How Sample Size Changes Mean Variability

Why Sample Size Matters for Sample Means

A sample mean varies from sample to sample: two random samples from the same population will usually have different averages. The standard error measures the standard deviation of this sampling distribution, so it describes the typical amount of variation among sample means. A larger sample generally makes the mean more stable, but the size of that improvement follows a square-root rule rather than a direct one-for-one relationship.

This tutorial focuses on quantifying that relationship. The standard-error formula lets us compare sample sizes, describe the resulting change in variability, and work backward from a desired precision to a required sample size.

Definition: For independent observations from a population with mean \(\mu\) and finite standard deviation \(\sigma\), the standard error of the sample mean \(\bar{x}\) is the standard deviation of its sampling distribution. For a sample of size \(n\), \(SE_{\bar{x}}=\frac{\sigma}{\sqrt{n}}\). This formula assumes the observations are independent and have the same population mean and variance.

The value \(\sigma\) describes the spread of individual observations in the population. In contrast, \(SE_{\bar{x}}\) describes the spread of sample means across repeated samples of size \(n\). Their units are the same as the units of the measured variable. If the population standard deviation is not known, \(s/\sqrt{n}\) is commonly used as an estimate of the standard error; the formula involving \(\sigma\) gives the standard error itself under the stated model.

The Square-Root Relationship

Write \(SE_n\) for the standard error when the sample size is \(n\). If the population standard deviation stays the same, comparing two sample sizes gives

\frac{SE_{n_2}}{SE_{n_1}} = \frac{\sigma/\sqrt{n_2}}{\sigma/\sqrt{n_1}} = \sqrt{\frac{n_1}{n_2}}.

The population spread cancels from the ratio. The relative change in standard error therefore depends only on the relative change in sample size, provided the same population and sampling conditions apply.

Key result: Multiplying sample size by a factor \(k\) divides the standard error by \(\sqrt{k}\). For example, doubling \(n\) multiplies the standard error by \(1/\sqrt{2}\), while quadrupling \(n\) halves it. To reduce the standard error by a factor \(c\), where \(0<c<1\), multiply the sample size by \(1/c^2\).

This result follows directly from the formula. If \(n_2=kn_1\), then

SE_{n_2} = \frac{\sigma}{\sqrt{kn_1}} = \frac{1}{\sqrt{k}}\frac{\sigma}{\sqrt{n_1}} = \frac{SE_{n_1}}{\sqrt{k}}.

For a target standard error \(E\), the same formula can be rearranged to find the minimum sample size needed under the model:

\frac{\sigma}{\sqrt{n}}\le E \quad\Longleftrightarrow\quad n\ge \left(\frac{\sigma}{E}\right)^2.

Because a sample size must be a whole number, round the calculated requirement up if it is not already an integer. This calculation assumes that the value used for \(\sigma\) is appropriate for the population and that the independence conditions hold.

Comparing Standard Errors in Practice

Worked Example: Doubling the Sample Size

Suppose a hypothetical population of delivery times has a standard deviation of 12 minutes. Independent random samples are drawn from this population. Compare the standard errors of the sample mean for \(n=36\) and \(n=72\).

The formula applies because the example specifies independent observations from the same population, with population standard deviation \(\sigma=12\) minutes. For \(n=36\),

SE_{36} = \frac{\sigma}{\sqrt{n}} = \frac{12\text{ minutes}}{\sqrt{36}} = \frac{12}{6}\text{ minutes} = 2\text{ minutes}.

For \(n=72\),

SE_{72} = \frac{12\text{ minutes}}{\sqrt{72}} \approx 1.414\text{ minutes}.

Doubling the sample size changed the standard error by a factor of \(1/\sqrt{2}\approx0.7071\). Thus the means from repeated samples of 72 delivery times have about 70.7% of the standard error of means from samples of 36. The standard error falls by about 29.3%, not by 50%. This describes the sampling variability of the mean, not a guarantee about the difference between any particular two sample means.

Worked Example: How Much Sample Size to Halve the Standard Error?

A hypothetical population of measurement errors has a standard deviation of 10 units. A researcher uses independent observations and currently plans to take \(n=25\). How large a sample is needed to cut the standard error in half?

First calculate the current standard error:

SE_{25} = \frac{10\text{ units}}{\sqrt{25}} = 2\text{ units}.

To halve it, the target is \(1\) unit. Using the target-size relationship,

n \ge \left(\frac{\sigma}{E}\right)^2 = \left(\frac{10\text{ units}}{1\text{ unit}}\right)^2 = 100.

A sample size of 100 gives \(SE_{100}=10/\sqrt{100}=1\) unit, exactly half the original standard error. The required sample size is four times 25, because reducing standard error by a factor of \(1/2\) requires multiplying sample size by \(1/(1/2)^2=4\). The calculation assumes that the population standard deviation is 10 units and that the observations can be treated as independent.

Worked Example: Planning for a Maximum Standard Error

For a hypothetical population of package weights, suppose the population standard deviation is 18 grams. What sample size is needed to make the standard error no greater than 1.5 grams?

Assume the package weights in the sample are independent observations from the same population. Rearrange the standard-error formula:

\frac{18\text{ grams}}{\sqrt{n}} \le 1.5\text{ grams} \quad\Longleftrightarrow\quad n \ge \left(\frac{18}{1.5}\right)^2 = 12^2 = 144.

Thus, at least 144 independent observations are required under these assumptions. Checking the boundary value confirms the result:

SE_{144} = \frac{18\text{ grams}}{\sqrt{144}} = \frac{18}{12}\text{ grams} = 1.5\text{ grams}.

A smaller sample would have a standard error greater than 1.5 grams according to the formula. This is a precision calculation for the sampling distribution; it does not determine the sample mean that will be observed or guarantee that a particular sample mean is within 1.5 grams of \(\mu\).

What Gets Less Variable—and What Does Not

Increasing \(n\) reduces the standard deviation of the sampling distribution of \(\bar{x}\). It does not, by itself, change the spread of individual values in the population. For example, collecting more delivery times does not make delivery times themselves less variable. It makes the average of a larger sample less variable from one repeated sample to another.

The standard-error formula also does not say that every sample mean from a larger sample will be closer to \(\mu\) than every sample mean from a smaller sample. Individual random samples can still produce unusual means. The formula describes the typical spread over repeated samples, rather than the exact outcome or error of a specific sample.

The square-root relationship has a practical consequence for planning. A modest reduction in standard error can require a substantial increase in sample size: halving the standard error requires four times as many independent observations, and reducing it to one-third requires nine times as many. This helps make the trade-off between precision and data collection explicit.

The calculations rely on the independent-observation model stated above. If observations are dependent, or if a sampling design requires a different variance calculation, simply applying \(\sigma/\sqrt{n}\) may not give the correct standard error. Also, when \(\sigma\) is unknown, using \(s\) gives an estimated standard error rather than a known population-based value.

Common Mistakes

Assuming twice the sample size halves the standard error. The denominator is \(\sqrt{n}\), so doubling \(n\) divides the standard error by \(\sqrt{2}\), not by 2. A fourfold sample size is needed to halve it.
Confusing individual spread with mean variability. The population standard deviation \(\sigma\) describes individual observations; \(\sigma/\sqrt{n}\) describes the variability of sample means.
Interpreting a smaller standard error as a guarantee. A smaller standard error means less spread across repeated sample means. It does not ensure that each observed mean is close to the population mean.
Forgetting the assumptions or units. State that the formula is being used for independent observations from the same population, and retain the measurement units in the standard error.

Key takeaway: Under independent sampling with population standard deviation \(\sigma\), the standard error of a sample mean is \(\sigma/\sqrt{n}\). Multiplying the sample size by \(k\) multiplies the standard error by \(1/\sqrt{k}\); use this relationship to compare precision or plan a sample size.

Check Your Understanding

Use the standard-error formula and its assumptions to answer these questions.

A population has standard deviation 8 minutes. What is the standard error of the sample mean when \(n=16\), assuming independent observations?
If the sample size increases from 25 to 100 and the population spread is unchanged, by what factor does the standard error change?
Why does doubling a sample size not halve the standard error?
If \(\sigma=15\) units, what is the smallest sample size that gives a standard error no greater than 3 units under the independent-sampling model?
Explain why a smaller standard error does not mean that the spread of individual population values has decreased.