A margin of sampling error describes uncertainty associated with estimating a population's views from a sample under the survey's statistical assumptions. It does not measure every way a poll can be wrong.
A poll can also be affected by who answers, who is missed, the wording of questions, and turnout assumptions. Those problems do not vanish because a release includes a familiar plus-or-minus figure.
The margin usually belongs to an estimate
Suppose a hypothetical poll reports Candidate A at 48% and provides a three-percentage-point margin of sampling error for that estimate. It would be incorrect to assume that a two-point lead over Candidate B has the same three-point uncertainty.
A candidate difference requires its own calculation because the two estimates come from the same survey and are statistically related. Pew explains this distinction in its discussion of election poll margins. Source: Pew Research Center.
Why “within the margin” is not a tie declaration
A survey can show a numerical lead that is not precise enough to support a confident claim about the population. That does not prove the candidates have exactly equal support.
The responsible wording describes the uncertainty: the poll shows a small estimated advantage, but the evidence is insufficient to establish a clear lead at the stated confidence level.
That sentence is less dramatic than “the race is tied,” but more accurate.
Check the subgroup size
A poll of 1,000 people may include far fewer respondents in a particular age or regional subgroup. The overall uncertainty figure should not automatically be applied to that smaller group.
When a headline says a candidate surged among young voters, look for the subgroup count and an appropriate uncertainty measure.
Practical reading rule
Ask what estimate the margin covers, what confidence level is used, and whether the survey design supports the reported calculation. A nonprobability sample needs a clearly explained uncertainty approach rather than an unexplained conventional claim.
Use margins to temper conclusions, not to dismiss every poll or declare every close race unknowable.
Related reading: small polling leads and subgroup preferences.