A survey reports that 52% of respondents support a proposal, with a margin of error of three percentage points. The accompanying range is 49% to 55%, at the stated confidence level and under the survey’s assumptions. The figure describes a component of uncertainty, not the maximum possible error in the survey.
The margin of sampling error is commonly reported as the half-width of a symmetric confidence interval. Some intervals are asymmetric, in which case the actual lower and upper bounds communicate the result more accurately than forcing it into a single plus-or-minus number.
Know which uncertainty the number covers
Sampling variation arises because different samples can produce different estimates. A conventional margin of error describes that variation under the sampling design or model. It does not automatically account for missing groups, non-response, misunderstood questions or inaccurate answers.
Pew Research Center’s explanation of polling margins discusses these limits and the effect of survey design. The distinction matters in product research too: a voluntary customer poll should not be given a probability-sample margin merely because the number of replies is known.
If a model-based uncertainty measure is used for a non-probability sample, name it and state the assumptions. The calculation needs a justification beyond resemblance to a familiar polling formula.
Plan precision before collecting responses
Under a simple random-sampling approximation for a proportion near 50%, at 95% confidence and with a large population, 100 observations give a margin of about 9.8 percentage points; 400 give about 4.9 points. Quadrupling the sample roughly halves the margin under those conditions.
These are illustrative calculations, not universal recruitment targets. The estimate, confidence level, weighting, clustering and finite population can all affect the result. When a sample is a substantial share of the population, the finite-population correction may matter; for a small fraction of a very large population, it usually makes little difference.
Work backwards from the decision. A study intended to distinguish a small difference requires more precision than one intended to identify a broad pattern. Increasing sample size can be useful, but it should follow the uncertainty that matters rather than a conventional round number.
Recalculate for the comparison you report
The margin quoted for a full sample does not apply unchanged to a smaller subgroup. If only a fraction of respondents use a particular plan, estimates for that group have their own uncertainty and may also be affected by weighting or the sampling design.
Likewise, a difference between two estimates needs an appropriate analysis of the difference. Overlapping separate intervals do not automatically establish the absence of a statistically detectable difference. Confidence intervals and statistical significance provide related tools, each requiring the correct comparison and assumptions.
Report the estimate, confidence level, sample or effective sample information, and calculation method. Keep non-response bias and other quality concerns visible beside them. The margin is valuable because it describes a particular uncertainty clearly, provided it is not asked to certify the entire study.