Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Biostatistics / Sampling and estimation

Sampling variation, standard error and the central limit theorem

13 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Explain what a sampling distribution is and why it matters

  • Calculate and interpret a standard error of a mean and of a proportion

  • State the central limit theorem and why it justifies many methods

Imagine drawing a sample of 100 adults, measuring systolic blood pressure (SBP) and calculating the mean. Draw another sample of 100 and the mean will differ a little. Repeat thousands of times: the collection of those sample means is the sampling distribution of the mean. Its centre is the true population mean; its spread tells us how much any one sample mean is likely to miss the truth.

The standard deviation of the sampling distribution is called the standard error (SE).

SD describes people. SE describes how precisely we have estimated a statistic.

The formulas

Standard errors

Mean: SE = (s)/(sqrt(n)) where s is the sample SD.

Proportion: SE = sqrt((p(1 - p))/(n))

Difference between two independent means: SE = sqrt((s_1^2)/(n_1) + (s_2^2)/(n_2))

Worked example: how SE shrinks

SBP in adults has SD = 15 mmHg. The SE of the sample mean is:

  • n = 25: 15 / 5 = 3.0 mmHg
  • n = 100: 15 / 10 = 1.5 mmHg
  • n = 400: 15 / 20 = 0.75 mmHg

Quadrupling the sample size only halves the SE, because of the square root. Going from 100 to 400 participants buys a halving of the uncertainty at four times the cost, which is why sample size planning always asks how much precision is worth paying for.

0123Standard error (mmHg), SD = 153n = 251.5n = 1000.75n = 4000.38n = 1,600
Diminishing returns: precision improves with the square root of sample size.

The central limit theorem (CLT)

The central limit theorem is the most useful result in statistics. It says that, for reasonably large samples, the sampling distribution of a mean is approximately normal, whatever the shape of the original data, with mean equal to the population mean and SD equal to σ / sqrt(n).

This is why we can use normal-based confidence intervals and tests for means of skewed variables such as hospital stay, provided the sample is not tiny. A rough rule: n ≥ 30 is often adequate for moderately skewed data; strongly skewed data (income, viral load) need more, or a log transformation. For proportions the normal approximation works well when both np and n(1-p) are at least about 10.

95% of sample means115.5117118.5120121.5123124.5Sample mean SBP (mmHg), n = 100
The sampling distribution of the mean for n = 100 is centred on the true mean and has SE 1.5. About 95% of sample means fall between 117.1 and 122.9.

Two things the CLT does not do

It does not make the data normal; individual observations stay as skewed as before. And it does not repair bias: a sampling distribution centred on the wrong value is still wrong.

Standard deviation or standard error: which to report?

  • Use SD to describe the spread of the observations in your sample ("mean age 34.2, SD 8.1 years").
  • Use SE (or, better, a confidence interval) to express uncertainty in an estimate ("mean difference 4.2 mmHg, 95% CI 1.1 to 7.3").

Journals warn against presenting "mean ± SE" as a description of the data, because the SE looks deceptively small and shrinks as n grows, while the spread of people does not.

Finite populations

The formulas assume the sample is small compared with the population. If you sample more than about 5% of a small population (say 60 of 200 health workers), apply a finite population correction sqrt(1 - n/N), which narrows the SE.

Summary

  • The sampling distribution is how a statistic would vary over repeated samples; its SD is the standard error.

  • SE of a mean = s/√n; SE of a proportion = √(p(1 − p)/n).

  • Quadrupling n halves the SE.

  • The central limit theorem makes sample means approximately normal for moderate to large n.

  • Use SD to describe people and SE or a confidence interval to describe an estimate.

Check your understanding

Question 1
1

The SE of a mean from n = 64 with SD = 16 is:

Question 2
2

To halve the standard error you must:

Question 3
3

The central limit theorem says that sample means:

Question 4
4

A table of baseline characteristics should describe participants with:

Question 5
5

A proportion of 0.20 from n = 100 has SE:

Question 6
6

Which does a larger sample NOT improve?

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

Populations, samples and the ways studies go wrongConfidence intervals for means and proportions