How to interpret a confidence interval for a mean

The one sentence to write

A confidence interval for a mean gets interpreted in a single sentence, and the sentence has a fixed shape:

We are C% confident that the true mean [quantity] for [population] is between [lower limit] and [upper limit] [units].

Three pieces have to be present. The confidence level C, a reference to the parameter written as the mean of a specific quantity for a specific population, and both endpoints carrying the units of the measurement. The Fall 2026 AP course states this directly in Unit 4 topic 4.3: an interpretation of a confidence interval for a population mean includes a reference to the parameter.

Filled in with the numbers from the worked example below, where 16 randomly chosen members of a climbing gym averaged 50 minutes on the wall:

We are 95% confident that the true mean session length for all members of this climbing gym is between 45.738 and 54.262 minutes.

The parameter you are estimating is $\mu$ (mu), the mean of the entire population. The number you actually measured is $\bar{x}$ (x-bar), the mean of your sample. Your interval is built from $\bar{x}$ and it is an estimate of $\mu$, so the sentence has to talk about $\mu$, not about $\bar{x}$.

What the confidence level describes

The 95% is a property of the procedure, not of the interval sitting in front of you. Imagine taking random samples of the same size from the same population over and over, building an interval each time by the same recipe. About 95% of those intervals would capture the population mean, and about 5% would miss it.

That long-run capture rate is the whole content of the phrase "95% confident". Your particular interval, 45.738 to 54.262 minutes, either contains $\mu$ or it does not. You cannot tell which, because $\mu$ is unknown, and that uncertainty is exactly what the confidence level prices in.

So the randomness lives in the sampling, not in $\mu$. Before you draw the sample, there is a 95% chance the interval you are about to build will cover $\mu$. After you draw it and compute two specific numbers, nothing is random anymore. The guide on what 95 percent confidence means works through the many-intervals picture in more detail.

Four wrong readings, corrected

Each of these sounds reasonable and each one is wrong. The interval below is the 95% interval (45.738, 54.262) minutes from a sample with $\bar{x} = 50$ minutes and $s = 8$ minutes.

"95% of the data fall between 45.738 and 54.262 minutes." No. The interval estimates the mean session length, not the spread of individual sessions. Individual sessions vary by roughly $s = 8$ minutes, so a rough picture of where most individuals sit would be $50 \pm 2(8)$, or about 34 to 66 minutes, much wider than the confidence interval. Confidence intervals get narrower as $n$ grows; the spread of the data does not.

"95% of sample means fall between 45.738 and 54.262 minutes." No. Sample means from repeated samples cluster around $\mu$, which is unknown. Your interval is centered at $\bar{x} = 50$, a value that itself came from one sample and is almost certainly not equal to $\mu$. There is no reason 95% of future sample means would land inside an interval centered on one particular sample's result.

**"There is a 95% probability that $\mu$ is between 45.738 and 54.262."** No. Once the interval is computed, both endpoints are fixed numbers and $\mu$ is a fixed number. The probability that a fixed number lies between two other fixed numbers is 0 or 1, not 0.95. Assign the 95% to the method, not to this interval.

"We are 95% confident the sample mean is between 45.738 and 54.262." No. You know the sample mean exactly: it is 50 minutes, and it sits at the dead center of the interval. There is nothing to be uncertain about. The unknown quantity is the population mean.

How the interval gets built

When the population standard deviation is unknown, which is nearly always, you use a one-sample t-interval for a population mean:

Here $s$ is the sample standard deviation, $n$ is the sample size, and $t^*$ (t-star) is the critical value that captures the central C% of a t-distribution with $n - 1$ degrees of freedom. The piece $s/\sqrt{n}$ is the standard error of $\bar{x}$, the typical distance between a sample mean and the population mean. The product $t^* \cdot s/\sqrt{n}$ is the margin of error.

Three conditions come first: the data must come from a random sample or randomized experiment, the population must be at least 10 times the sample size when sampling without replacement, and the sample data must either come from a population stated to be approximately normal, or have $n \ge 30$, or, for smaller $n$, show no strong skewness and no outliers.

Read $t^*$ off the t-table using $n - 1$ degrees of freedom, or let the confidence interval calculator do the arithmetic. If you are unsure whether the situation calls for $t$ or $z$, see z-interval vs t-interval, and for the difference between $s$ and $s/\sqrt{n}$, see standard error vs standard deviation.

Using the interval to judge a claimed value

A confidence interval is a list of plausible values for $\mu$, which makes it a tool for testing someone's claim. The rule is simple, and it works in both directions.

• If the claimed value is outside the interval, you have convincing evidence against the claim at the matching significance level, $\alpha = 1 - C$. The claim is not among the plausible values your data support.
• If the claimed value is inside the interval, you do not have convincing evidence against the claim. The claim stays plausible.

The gym's website says the average member session lasts 60 minutes. Since 60 is above the upper limit of 54.262, the interval gives convincing evidence that the true mean session length is not 60 minutes, and every plausible value in the interval is shorter than 60.

The second half of the rule is where answers go wrong. Suppose the claim had been 48 minutes. That value falls inside (45.738, 54.262), so the data do not contradict it.

That is not the same as proving the mean is 48. Every value in the interval is plausible, and 48 is only one of them, so the honest conclusion is that the data are consistent with a mean of 48 minutes.

Why a wider interval means less precision

The width of an interval is twice the margin of error, and it measures how precisely you have pinned down $\mu$. The 95% interval above runs from 45.738 to 54.262 minutes, a width of 8.524 minutes.

Raising the confidence level buys certainty with precision. At 99% confidence the critical value grows from $t^* = 2.131$ to $t^* = 2.947$, the margin of error grows from 4.262 to 5.894 minutes, and the interval stretches to (44.106, 55.894), a width of 11.788 minutes. You are more confident and less precise, and the two move in opposite directions every time.

Sample size pushes the other way. A larger $n$ shrinks the standard error $s/\sqrt{n}$, so the interval narrows. The AP course notes that the width of a t-interval is approximately proportional to $1/\sqrt{n}$, which means quadrupling the sample size roughly halves the width. Collecting more data is the only way to gain precision without giving up confidence.

A very wide interval is not a mistake. It is an honest report that a small or highly variable sample cannot say much, and treating a 40-minute-wide interval as a useful estimate is the real error.

Check your work and practice

Build a few intervals with the confidence interval calculator and watch how the endpoints move as you change $n$ and the confidence level. Then write the interpretation sentence from memory each time, because the sentence is the part that has to be exactly right.

For the mechanics of building intervals for both means and proportions, see how to calculate a confidence interval. When you are ready to be tested on it, work through the confidence interval for means practice set.