Survey margin of error: formula, examples and interpretation checklist

A survey margin of error is a confidence multiplier times an estimated standard error for a named estimate under a stated sampling design. For a simple random sample proportion, an approximate 95% margin is 1.96 × √[p(1−p)/n]. The interval describes sampling variability under the design; it does not include nonresponse, frame gaps, measurement error, processing mistakes or every other source of uncertainty.

What is survey margin of error?

Margin of error is the half-width added to and subtracted from an estimate to form a confidence interval. It is calculated from an estimated standard error and a confidence multiplier. The U.S. Census Bureau describes a 90% margin as 1.645 times the estimated standard error. A two-sided 95% normal-approximation interval uses about 1.96 standard errors.

Margin of error is a post-collection uncertainty measure. It differs from the sample-size calculation used before fieldwork and from a nonresponse-bias assessment, which asks whether responders and nonresponders differ systematically.

What is the formula for a sample proportion?

For an unweighted simple random sample with a sufficiently large sample, SE(p) = √[p(1−p)/n] and MOE = z × SE(p). Here, p is the observed proportion, n is the number of valid independent observations for that estimate, and z is the confidence multiplier. Common normal multipliers are 1.645 for 90%, 1.96 for 95% and 2.576 for 99%.

Worked 95% example

In 800 valid responses, 480 respondents select Yes, so p = 480 ÷ 800 = 0.60. The standard error is √[(0.60 × 0.40) ÷ 800] = 0.01732. The approximate 95% margin is 1.96 × 0.01732 = 0.03395, or 3.4 percentage points. Report the approximate interval as 60.0% ± 3.4 points, from 56.6% to 63.4%.

This does not mean 95% of respondents fall inside the interval. Under repeated samples taken and analyzed with the same design, about 95% of intervals produced by the method would contain the target parameter, subject to the method's assumptions.

Observed proportion or conservative 50%?

The product p(1−p) is largest at p = 0.50. At n = 800, the conservative 95% margin is 1.96 × √(0.25 ÷ 800) = 3.5 points. Use that value for planning when the proportion is unknown. When reporting the observed 60% estimate, calculate uncertainty for that estimate or apply the survey's official design-aware variance method.

When does finite population correction apply?

If a simple random sample is drawn without replacement from a finite eligible population N and the sampling fraction is material, multiply the standard error by the finite population correction √[(N−n)/(N−1)]. With N = 2,000, n = 800 and p = 0.60, the factor is √(1,200 ÷ 1,999) = 0.775. The illustrative margin becomes about 2.6 points. Apply the correction only when N is the actual eligible finite population and the sample was drawn accordingly; it does not repair nonresponse or frame error.

Design-aware decision table

How do you calculate a difference between groups?

Suppose independent group A is 60% of 400 and group B is 52% of 400. The estimated difference is 8 points. Its standard error is √[(0.60 × 0.40 ÷ 400) + (0.52 × 0.48 ÷ 400)] = 0.0350. The approximate 95% margin for the difference is 6.9 points, so the interval is about 1.1 to 14.9 points. Calculate the contrast directly; separate intervals around each proportion do not answer the same question.

Reusable reporting worksheet

Store estimate_name, numerator, valid_n, estimate, population, sampling_frame, sample_design, weight, strata, clusters, variance_method, confidence_level, multiplier, standard_error, margin_of_error, confidence_interval, finite_population, design_effect, effective_sample_size, field_period, nonsampling_limitations, analyst, code_version and verification_date.

Eight interpretation checks

1. State whether the margin is in percentage points or percent of the estimate.
2. Match the denominator to the specific estimate, not the total survey count.
3. Use the actual sample design and analysis weights.
4. Check whether small counts or proportions near zero or one need a non-normal interval.
5. Calculate differences and trends directly.
6. Separate sampling uncertainty from coverage, nonresponse, measurement and processing error.
7. Avoid a conventional sampling MOE for nonprobability samples without a defensible model.
8. Preserve code, design variables and replicate weights needed to reproduce the interval.

Transparent method

The core definition and multiplier relationship follow the current U.S. Census Bureau methodology page. The complex-design and boundary cautions follow current CDC/NCHS reliability guidance, which evaluates effective sample size, design effects, interval width and degrees of freedom. Statistics Canada's current Quality Guidelines, Sixth edition directs producers to use variance estimation that accounts for the sample design and adjustments such as nonresponse, imputation and calibration. The worksheet, examples and release checks are editorial synthesis, not a substitute for the survey's official variance specification.

Limitations

The normal approximation can perform poorly with small samples, rare outcomes or proportions near zero or one. CDC/NCHS notes that the common Wald interval can extend outside the valid 0-to-1 range and can be too narrow, especially near the boundaries. A finite population correction is not a remedy for nonresponse or frame error, and a narrow interval around a biased estimate remains biased. Replicate weights, imputation, multistage selection, calibration and small-area models require specialized methods. Consequential or regulated studies need review by a survey statistician.