A three-point poll lead can still be a tie

If a poll says 51% versus 48% with a plus-or-minus 3 point margin of error, the headline leader may not really be ahead.

Margins of error remind us that poll numbers are estimates. When uncertainty ranges overlap, the apparent ranking can be too close to call.

By the Math Says Yes editorial team

Human-reviewed under our source and correction standards.

How we review content

THE TRAP

Reading the bigger point estimate as a definite lead and ignoring the plus-or-minus uncertainty around both estimates.

Two poll bars with overlapping margin-of-error brackets.

Poll numbers are estimates

A poll asks a of people and uses that sample to estimate what a larger thinks. The reported percentages are not exact measurements of the population. If a different random sample had been drawn, the percentages would usually shift a little. The is the pollster's way of showing the size of that ordinary sampling wobble.

What the Numbers Show

A (51%) could really be at

48%

B (48%) could really be at

51%

Illustrative: a 51–48 result with ±3-point uncertainty can still be tied or reversed.

The tie hidden inside the lead

Suppose one side is at 51% and the other is at 48%, with a of plus or minus 3 percentage points. The first estimate could plausibly be around 48% to 54%, while the second could plausibly be around 45% to 51%. Those ranges overlap. The 's point estimate has a leader, but the uncertainty still leaves room for a practical tie or even a reversed order.

Why headlines overstate it

Rankings make simple stories: ahead, behind, winner, loser. Uncertainty makes messier stories: maybe ahead, maybe tied, maybe too close to call. The simple story is easier to headline, so point estimates get treated like exact scores. That turns noise into drama. The is not a footnote for statisticians; it is part of the result and should shape the headline.

How to use it

When reading a poll, compare the uncertainty ranges, not just the point estimates. Treat small leads inside the as uncertain. Also remember that the margin of error covers only random ; it does not automatically fix nonresponse, bad weighting, question wording, or turnout assumptions. A good poll is evidence, but it is still evidence with a range around it.

Worked example

A poll can report Candidate A at 51% and Candidate B at 48%, a three-point lead. If the uncertainty around each estimate is roughly three points, the headline lead is smaller than the ordinary sampling uncertainty around the estimates. The exact uncertainty around the difference can depend on the poll design, but the practical reading is the same: the point estimate has an order, while the evidence is still too close to call confidently.

When it applies

is useful for random sampling uncertainty, not for every polling problem. Turnout modeling, question wording, late opinion changes, nonresponse, weighting, and mode effects can all move the result. Treat the published margin as one visible piece of uncertainty. A responsible reading combines it with poll quality, trend, composition, and whether other polls show the same direction.

What people get wrong

People read the poll like a scoreboard: 51 beats 48, so one side is winning. Polls are estimates, not final scores. A small lead can be the best single estimate and still not be strong evidence of a real lead once sampling uncertainty and other survey errors are included.

Source note

The Dominitz and Manski source is used because it broadens the discussion from the familiar sampling to total survey error. That is the quality point for this page: a close poll lead can be uncertain even before considering non-, and real polling uncertainty is often larger than the simple number in a graphic.

FAQ

Does margin of error include every possible polling mistake?

No. The usual margin of error reflects random sampling uncertainty. Other errors, such as nonresponse, likely-voter modeling, wording, and weighting, can still matter.