is more quotable than a narrow one.

Judging importance requires a scale outside the statistics: how much of a change would matter, in the units people actually care about. That threshold has to come from the subject matter, and it should ideally be stated before the result is seen. Without it, any effect can be described as meaningful, and usually is.

Why more precision isn’t always the goal

It is tempting to conclude that the aim of research is ever-narrower intervals, but precision about the wrong quantity is worse than vagueness about the right one. A tightly estimated effect in a sample that does not resemble the people you care about tells you very little, and the narrowness of the interval can make it more persuasive than it deserves to be.

This is the trade every study makes. Tight control gives precision and narrows who the finding applies to; realistic conditions give relevance and widen the range. Neither is a mistake, and a reader’s job is to notice which one was bought and at what cost.

A reading routine

Find the estimate. Find the range. Ask whether the two ends of the range would lead to different decisions. Then ask what sources of error the range does not include, which is nearly always the more important question and nearly always the one nobody asks.

None of this requires statistical training. It requires the habit of refusing to read a single number as a fact, which is a small act of resistance against how results are presented almost everywhere.