What does 'statistically tied' mean in an election survey?

Two candidates are statistically tied when their 95% confidence intervals overlap — the gap between them is smaller than the combined margin of error, so the survey cannot tell who is actually ahead.

Election survey reliability

1,200 voices. 60 million voters. How much can a survey know?

Q: Why can 1,200 respondents represent 60 million voters?

Margin of error depends on sample size, not population, once the population is far larger than the sample. At N=1,200 the 95% margin of error is about ±2.83%.

Each bar is a candidate's surveyed share. The whiskers are the 95% confidence interval — the range the true vote almost certainly falls in. Drag the sample size and watch the whiskers breathe.

Sample size1,200

Margin of error±2.83%

Confidence95%

Top-2 statistical tie?No

0%vote share →50%

Survey design

Sample size · N1,200

±2.83% margin of error · halve the error by quadrupling N

Confidence level95%

z = 1.96 · higher confidence → wider intervals

How to read this

Drag sample size (400–5,000) and confidence level (90–99%). Each candidate's bar redraws with its confidence interval as whiskers. When the top two intervals overlap, the survey flags a statistical tie — it cannot tell who is really ahead.

Try the presets: SWS and Pulse Asia field roughly 1,200–3,000 respondents to speak for ~60M registered voters.

Why N=1,200 represents 60M voters

Margin of error depends on sample size, not population — once the population is far larger than the sample, the population barely enters the formula at all. The margin of error is MoE = z·√(p(1−p)/n).

At N=1,200, p=50%, 95% confidence: MoE ≈ ±2.83%. Doubling to 2,400 cuts it to ≈±2.0% — a reduction of exactly √2. Doubling again to 4,800 reaches ≈±1.4%. Each halving of error costs four times the sample. Diminishing returns set in fast, which is why pollsters stop near 1,200–3,000.

What "statistical tie" means

If candidate A leads B by less than 2× the MoE, their confidence intervals overlap and you cannot statistically distinguish them — the survey reports them as tied.

A 3-point lead with a ±2.5% MoE is not a tie: 3 > 2×2.5 is false, but the intervals only just touch — the lead is real but fragile. Media coverage frequently gets this boundary wrong in both directions.

What MoE doesn't capture

The margin of error only quantifies random sampling error. It says nothing about sampling bias, non-response bias, social-desirability bias, or late deciders.

The 2016 US polls carried ~2% MoE. Their final error was ~4% — the systematic bias was double the sampling noise the MoE measured. A tight margin of error is not the same thing as an accurate poll.

Citations

Cochran (1977) — Sampling Techniques, 3rd ed., Wiley. MoE = z·√(p(1-p)/n).
Pulse Asia / SWS methodology notes — N=1,200 stratified random sample, ±3% MoE @ 95%.
AAPOR Standards (2016) — survey reporting transparency.
Kennedy et al. (2018) — Pew, on 2016 US poll error post-mortem.
Groves (2006) — total survey error framework, Public Opin. Q. 70, 646.

Built by Joshua Chua · Fable rebuild · vanilla JS + Canvas