What does a confidence interval’s margin of error include—and what does it leave out?

It describes a particular kind of uncertainty: chance differences between a sample result and the population figure the survey aims to estimate. More precisely, a margin of error belongs to a specified interval-calculation method; it is not a universal measurement of survey quality. One example in OpenStax is for a population mean when the population standard deviation is known: the margin of error equals a critical z-value times the standard error, σ/√n. That method requires a random sample and known population standard deviation, so its formula should not be casually applied to every poll or interval. [S1][S2]

For your hypothetical poll, “48% ± 4 percentage points” gives an interval from 44% to 52% by subtraction and addition. Treat those numbers as an illustration, not a verified result. The margin describes the calculated interval’s sampling precision under its assumptions; it does not certify that the poll was well designed or that 48% is close to the population figure.

No. It is not a guarantee that this particular sample result is within four points of the population figure. The margin describes an interval produced by a method, whose interpretation depends on how it was calculated and the assumptions it uses. Sampling error concerns chance differences between a sample result and the population parameter; it does not make those differences impossible. [S1][S2]

When explaining a reported figure, keep the claim modest: “The poll reports 48%, with a stated margin of error of ±4 percentage points under its calculation method.” Avoid turning that into “The true figure must be between 44% and 52%.” To understand what the interval says, look for the poll’s confidence level, sampling approach, and calculation method—not just the plus-or-minus number.

A margin of error does not, by itself, quantify all the other ways a survey can go wrong. A sampling frame might omit parts of the target population, or a sample might be unrepresentative. A question can also produce inaccurate self-reports, or answers can be recorded or analyzed incorrectly. These are distinct concerns from chance differences caused by drawing a sample. [S2]

Use the poll’s documentation to ask practical questions: Who was eligible? How were people selected and contacted? Could parts of the target population have been left out? What exactly did the question ask? How were answers collected and processed? The margin alone cannot answer those questions, and a general explanation of error types cannot establish whether any particular poll has those problems. Look for survey-specific evidence rather than assuming either that problems occurred or that none did.

You could write:

“Suppose a poll reports 48% with a margin of error of ±4 percentage points. Under the poll’s stated interval method, that gives a range of 44% to 52%. The margin describes sampling uncertainty under that method; it does not measure every possible survey error. To judge the result, also check how the sample was selected and whether the question and data collection could have affected responses.” [S1][S2]

This separates the arithmetic from the interpretation and makes clear that 48% and ±4 are hypothetical. If the poll does not explain its method, confidence level, or sample, say those details are unavailable rather than filling them in. The interval’s width alone cannot establish whether the sample represents the target population or whether answers were measured accurately. [S1][S2]