Sampling Distributions

If you take two different random samples from the same population, you will get different sample means. Take a third sample, and you will get yet another mean. This natural variability is not a flaw in the sampling process — it is an unavoidable feature of working with samples instead of entire populations. Sampling distributions describe exactly how much this variability is, what shape it takes, and how it shrinks as your sample size grows. Understanding sampling distributions is the key to understanding confidence intervals, hypothesis tests, and virtually all of statistical inference.

What Is a Sampling Distribution?

A sampling distribution is not the distribution of data in a single sample. That is a common and important misconception to clear up right away. A sampling distribution is the distribution of a statistic — such as the sample mean $\bar{x}$ or the sample proportion $\hat{p}$ — computed across many possible samples of the same size drawn from the same population.

Here is the thought experiment that makes this concrete:

1. You have a population with some mean $\mu$ and standard deviation $\sigma$.
2. You draw a random sample of size $n$ and compute the sample mean $\bar{x}$.
3. You put that sample back (or draw from a very large population) and draw another sample of size $n$. You compute $\bar{x}$ again.
4. You repeat this process 1,000 times — or 10,000 times, or a million times.
5. Now you have 1,000 (or 10,000, or a million) values of $\bar{x}$. Plot them in a histogram.

That histogram is the sampling distribution of $\bar{x}$. It shows you all the possible values the sample mean could take and how likely each value is. In practice, you never actually draw thousands of samples — but the sampling distribution tells you what would happen if you did, and that knowledge is the foundation of statistical inference.

Sampling Distribution of the Sample Mean

The sampling distribution of $\bar{x}$ has three key properties:

Center: The mean of the sampling distribution equals the population mean:

$\mu_{\bar{x}} = \mu$

This says that the sample mean is an unbiased estimator of the population mean. If you averaged all possible sample means, you would get exactly $\mu$. Individual samples may overshoot or undershoot, but on average they hit the target.

Spread: The standard deviation of the sampling distribution is:

$\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}}$

This quantity is called the standard error of the mean. Notice the critical role of $n$: as the sample size increases, the standard error decreases. Larger samples produce more consistent (less variable) estimates.

Shape: When $n$ is large enough, the sampling distribution of $\bar{x}$ is approximately normal — regardless of the shape of the population. This remarkable fact is the Central Limit Theorem, covered on the next page. If the population itself is normal, then the sampling distribution of $\bar{x}$ is exactly normal for any sample size.

Example 1: Adult Heights

Suppose adult heights in a population have $\mu = 170$ cm and $\sigma = 8$ cm. You plan to take random samples of $n = 64$ adults and compute $\bar{x}$ for each sample.

The sampling distribution of $\bar{x}$ has:

• Mean: $\mu_{\bar{x}} = 170$ cm
• Standard error: $\sigma_{\bar{x}} = \frac{8}{\sqrt{64}} = \frac{8}{8} = 1$ cm

Individual heights vary with a standard deviation of 8 cm. But sample means (with $n = 64$) vary with a standard deviation of only 1 cm. The averaging process dramatically reduces variability — your sample mean will almost certainly be within a few centimeters of the true population mean.

The diagram above illustrates the key idea: while individual heights (dashed blue curve) spread widely around 170 cm, sample means with $n = 64$ (solid green curve) cluster tightly around 170 cm. The sampling distribution is far narrower than the population distribution.

Standard Error

The standard error (SE) is the standard deviation of a sampling distribution. It measures how much a statistic typically varies from sample to sample.

Standard error of the mean:

$SE = \frac{\sigma}{\sqrt{n}}$

When the population standard deviation $\sigma$ is unknown (which is typical in practice), we estimate it from the sample:

$SE \approx \frac{s}{\sqrt{n}}$

where $s$ is the sample standard deviation.

The key insight is the $\sqrt{n}$ in the denominator: to cut the standard error in half, you must quadruple the sample size. This is the law of diminishing returns for sampling — initial increases in sample size produce large gains in precision, but each further increase yields less additional benefit.

Example 2: Effect of Sample Size

A population has $\sigma = 20$. How does the standard error change as $n$ increases?

Going from $n = 25$ to $n = 100$ (a fourfold increase) cuts the SE from 4.0 to 2.0 — halving it. Going from $n = 100$ to $n = 400$ (another fourfold increase) cuts the SE from 2.0 to 1.0 — halving it again. Each time you want to halve the SE, you need four times as many observations. This relationship governs how researchers choose sample sizes for studies.

Sampling Distribution of the Sample Proportion

When working with categorical data (yes/no, pass/fail, support/oppose), the relevant statistic is the sample proportion $\hat{p}$. Its sampling distribution has analogous properties:

Center: The mean of $\hat{p}$ equals the population proportion:

$\mu_{\hat{p}} = p$

The sample proportion is an unbiased estimator of $p$.

Spread: The standard error of $\hat{p}$ is:

$\sigma_{\hat{p}} = \sqrt{\frac{p(1-p)}{n}}$

Shape: The sampling distribution of $\hat{p}$ is approximately normal when both of these conditions are met:

$np \geq 10 \quad \text{and} \quad n(1-p) \geq 10$

These conditions ensure that there are enough “successes” and “failures” in the expected sample for the normal approximation to work well.

Example 3: Voter Support

A candidate has true support of $p = 0.60$ in a population. A pollster takes a random sample of $n = 200$ voters.

Standard error:

$\sigma_{\hat{p}} = \sqrt{\frac{0.60 \times 0.40}{200}} = \sqrt{\frac{0.24}{200}} = \sqrt{0.0012} \approx 0.0346$

Normality check:

• $np = 200 \times 0.60 = 120 \geq 10$ --- condition met
• $n(1-p) = 200 \times 0.40 = 80 \geq 10$ --- condition met

So $\hat{p}$ is approximately normal with mean $0.60$ and standard error $0.0346$.

Interpretation: In repeated polls of 200 voters, the sample proportion would typically be within about $2 \times 0.0346 = 0.0692$ (about 7 percentage points) of the true proportion. This is why polls report a “margin of error” — it comes directly from the standard error of the sampling distribution.

Why Sampling Distributions Matter

Sampling distributions are not just an abstract concept — they are the engine that powers statistical inference. Here is why:

• Confidence intervals use the standard error to build a range of plausible values around a sample statistic. A 95% confidence interval for a mean is approximately $\bar{x} \pm 2 \times SE$, which comes directly from the sampling distribution.
• Hypothesis tests ask: “If the true parameter were some specific value, how likely is the sample result we observed?” The sampling distribution provides the answer by telling us what values are typical and what values are surprising.
• Margin of error in polls and surveys is a direct application of the standard error of $\hat{p}$.

Without understanding sampling distributions, confidence intervals and p-values are just mysterious formulas. With this understanding, they become logical consequences of how statistics vary from sample to sample.

Real-World Application: Nursing — Precision of Blood Pressure Estimates

A hospital clinic wants to estimate the average systolic blood pressure of its patient population. From prior studies, $\sigma = 18$ mmHg is a reasonable estimate of the population standard deviation.

How precise is the estimate for different sample sizes?

With only 9 patients, the estimate could easily be off by 12 mmHg — that is clinically meaningless because it spans the difference between normal and hypertensive. With 100 patients, the margin shrinks to 3.6 mmHg, which is precise enough for most clinical decisions. With 225 patients, the margin is 2.4 mmHg — very precise, but the improvement from 100 to 225 patients is modest compared to the improvement from 9 to 36.

This illustrates the practical tradeoff that nurses, physicians, and public health researchers face: more measurements improve precision, but with diminishing returns. The sampling distribution quantifies exactly how much precision you gain for each additional measurement.

Practice Problems

Test your understanding with these problems. Click to reveal each answer.

Key Takeaways

• A sampling distribution is the distribution of a statistic (like $\bar{x}$ or $\hat{p}$) across all possible samples of a given size — it describes how the statistic varies from sample to sample
• The sampling distribution of $\bar{x}$ is centered at $\mu$ (unbiased) with standard error $\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}}$
• The sampling distribution of $\hat{p}$ is centered at $p$ (unbiased) with standard error $\sigma_{\hat{p}} = \sqrt{\frac{p(1-p)}{n}}$
• Standard error measures the typical distance between a sample statistic and the population parameter — it is the standard deviation of the sampling distribution
• Larger samples produce smaller standard errors — but you must quadruple $n$ to halve the SE
• The normal approximation for $\hat{p}$ requires $np \geq 10$ and $n(1-p) \geq 10$
• Sampling distributions are the foundation of confidence intervals and hypothesis tests — they tell us what to expect when sampling from a population

Return to Statistics for more topics in this section.