New: 6 free SQL practice datasets with 300+ questions — try the SQL Compiler →
Statistics

Sampling and the Central Limit Theorem

Sampling methods, the standard error of the mean, and the Central Limit Theorem explained with worked numbers and why it works for any population shape.

Upskly AI Team October 5, 2026 7 min read
Sampling and the Central Limit Theorem

Every statistic calculated from a sample is an estimate, not the truth, and how much that estimate can be trusted depends on how the sample was drawn and how large it is. This covers the four common sampling methods, the standard error of the mean, and the Central Limit Theorem, the result that makes normal-distribution-based methods usable even on data that is not normal at all.

In this guide

Four sampling methods

Four ways to draw a sample
MethodHow it worksGood fit when
Simple randomEvery individual has an equal chance of selectionNo meaningful subgroups to protect
StratifiedPopulation split into subgroups, then sampled from each oneImportant subgroups could otherwise be under-represented by chance
SystematicEvery nth individual on an ordered list, after a random startA practical, quick-to-execute alternative to full random sampling
ClusterWhole groups (clusters) chosen at random, everyone inside measuredThe population is naturally grouped, such as by school or city block

Telling cluster and stratified apart

A national survey randomly selects 20 schools, then surveys every student within each chosen school. This is cluster sampling, not stratified: whole natural groups (the schools) are selected first and then fully measured, rather than individuals being drawn from every group across the population.

Bias, variability and representativeness

Bias is a systematic tendency to miss the true value in one direction. Variability is how much sample results differ from each other purely by chance, with no particular direction. A representative sample is one whose key characteristics, such as its age or income mix, closely mirror the population’s own, which is why a biased selection method can still produce a large sample that is not representative.

True random sampling removes selection bias, but it does not eliminate sampling error entirely. Any sample smaller than the full population will still differ from it somewhat, just by chance rather than by a systematic skew.

The standard error of the mean

The standard error of the mean is the standard deviation of the sampling distribution of the sample mean, how much sample means themselves vary from sample to sample, a different quantity from the spread of individual data points. The formula divides the population standard deviation by the square root of the sample size.

Standard error shrinks with the square root of n, not with n itself
Population SDSample size (n)Standard error
202520 / sqrt(25) = 4.0
1510015 / sqrt(100) = 1.5
8168 / sqrt(16) = 2.0
864 (4x the sample size above)8 / sqrt(64) = 1.0 (half, not a quarter)
30930 / sqrt(9) = 10.0

That last comparison is the key pattern: multiplying the sample size by 4 only cuts the standard error in half, because the formula depends on the square root of n, not n directly. Quadrupling a sample does not quarter the uncertainty in the mean.

The Central Limit Theorem

The Central Limit Theorem states that as sample size grows, the sampling distribution of the mean approaches a normal shape, whatever shape the underlying population itself has. It does not require the original population to already be normal, it works for almost any population shape, provided the sample size is reasonably large, with around 30 commonly used as a rough rule of thumb, though a more strongly skewed population may need an even larger sample.

The sampling distribution of the mean is also typically far less spread out than the distribution of individual values, because averaging within a sample smooths out extremes: a single unusually high or low value barely shifts an average taken across many values.

Law of large numbers versus the Central Limit Theorem

These two results are easy to conflate but answer different questions.

Convergence versus shape
ResultWhat it is about
Law of large numbersConvergence: as a sample grows, its mean gets closer to the true population mean
Central Limit TheoremShape: as sample size grows, the sampling distribution of the mean becomes normal

Using the normal distribution on sample means

Because the Central Limit Theorem justifies treating the sampling distribution of the mean as approximately normal, the same Z-score and normal-curve methods used for individual values apply directly to sample means, just using the standard error in place of the population standard deviation.

Probabilities for the sample mean, not an individual value
ScenarioStandard errorZ-scoreResult
Mean 50, SD 20, n=25: P(sample mean above 54)20/5 = 4(54-50)/4 = 1.00.1587
Mean 100, SD 15, n=36: P(sample mean between 97 and 103)15/6 = 2.5-1.2 and 1.20.7699
Mean 40, SD 12, n=36: P(sample mean below 37)12/6 = 2(37-40)/2 = -1.50.0668

This is exactly why the Central Limit Theorem matters so much in practice: many common statistical tools assume normality, and the theorem is what makes applying them to sample means reasonable even when the underlying population looks nothing like a normal curve.

Common mistakes

  • Using the population standard deviation where the standard error is needed. A question about the sample mean’s spread calls for dividing by the square root of n first.
  • Assuming quadrupling the sample size quarters the standard error. It only halves it, since the formula depends on the square root of n.
  • Thinking the Central Limit Theorem requires a normal population to start with. It works for almost any population shape, that is the actual point of the theorem.
  • Confusing the law of large numbers with the Central Limit Theorem. One is about the mean converging to the truth; the other is about the sampling distribution’s shape.
  • Treating cluster sampling and stratified sampling as the same thing. Stratified sampling draws individuals from every subgroup; cluster sampling studies everyone inside only a few whole, randomly chosen groups.

Cheat sheet

Quick answers
QuestionAnswer
What gives every individual an equal chance of selection?Simple random sampling
What samples whole natural groups rather than individuals?Cluster sampling
What divides the population into subgroups before sampling each?Stratified sampling
What is the standard error of the mean formula?Population SD / sqrt(sample size)
Does quadrupling n quarter the standard error?No, it only halves it
Does the Central Limit Theorem require a normal population?No, it works for almost any shape, given a large enough sample
What sample size is a common rule-of-thumb minimum for the CLT?Around 30
What does the law of large numbers describe?Convergence of the sample mean to the true mean
What does the Central Limit Theorem describe?The normal shape of the sampling distribution of the mean
Does true random sampling eliminate sampling error?No, only selection bias; chance variation remains

Try it yourself

Quick check

A population has a mean of 200 and a standard deviation of 40. For random samples of size 64, what is the probability that the sample mean is above 210 (4 dp)?

Show answer

0.0228. The standard error is 40 divided by the square root of 64, which is 5.0. A sample mean of 210 is then 2.0 standard errors above the population mean ((210 minus 200) divided by 5), and the area of the standard normal curve beyond a Z-score of 2.0 is 0.0228.

Frequently asked questions

What is the standard error of the mean?

The standard error of the mean is the standard deviation of the sampling distribution of the sample mean, calculated as the population standard deviation divided by the square root of the sample size.

Does the Central Limit Theorem require the original population to be normally distributed?

No. It works for almost any population shape, as long as the sample size is reasonably large, around 30 is a common rough rule of thumb. This is the theorem’s main practical value.

What is the difference between the law of large numbers and the Central Limit Theorem?

The law of large numbers is about convergence: a sample mean gets closer to the true population mean as the sample grows. The Central Limit Theorem is about shape: the sampling distribution of the mean becomes approximately normal as the sample grows.

What is the difference between stratified sampling and cluster sampling?

Stratified sampling divides the population into subgroups and draws individuals from every subgroup. Cluster sampling randomly selects a few whole groups and measures everyone inside only those groups.

If I multiply my sample size by 4, does the standard error drop by the same factor?

No. Standard error depends on the square root of the sample size, so multiplying the sample size by 4 only cuts the standard error in half, not to a quarter.

Does truly random sampling eliminate all sampling error?

No. Random sampling removes systematic selection bias, but any sample smaller than the full population will still differ from it somewhat just by chance.

Test yourself

Timed questions on Sampling and the Central Limit Theorem, with an explanation for every answer.

Take the Quiz
Upskly AI Team
Learning made simple
Scroll to Top