Probability vs sampling distribution

Q: Does the sampling distribution have the same center as the population?

For the sample mean, yes: μ_x-bar = μ, so the sample mean neither systematically overshoots nor undershoots. For the sample proportion, the sampling distribution of p-hat is centered at the population proportion p. The center is inherited; the spread is not.

Q: Why does the spread shrink by the square root of n rather than by n?

Averaging n independent values lets high and low draws offset each other, and the arithmetic of that cancellation gives σ/sqrt(n). Quadruple the sample size and the spread halves. To cut it to a third you need 9 times as many observations, so each extra observation buys less than the one before it.

By Jude Wallis · Published August 11, 2026

A probability distribution gives the possible values of a random variable and their probabilities, usually for one observation. A sampling distribution is the probability distribution of a statistic from a whole sample of size n. For a mean or proportion: same center, spread smaller by root n.

AP Statistics: Unit 2 (topics 2.8 Introduction to Random Variables and Probability Distributions, 2.12 Sampling Distributions and the Central Limit Theorem, 3.2 Sampling Distributions for Sample Proportions, 4.1 Sampling Distributions for Sample Means). In the Fall 2026 AP Statistics course, probability distributions arrive in Unit 2 topic 2.8, Introduction to Random Variables and Probability Distributions. Sampling distributions come four topics later in 2.12, Sampling Distributions and the Central Limit Theorem, then return for proportions in Unit 3 topic 3.2 and for means in Unit 4 topic 4.1. The ordering reflects the relationship: the sampling distribution is the probability distribution idea applied to a statistic.

Probability vs sampling distribution: the short answer

Start with the fact that clears up most of the confusion: a sampling distribution is a probability distribution. It is not a rival idea or a separate species. The only thing that changes is what the random variable is.

A probability distribution lists the possible values of a random variable and the probability of each. The AP course introduces it for a single random variable, usually one observation from a chance process: one die roll, one customer, one measured height.

A sampling distribution is the probability distribution of a statistic. Its random variable is not one observation, it is a summary number such as $\bar{x}$ ("x-bar") or $\hat{p}$ ("p-hat") computed from an entire sample of size $n$ . One value of a sampling distribution stands for one whole sample, not one outcome.

So one population can generate both at once. The distribution of a single draw and the distribution of the mean of $n$ draws sit on the same center and differ in spread by a factor of $\sqrt{n}$ .

What a probability distribution is

A probability distribution assigns probability to the values of a random variable. For a discrete random variable it is a table or formula giving $P(X = x)$ for every possible $x$ , and two conditions define it: every probability sits between 0 and 1, and the probabilities add to exactly 1. For a continuous random variable there is no table. Probability is area under a density curve, and the area under the whole curve is 1.

From the distribution you can read the mean $\mu$ ("mu"), the standard deviation $\sigma$ ("sigma"), and the probability of any range of values. Nothing about it depends on a sample size, because there is no sample: the distribution describes the chance process itself, and it is fixed before any data exist.

Read the definition closely and one word is missing from it: nowhere does it say the random variable has to be a single observation. That omission is the door a sampling distribution walks through.

What a sampling distribution is

A sampling distribution is the distribution of a statistic across all possible samples of a fixed size $n$ from a population. Fix $n$ , consider every sample you could draw, compute the statistic for each, and record how likely each value is. The result satisfies the same rules as any probability distribution, because it is one.

For the sample mean $\bar{x}$ taken from a population with mean $\mu$ and standard deviation $\sigma$ , when the sampled values are independent:

Center: $\mu_{\bar{x}} = \mu$ , the same center as the population.
Spread: $\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}}$ , smaller than $\sigma$ for every $n$ greater than 1. That formula needs independent draws, which for sampling without replacement means the sample is at most 10% of the population.
Shape: normal when the population is normal, and approximately normal for a large $n$ whatever the population shape, which is the central limit theorem.

Unlike a probability distribution for one observation, a sampling distribution cannot be stated without naming $n$ . Ask for "the sampling distribution of the sample mean" and the honest reply is: of what sample size? Full detail is in sampling distributions explained, and the shape result in the central limit theorem guide.

One population, two distributions

Build both from the same process. At a coffee stand, let $X$ be the number of drinks in a single order, with $P(1) = 0.5$ , $P(2) = 0.3$ , and $P(3) = 0.2$ . Those add to 1, so this is a probability distribution. Working from it, $\mu = 1.7$ drinks and $\sigma \approx 0.7810$ drinks.

Now change the random variable rather than the population. Take $n = 2$ independent orders and record their mean $\bar{x}$ . The possible values are 1, 1.5, 2, 2.5, and 3, and their probabilities come from the same three numbers:

Value of $\bar{x}$	1	1.5	2	2.5	3
Probability	0.25	0.30	0.29	0.12	0.04

Those probabilities add to 1 as well, which is the whole point: this table is a probability distribution, and it is also the sampling distribution of $\bar{x}$ for $n = 2$ . Its mean is 1.7, unchanged. Its standard deviation is about 0.5523, which is $0.7810 / \sqrt{2}$ . The center held and the spread shrank. Every digit is worked out below.

The differences side by side

Feature	Probability distribution of one observation	Sampling distribution
Random variable	One observation, $X$	A statistic, $\bar{x}$ or $\hat{p}$
One value represents	One outcome	One entire sample of size $n$
Needs a sample size	No	Yes, always
Center, for a mean	$\mu$	$\mu$ , the same
Standard deviation, for a mean	$\sigma$	$\frac{\sigma}{\sqrt{n}}$
Effect of a larger $n$	None, $n$ does not appear	Narrower, and closer to normal unless it already is normal
Total probability is 1	Yes	Yes
AP topics	2.8	2.12, then 3.2 and 4.1

The last two rows are the ones students skip. A sampling distribution obeys every rule a probability distribution obeys, because the second column is the first column applied to a summary of $n$ observations instead of to one.

When to use which

Use the probability distribution when the question is about one outcome: the chance a single order is 3 drinks, a single battery lasts past 40 hours, a single student scores above 90. Compute directly from the table or the density curve, with no $n$ anywhere in the work.

Use the sampling distribution when the question is about a statistic from a sample: the chance the mean of 25 batteries exceeds 40 hours, or the chance a poll of 500 people gives $\hat{p}$ above 0.55. Every confidence interval and every p-value in this course is read off a sampling distribution, which is why this one idea carries the second half of the course.

The tell is in the wording. "The probability that one adult is taller than 71 inches" uses $\sigma$ . "The probability that the mean of 25 adults is taller than 71 inches" uses $\frac{\sigma}{\sqrt{25}} = \frac{\sigma}{5}$ . The sentences look almost identical, and unless the threshold sits right on $\mu$ the two answers are far apart: at $\mu = 70$ and $\sigma = 3$ inches they are 0.369 and 0.048. You can watch the second distribution assemble itself draw by draw in the sampling distribution and CLT interactive.

The classic mix-up and how to avoid it

The expensive error is standardizing a sample mean with $\sigma$ instead of $\frac{\sigma}{\sqrt{n}}$ . It looks like a small slip and it is not: it treats a question about a statistic as a question about one observation, and the z-score comes out too small by a factor of $\sqrt{n}$ , so a genuinely unusual result reads as ordinary. Before dividing, ask whether the number in the question is one value or an average.

A second mix-up confuses the sampling distribution with the distribution of the one sample you collected. A histogram of your 40 data points is a picture of that sample. As $n$ grows it comes to resemble the population, skew and all, and it does not narrow. The sampling distribution is the thing that narrows.

The third is treating the two ideas as opposites, which usually shows up as a student refusing to apply ordinary probability rules to $\bar{x}$ . They apply. Once you know that $\bar{x}$ has a distribution with a center and a standard deviation, it behaves like any other random variable, which is exactly the move made in expected value vs sample mean.

One population, both distributions written out in full

At a coffee stand a single order is 1 drink with probability 0.5, 2 drinks with probability 0.3, and 3 drinks with probability 0.2. Find the mean and standard deviation of this probability distribution. Then build the sampling distribution of the sample mean for n = 2 independent orders, and find its mean and standard deviation.

Confirm it is a probability distribution: $0.5 + 0.3 + 0.2 = 1$ , and every probability is between 0 and 1.
Mean of one order: $\mu = 1(0.5) + 2(0.3) + 3(0.2) = 0.5 + 0.6 + 0.6 = 1.7$ drinks.
For the standard deviation, first $E(X^2) = 1(0.5) + 4(0.3) + 9(0.2) = 0.5 + 1.2 + 1.8 = 3.5$ . Then the variance is $3.5 - 1.7^2 = 3.5 - 2.89 = 0.61$ , so $\sigma = \sqrt{0.61} \approx 0.7810$ drinks.
Now switch the random variable to $\bar{x}$ for $n = 2$ . There are $3 \times 3 = 9$ ordered pairs of independent orders, with means 1, 1.5, 2, 1.5, 2, 2.5, 2, 2.5, and 3.
The pairs are not equally likely, so add probabilities rather than counting pairs: $P(\bar{x} = 1) = 0.5^2 = 0.25$ ; $P(\bar{x} = 1.5) = 2(0.5)(0.3) = 0.30$ ; $P(\bar{x} = 2) = 2(0.5)(0.2) + 0.3^2 = 0.20 + 0.09 = 0.29$ ; $P(\bar{x} = 2.5) = 2(0.3)(0.2) = 0.12$ ; $P(\bar{x} = 3) = 0.2^2 = 0.04$ .
Check the total: $0.25 + 0.30 + 0.29 + 0.12 + 0.04 = 1$ . This sampling distribution is a probability distribution.
Its mean: $1(0.25) + 1.5(0.30) + 2(0.29) + 2.5(0.12) + 3(0.04) = 0.25 + 0.45 + 0.58 + 0.30 + 0.12 = 1.70$ , matching $\mu = 1.7$ .
Its variance: $E(\bar{x}^2) = 1(0.25) + 2.25(0.30) + 4(0.29) + 6.25(0.12) + 9(0.04) = 0.25 + 0.675 + 1.16 + 0.75 + 0.36 = 3.195$ , so the variance is $3.195 - 2.89 = 0.305$ and the standard deviation is $\sqrt{0.305} \approx 0.5523$ .
Check against the formula: $\frac{\sigma}{\sqrt{n}} = \frac{0.7810}{\sqrt{2}} = \frac{0.7810}{1.4142} \approx 0.5523$ . The direct computation and the formula agree.

One order has mean 1.7 drinks and standard deviation about 0.7810 drinks. The mean of two orders has the same center, 1.7 drinks, and standard deviation about 0.5523 drinks, which is $0.7810 / \sqrt{2}$ . Both tables are probability distributions and both come from the same population. Changing the random variable from one observation to the mean of two left the center alone and shrank the spread by a factor of $\sqrt{2}$ .

Same threshold, two very different probabilities

Using the same coffee stand, where an order is 1 drink with probability 0.5, 2 with probability 0.3, and 3 with probability 0.2, find the probability that a single order is 3 drinks or more. Then find the probability that the mean of 4 independent orders is 3 drinks or more, and give the standard deviation of that sample mean.

One observation first. Since 3 is the largest value $X$ can take, $P(X \geq 3) = P(X = 3) = 0.20$ , read straight off the probability distribution.
Now the statistic. For $\bar{x} \geq 3$ with $n = 4$ , the mean of four values none of which exceeds 3 must reach 3, which happens only when all four orders are 3.
So $P(\bar{x} \geq 3) = 0.2^4 = 0.2 \times 0.2 \times 0.2 \times 0.2 = 0.0016$ . This is exact; no normal approximation is used, and at $n = 4$ none would be appropriate.
Compare the two: $\frac{0.20}{0.0016} = 125$ , so a single order is 125 times as likely to reach 3 drinks as the mean of four orders is.
Standard deviation of the sample mean: $\frac{\sigma}{\sqrt{n}} = \frac{0.7810}{\sqrt{4}} = \frac{0.7810}{2} \approx 0.3905$ , exactly half the population standard deviation, because $\sqrt{4} = 2$ .

$P(X \geq 3) = 0.20$ for one order and $P(\bar{x} \geq 3) = 0.0016$ for the mean of four, a factor of 125. The sample mean has standard deviation about 0.3905 drinks, half of $\sigma \approx 0.7810$ . Same population and same threshold: which distribution the question is about decides the answer.

Frequently asked questions

Is a sampling distribution a probability distribution?

Yes. A sampling distribution obeys the same rules as any probability distribution: if it is discrete, probabilities between 0 and 1 that add to 1; if it is continuous, total area 1 under a density curve. The only difference is its random variable, which is a statistic computed from a sample of size n rather than a single observation.

What is the difference between a sampling distribution and the distribution of my sample?

The distribution of your sample is the histogram of the values you actually collected. As the sample grows it comes to look more like the population, skew included, and it does not get narrower. The sampling distribution describes how a statistic such as $\bar{x}$ varies across all possible samples of that size, and it is the one that narrows as $\frac{\sigma}{\sqrt{n}}$ .

Does the sampling distribution have the same center as the population?

For the sample mean, yes: $\mu_{\bar{x}} = \mu$ , so the sample mean neither systematically overshoots nor undershoots. For the sample proportion, the sampling distribution of $\hat{p}$ is centered at the population proportion $p$ . The center is inherited; the spread is not.

Why does the spread shrink by the square root of n rather than by n?

Averaging $n$ independent values lets high and low draws offset each other, and the arithmetic of that cancellation gives $\frac{\sigma}{\sqrt{n}}$ . Quadruple the sample size and the spread halves. To cut it to a third you need 9 times as many observations, so each extra observation buys less than the one before it.

Can a probability distribution describe something other than one observation?

Yes, and that is exactly why the sampling distribution fits inside the definition. A random variable is any numeric result of a chance process, and a statistic computed from a random sample qualifies. Once $\bar{x}$ is recognized as a random variable, its distribution is just a probability distribution with a familiar center and a smaller spread.