Normal vs t distribution: the difference

By Jude Wallis · Published

The standard normal and t distributions are both symmetric, bell-shaped, and centered at 0. A t distribution has heavier tails, set by its degrees of freedom, and it approaches the standard normal as the degrees of freedom grow. Use t for inference about means when ss replaces σ\sigma.

AP Statistics: Unit 4 (topics 2.11 The Normal Distribution, 4.2 Constructing a Confidence Interval for a Population Mean or Population Mean Difference). In the Fall 2026 AP Statistics course, the normal distribution is Unit 2 topic 2.11, and t distributions are introduced in Unit 4 topic 4.2, Constructing a Confidence Interval for a Population Mean or Population Mean Difference. Unit 4, Inference for Quantitative Data: Means, is 10% to 20% of the multiple-choice section, and its confidence intervals and tests for means are t procedures.

Normal vs t distribution: the short answer

The fair comparison is between a tt distribution and the standard normal distribution, the normal curve with mean 0 and standard deviation 1, because both describe standardized statistics.

They share a shape: symmetric, single-peaked, bell-shaped, centered at 0. They differ in the tails. A tt distribution puts more probability far from 0 than the standard normal does, so values beyond 2 or 3 are more common under tt.

How much heavier the tails are depends on one number, the degrees of freedom (df). With few degrees of freedom the difference is large. As the degrees of freedom increase, the tt curve moves toward the standard normal curve, and for large df the two are nearly the same.

What the normal distribution is

A normal distribution is a continuous, symmetric, bell-shaped curve identified by two parameters: its mean μ\mu ("mu") and its standard deviation σ\sigma ("sigma"). The standard normal distribution is the member with μ=0\mu = 0 and σ=1\sigma = 1, and its values are called z-scores.

Normal distributions do two jobs in the course. They model data, such as heights or measurement errors. They also model statistics: when certain conditions hold, a standardized sample proportion follows the standard normal distribution closely, which is why inference for proportions uses zz.

The standard normal has fixed critical values. The middle 95% lies between −1.960-1.960 and 1.9601.960, whatever the sample size. Areas come from the z-table or from technology.

What the t distribution is

The tt distributions, also called Student's tt distributions, are a family of curves, one for each number of degrees of freedom. Each is symmetric, bell-shaped, and centered at 0, with a lower peak and heavier tails than the standard normal.

The family exists because of one substitution. To standardize a sample mean you would divide by σn\frac{\sigma}{\sqrt{n}}, but the population standard deviation σ\sigma is almost never known. Replacing it with the sample standard deviation ss gives the statistic

t=xˉ−μs/nt = \frac{\bar{x} - \mu}{s / \sqrt{n}}

Because ss varies from sample to sample, this ratio varies more than a z-score does. When the population is normal, the statistic follows a tt distribution with df=n−1df = n - 1. Small samples estimate σ\sigma poorly, so they have the heaviest tails. Critical values come from the t-table. The meaning of df is covered in what are degrees of freedom.

The differences side by side

FeatureStandard normal (zz)tt distribution
ShapeSymmetric, bell-shapedSymmetric, bell-shaped
Center00
TailsLighterHeavier, more so for small df
ParametersNone once standardizedDegrees of freedom
Standard deviation1dfdf−2\sqrt{\frac{df}{df - 2}} for df>2df > 2, which is greater than 1
Area beyond 2 in the right tail0.02280.0510 at df=5df = 5, 0.0273 at df=30df = 30
95% critical value1.9602.776 at df=4df = 4, 2.262 at df=9df = 9, 2.045 at df=29df = 29
Used forInference about proportionsInference about means when ss estimates σ\sigma

The critical values show the convergence. At 95% confidence, t∗t^* falls from 2.776 at 4 degrees of freedom to 2.045 at 29 and 1.984 at 99, approaching the normal value of 1.960 from above. A t∗t^* is always larger than the z∗z^* for the same confidence level.

When to use which

Use the standard normal distribution for inference about proportions: one-proportion and two-proportion zz intervals and zz tests. Use it also for probability questions about a variable or a sample mean when the population standard deviation σ\sigma is given.

Use a tt distribution for inference about means when σ\sigma is unknown and you use ss in its place, which is nearly always the case with real data. A one-sample or paired procedure uses df=n−1df = n - 1. The choice between procedures is covered in t-test vs z-test and z-interval vs t-interval.

Sample size does not change the rule. With a large sample and ss, the correct distribution is still tt. The tt and zz values are simply close together by then: at 99 degrees of freedom the 95% critical value is 1.984 against 1.960.

The classic mix-up and how to avoid it

The main error is using z∗=1.960z^* = 1.960 for a confidence interval for a mean built from ss. With a small sample this makes the interval too narrow, so it captures the true mean less often than the stated 95%. With n=10n = 10 the correct multiplier is t∗=2.262t^* = 2.262, about 15% larger than 1.960.

The second error is the reverse: using tt for a proportion. Proportion procedures have no sample standard deviation ss to estimate separately, and they use zz.

The third error is believing a tt distribution turns into a normal distribution at n=30n = 30. The number 30 is a guideline for a different question, whether the sampling distribution of xˉ\bar{x} is close enough to normal for the procedure to be valid. The tt curve approaches the standard normal gradually and never switches over.

The fourth is confusing the two roles of a distribution. Describing the data as roughly normal is a statement about the population or sample. Using a tt distribution is a statement about the standardized statistic. A problem can involve both: normal data, and a tt statistic computed from it.

The same data with t and with z

A random sample of 10 batteries has a mean life of 52 hours and a sample standard deviation of 6 hours. Assume battery life is approximately normal. Build the 95% confidence interval for the mean life using the correct t critical value, then show what the interval would be if 1.960 were used by mistake.

  1. Identify the setting: a mean, with s=6s = 6 standing in for the unknown σ\sigma. The correct distribution is tt with df=n−1=10−1=9df = n - 1 = 10 - 1 = 9.

  2. Standard error: SE=sn=610≈1.8974SE = \frac{s}{\sqrt{n}} = \frac{6}{\sqrt{10}} \approx 1.8974 hours.

  3. Critical value from the t-table at 95% confidence and 9 degrees of freedom: t∗=2.262t^* = 2.262.

  4. Margin of error with tt: 2.262×1.8974≈4.2922.262 \times 1.8974 \approx 4.292 hours. Interval: 52−4.292=47.70852 - 4.292 = 47.708 to 52+4.292=56.29252 + 4.292 = 56.292 hours.

  5. Margin of error if z∗=1.960z^* = 1.960 is used by mistake: 1.960×1.8974≈3.7191.960 \times 1.8974 \approx 3.719 hours. Interval: 52−3.719=48.28152 - 3.719 = 48.281 to 52+3.719=55.71952 + 3.719 = 55.719 hours.

  6. Compare the widths. The tt interval is 2×4.292=8.5842 \times 4.292 = 8.584 hours wide and the zz interval is 2×3.719=7.4382 \times 3.719 = 7.438 hours wide. The tt interval is wider by the ratio 2.2621.960≈1.154\frac{2.262}{1.960} \approx 1.154.

The correct 95% interval is about 47.71 to 56.29 hours, using t∗=2.262t^* = 2.262 with 9 degrees of freedom. Using 1.960 would give about 48.28 to 55.72 hours, an interval that is too narrow for 95% confidence because it ignores the extra uncertainty from estimating σ\sigma with ss.

A p-value that changes with the distribution

A random sample of 9 values from an approximately normal population has mean 54 and sample standard deviation 6. Test the null hypothesis that the population mean is 50 against a two-sided alternative at the 0.05 significance level. Find the p-value from the t distribution, and compare it with the p-value the standard normal would give.

  1. Compute the test statistic: t=xˉ−μ0s/n=54−506/9=42=2t = \frac{\bar{x} - \mu_0}{s/\sqrt{n}} = \frac{54 - 50}{6/\sqrt{9}} = \frac{4}{2} = 2.

  2. Degrees of freedom: df=9−1=8df = 9 - 1 = 8.

  3. Two-sided p-value from the tt distribution with 8 degrees of freedom: 2×P(T>2)≈0.08052 \times P(T > 2) \approx 0.0805.

  4. Decision with tt: 0.0805>0.050.0805 > 0.05, so fail to reject the null hypothesis.

  5. For comparison, the standard normal gives 2×P(Z>2)=2×0.02275=0.04552 \times P(Z > 2) = 2 \times 0.02275 = 0.0455, which is below 0.05.

  6. Check with critical values: the 5% two-sided cutoff is t∗=2.306t^* = 2.306 at 8 degrees of freedom and z∗=1.960z^* = 1.960 for the standard normal. The statistic 2 is inside the first cutoff and outside the second.

With the correct tt distribution at 8 degrees of freedom, the p-value is about 0.0805, and the null hypothesis is not rejected at the 0.05 level. Treating the statistic as a z-score would give about 0.0455 and the opposite decision. The heavier tails of the tt distribution make a value of 2 less surprising.

Frequently asked questions

Why does the t distribution have heavier tails than the normal?

Because the tt statistic divides by a standard error built from ss, and ss varies from sample to sample. That extra source of variation makes large values of the statistic more likely than they are for a z-score, most of all in small samples.

Is the t distribution the same as the normal distribution for large samples?

Very close, but not identical. As the degrees of freedom increase, the tt curve approaches the standard normal. At 99 degrees of freedom the 95% critical value is 1.984 against 1.960. You still use tt for a mean whenever ss replaces σ\sigma.

Do I use z or t for a proportion?

Use zz. Inference about proportions in this course is based on the standard normal distribution. The tt distributions are for means, where the sample standard deviation ss is used in place of the unknown population standard deviation.