z-interval vs t-interval: when to use each

Q: When would you ever use a z-interval for a mean?

Only if the population standard deviation σ is known, which almost never happens with real data. Because σ is nearly always unknown and estimated by s, AP Statistics builds intervals for a mean with the t distribution.

By Jude Wallis · Published July 28, 2026

Use a z-interval for a proportion, where the standard error comes from p-hat alone; use a t-interval for a mean, where the population standard deviation is unknown and estimated by s, so the t distribution absorbs the extra uncertainty.

In the Fall 2026 AP Statistics course, the z-interval for a proportion is built in Unit 3 topic 3.3, and the t-interval for a mean is built in Unit 4 topic 4.2 because the population standard deviation is unknown; this comparison spans both units.

What a z-interval is

A z-interval is a confidence interval built with critical values from the standard normal distribution. In AP Statistics it is the interval for a population proportion $p$ , using $\hat{p} \pm z^* \sqrt{\frac{\hat{p}(1 - \hat{p})}{n}}$ , where $\hat{p}$ (p-hat) is the sample proportion and $z^*$ is the normal critical value. There is no separate spread parameter to estimate, because the standard error of a proportion depends only on $\hat{p}$ and the sample size $n$ . That is why the normal distribution applies directly.

What a t-interval is

A t-interval is a confidence interval built with critical values from a t distribution. In AP Statistics it is the interval for a population mean $\mu$ , using $\bar{x} \pm t^* \frac{s}{\sqrt{n}}$ , where $\bar{x}$ (x-bar) is the sample mean, $s$ is the sample standard deviation, and $t^*$ has $n - 1$ degrees of freedom. Because the population standard deviation $\sigma$ is unknown, you estimate it with $s$ , and that added uncertainty makes the t distribution's tails heavier than the normal. Heavier tails give a larger critical value, so a t-interval is slightly wider than a z-interval at the same confidence level.

z-interval vs t-interval at a glance

The parameter you are estimating decides which interval you build.

Feature	z-interval	t-interval
Parameter	Population proportion $p$	Population mean $\mu$
Critical value	$z^*$ from the normal	$t^*$ from the t distribution
Standard error	$\sqrt{\frac{\hat{p}(1 - \hat{p})}{n}}$	$\frac{s}{\sqrt{n}}$
Degrees of freedom	None	$n - 1$
Why	SE depends only on $\hat{p}$	$\sigma$ unknown, estimated by $s$
AP unit	Unit 3	Unit 4

When to use which

Ask what kind of parameter you are estimating. If the data are categorical and summarized as a proportion, build a z-interval. If the data are quantitative and summarized as a mean, build a t-interval, because you almost never know the true $\sigma$ . The tie-breaker is the same one that governs z-tests and t-tests: proportions use z, and means use t; see t-test vs z-test.

The classic mix-up and how to avoid it

The usual mistake is using a normal critical value $z^*$ for a mean when $\sigma$ is unknown. That produces an interval that is too narrow, understating the uncertainty, especially at small sample sizes where $t^*$ is noticeably larger than $z^*$ . A second slip is forgetting the degrees of freedom and reading the wrong row of the t-table. Avoid both by naming the parameter first: a proportion means $z^*$ , and a mean means $t^*$ with $n - 1$ degrees of freedom.

A t-interval for a mean, and why z would mislead

A random sample of $n = 16$ has $\bar{x} = 50$ and $s = 8$ . Build a $95\%$ confidence interval for the population mean, then show what using a z critical value instead would do.

Standard error of the mean: $SE = \frac{s}{\sqrt{n}} = \frac{8}{\sqrt{16}} = \frac{8}{4} = 2$ .
Degrees of freedom: $n - 1 = 16 - 1 = 15$ . The t critical value for $95\%$ confidence with 15 degrees of freedom is $t^* = 2.131$ .
Margin of error: $t^* \times SE = 2.131 \times 2 = 4.262$ .
Confidence interval: $50 \pm 4.262$ , which is $(45.738, 54.262)$ .
Using $z^* = 1.96$ by mistake gives a margin of only $1.96 \times 2 = 3.92$ and the narrower interval $(46.08, 53.92)$ , which understates the true uncertainty from estimating $\sigma$ .

The 95% t-interval is $(45.738, 54.262)$ ; the incorrect z-interval $(46.08, 53.92)$ is too narrow because it ignores the extra uncertainty of estimating $\sigma$ with $s$ .

Frequently asked questions

When would you ever use a z-interval for a mean?

Only if the population standard deviation $\sigma$ is known, which almost never happens with real data. Because $\sigma$ is nearly always unknown and estimated by $s$ , AP Statistics builds intervals for a mean with the t distribution.

Why is a t-interval wider than a z-interval?

The t distribution has heavier tails than the normal, so its critical value $t^*$ is larger than the matching $z^*$ at the same confidence level. That larger multiplier makes the margin of error and the interval wider, which honestly reflects the uncertainty of estimating $\sigma$ .