Skip to content
PMMilestoneCalculators
Math

Sample size and confidence intervals without the jargon

Separate sample planning from interval estimation, and make the assumptions visible before reporting a margin of error.

By Last updated

A sample estimate with a confidence interval and a finite population sample planning table

Start with the quantity you want to estimate

“Enough responses” is not a statistical specification. First identify the population, the measurement and the precision needed for a decision. Estimating the share of customers who select an option is a proportion problem. Estimating average delivery time is a mean problem. Their planning inputs and interval formulas differ.

Sample size planning asks how much information to collect before the study. A confidence interval describes uncertainty after collecting the data, under a stated statistical model. Neither procedure repairs a biased recruitment process or an ambiguous question.

Use the sample size calculator for planning and the confidence interval calculator for estimation. Before entering values, identify whether the calculation concerns a proportion or a mean, and which method the tool implements.

Put precision in operational terms

For a proportion, a margin of error of 0.05 means five percentage points, not five percent of the observed percentage. An estimate of 40% with that margin spans 35% to 45%. That width may be acceptable for an exploratory decision and inadequate for a narrow compliance threshold.

Confidence level and precision are different controls. A higher confidence level requires a wider interval for the same data, or a larger planned sample for the same margin. A larger sample usually narrows random sampling uncertainty, but does not eliminate measurement error, coverage gaps or nonresponse bias.

Write the intended reporting statement before collecting responses: which population, which outcome, which confidence level, and what maximum margin? This catches requests for precise subgroup results that a total-sample calculation cannot support.

Plan a proportion sample using an explicit normal approximation

For the following planning calculation, use the two-sided normal z approximation, also called normal-approximation proportion planning. This is not a Wilson interval calculation. Assume a simple random sample from a large population, independent observations to a suitable approximation, and a binary outcome.

When there is no defensible preliminary proportion, use p = 0.5. This maximizes p(1 − p), making the planned sample conservative within this formula. It does not guarantee coverage for every eventual dataset or sampling design.

n₀ = z² × p(1 − p) / E²
n = ceiling(n₀)

For 95% confidence, take z = 1.95996398454, p = 0.5 and E = 0.05:

  • p(1 − p) = 0.25.
  • E² = 0.0025.
  • n₀ = 1.95996398454² × 0.25 / 0.0025 = approximately 384.145882.
  • Round upward to obtain 385 completed observations.

Rounding down would miss the requested planning threshold. The count refers to usable observations, not invitations, page views or partially completed records.

Tip: Keep the unrounded result for later adjustments. Rounding at every intermediate step makes independently produced plans harder to reconcile.

Correct for a small, known finite population

If the entire target population contains 1,000 units and you sample without replacement by simple random sampling, collecting hundreds of observations covers a meaningful share of the population. A finite population correction reduces the required sample for the same normal-approximation planning target.

nF = n₀ / (1 + (n₀ − 1) / N)
Required completed sample = ceiling(nF)

With N = 1,000 and the unrounded n₀ above, nF is approximately 277.733453. Round upward to 278 completed observations.

Planning caseConfidenceMarginCompleted sample
Large population, p = 0.595%5 percentage points385
Population 1,000, without replacement95%5 percentage points278

Use a defensible population count, such as a complete list of eligible accounts at a specified date. Do not enter the size of a convenient email list if the intended claim concerns all potential customers. The correction applies to the defined finite population; it does not turn convenience responses into a probability sample.

Invitations and subgroups need separate planning

Suppose the finite-population plan requires 278 usable responses. Under an explicitly hypothetical 50% completion assumption, inviting 556 eligible units gives an expected 278 completions. That is arithmetic, not a guarantee or a claim about typical survey behavior.

If completion differs by customer type, the responding sample may still be distorted. Track who was eligible, invited and retained. Increasing invitations can help reach a count without curing systematic nonresponse.

Similarly, a sample adequate for an overall proportion may leave very small subgroup counts. If the decision requires comparisons between regions, plan those estimates separately. Cluster sampling, weighting and repeated observations introduce design considerations that the simple random sample formula does not capture.

Calculate a mean interval without confusing spread and precision

Now switch questions. Imagine a hypothetical random sample of 100 independent measurements, with sample mean 50, and a genuinely known population standard deviation σ = 10. Assume a normal population, or conditions making the sampling distribution of the mean sufficiently close to normal.

Because σ is known in this example, use a two-sided normal z interval for the mean. The standard error is 10 / √100 = 1. Standard deviation describes the spread of individual measurements; standard error describes the sampling variability of the mean.

Mean interval = sample mean ± z × σ / √n
Mean interval = 50 ± 1.95996398454 × 1

The endpoints are 48.04003601546 and 51.95996398454, reported to two decimals as 48.04 to 51.96. The margin is approximately 1.96 measurement units, not 1.96% unless the measurement scale itself warrants that interpretation.

When the population standard deviation is unknown and 10 is instead the sample standard deviation, use an appropriate Student t interval, not the known-σ z calculation. With 100 observations, that conventional t interval uses 99 degrees of freedom and is slightly wider. The standard deviation calculator helps distinguish the sample and population spread calculations; it cannot establish that an unknown population σ is known.

Interpret confidence as a property of the method

A 95% confidence procedure is designed so that, across repeated samples meeting its assumptions, about 95% of the intervals contain the fixed population parameter. Once an interval has been calculated, the parameter is not moving randomly between its endpoints under this frequentist interpretation.

Avoid saying that 95% of individual measurements lie inside the mean interval. A confidence interval for the mean is not a prediction interval for the next observation, nor a range covering most observations. Individual measurements can be much more dispersed than their estimated average.

Also avoid equating a narrow interval with practical importance. A precisely estimated difference can be operationally trivial, while a wide interval can contain both valuable and unacceptable outcomes. Compare the interval with a decision threshold defined before seeing the result.

Keep proportion interval methods separate

The normal planning formula above is not an instruction to report every observed proportion with a symmetric Wald interval. Wald intervals use the observed proportion and a normal standard error; they can perform poorly with small samples or proportions near zero or one.

Wilson score intervals are a different method and often a better routine choice for reporting a single binomial proportion. Exact binomial intervals are another option, with different coverage and width behavior. Neither has been numerically substituted into the z-planning examples in this guide.

If a calculator offers Wilson, Wald or exact choices, name the selected method beside the result. For a finite-population sample, also check whether the estimation method accounts for sampling without replacement; a plain binomial interval does not automatically include that adjustment.

Record what the reader needs to audit

A useful report includes the population definition, recruitment method, usable sample count, estimate, confidence level, interval method and relevant exclusions. Record whether a finite population correction or survey design adjustment was applied. For means, state whether the spread input was a known population standard deviation or an estimate from the sample.

Tip: Save inputs and unrounded calculations with the analysis, then round only the displayed endpoints to a precision justified by the measurement. Extra decimal places cannot compensate for weak sampling.

Before acting, check missing responses, duplicates, impossible values and unusual observations against a documented rule. Do not discard inconvenient records merely to narrow an interval. The most useful uncertainty statement is one another analyst can reproduce and a decision-maker can interpret without guessing the method.

Calculators used in this guide

Frequently asked questions

Why does the example need 385 rather than 384 responses?

The unrounded normal-approximation planning result is about 384.145882. Rounding upward gives the smallest integer meeting that formula’s target. Rounding down would slightly exceed the requested margin.

When does the population of 1,000 reduce the target to 278?

The correction assumes a simple random sample without replacement from that defined finite population. It does not apply merely because a mailing list contains 1,000 addresses or because respondents volunteered.

Does 95% confidence mean the interval contains 95% of observations?

No. A confidence interval estimates a population parameter, such as a mean. An interval for individual observations answers a different question and generally has a different width.

Should the mean example use z or t?

It uses z because the population standard deviation is explicitly assumed known as 10. If 10 is a sample standard deviation estimating an unknown population value, the conventional mean interval uses Student t under the appropriate assumptions.

Are the proportion examples Wilson intervals?

No. The sample counts use a normal z planning formula with p equal to 0.5. Wilson is a separate method for estimating an interval from observed binomial data; its endpoints have not been calculated in these examples.

Sources and standards

About the author

Dr. Hassan Khames Eliwa holds a PhD in Construction & Built Environment and has more than 20 years of international project controls experience. He is Director of Eliwa Project Controls Limited in Auckland and founder of PMMilestone.org and PMMilestone.com.

Buy me a coffee