Last updated: 2026-06-23

Why sample size matters for a mean

When a survey or experiment aims to estimate an average — average wait time, average blood pressure, average household spend — the sample size decides how precise that estimate can be. Too few subjects and the confidence interval is so wide the result says little; too many and the study wastes time and money. Planning the sample size in advance fixes the precision you want before a single observation is collected.

The required sample size for estimating a population mean follows directly from the width of a confidence interval:

Each term has a plain meaning:

• $n$ — the minimum number of observations needed.
• $z$ — the standard-normal critical value for the confidence level. It is 1.645 for 90%, 1.96 for 95%, and 2.576 for 99%.
• $\sigma$ — the population standard deviation, a measure of how spread out the values are.
• $E$ — the margin of error, the ± precision you want around the estimated mean, in the same units as $\sigma$.

Precision tightens with the square of the margin: halving the margin of error quadruples the sample. Likewise, doubling the spread $\sigma$ quadruples the sample, and moving from 95% to 99% confidence raises the critical value from 1.96 to 2.576, multiplying the requirement by roughly 1.7.

Getting a value for σ

The formula needs the standard deviation, yet $\sigma$ is almost never known before the study runs. Three practical estimates cover most cases:

• Pilot study. Collect a small batch of observations and use their sample standard deviation as a stand-in for $\sigma$.
• Prior data. Reuse the standard deviation reported in earlier research or published statistics for a comparable population.
• Range rule. When you only know the plausible smallest and largest values, approximate $\sigma \approx (\text{max} - \text{min}) / 4$, since most of a roughly normal range spans about four standard deviations.

A slightly conservative (larger) estimate is the safe choice: it pushes the sample up rather than leaving the study underpowered.

Worked example

Suppose you want to estimate a mean within ±2 units at 95% confidence, and prior data suggest a standard deviation of about 15. With a critical value of 1.96, $\sigma = 15$, and a margin of error of 2:

The exact formula returns 216.09, but you cannot survey a fraction of a subject, and rounding down would leave the margin slightly wider than requested. The result is always rounded up — here to 217 observations.

When z is not enough

The formula above uses the normal value $z$, which is the standard convention for sample-size planning. Two caveats are worth keeping in mind.

Small samples and unknown σ. The $z$ value assumes a large sample. When the final dataset is small (roughly under thirty) and $\sigma$ is estimated from the data, the analysis itself should use the wider $t$-distribution instead of $z$. For planning, though, $z$ is the usual starting point because you do not yet know $n$; once $n$ is large the gap between $z$ and $t$ becomes negligible.

Finite populations. The plain formula assumes sampling from an effectively infinite population. If the sample is a sizeable fraction of a small finite population of size $N$, apply the finite population correction, which lowers the required sample:

For estimating a proportion rather than a mean, the planning formula changes shape — see the Sample Size Calculator (Proportion). To work in the other direction and find the margin of error implied by a sample you already have, use the Margin of Error Calculator.