T-statistics are more variable than z-scores because their formula contains an extra source of randomness. A z-score uses the true population standard deviation, which is a fixed number. A t-statistic replaces that fixed number with the sample standard deviation, which changes every time you draw a new sample. That additional moving part in the denominator makes the whole statistic wobble more from sample to sample, producing a distribution with more spread and heavier tails than the standard normal.
The Key Difference in the Formulas
Both statistics measure how far a sample mean falls from the population mean, scaled by some measure of spread. The z-score formula is z = (X̄ − μ) / (σ / √n), where σ is the known population standard deviation. The t-statistic formula is t = (X̄ − μ) / (s / √n), where s is the standard deviation calculated from your sample.
That single substitution, s for σ, is the entire reason t-statistics behave differently. The population standard deviation σ is a constant. It doesn’t change no matter how many samples you draw. The sample standard deviation s is itself a random variable. Sometimes your sample will underestimate the true spread, sometimes it will overestimate it. When s happens to be smaller than σ, it shrinks the denominator and inflates the t-value. When s is larger, it does the opposite. This means the t-statistic fluctuates for two reasons at once: the numerator (sample mean) varies AND the denominator (sample standard deviation) varies. The z-score only fluctuates because of the numerator.
Why This Creates Heavier Tails
Because the denominator of the t-statistic is itself random, extreme values become more likely. Imagine you draw a sample where the mean happens to land far from μ (a large numerator) and, by chance, the sample standard deviation comes out unusually small (a small denominator). That combination produces a very large t-value. With a z-score, the denominator is locked in place, so you can only get an extreme value if the numerator is extreme. With a t-statistic, the numerator and denominator can conspire together to push the value further into the tails.
This is why the t-distribution is shorter in the middle and fatter in the tails compared to the standard normal curve. The probability of landing far from zero is higher. As Penn State’s statistics reference puts it: since s is a random quantity varying with each sample, the variability in t is greater, resulting in a larger spread.
The Role of Degrees of Freedom
The t-distribution’s shape depends on degrees of freedom, which equal n − 1 for a one-sample t-test. With a small sample, your estimate of the standard deviation is unreliable. Five data points simply can’t pin down the population’s spread with much precision, so s bounces around a lot from sample to sample. That extra bouncing translates directly into thicker tails for the t-distribution.
As your sample size grows, s becomes a better and better estimate of σ. With 200 observations, the sample standard deviation barely changes from one sample to the next, so the denominator of the t-statistic is nearly constant. At that point, the t-statistic behaves almost identically to a z-score. Mathematically, as degrees of freedom approach infinity, the t-distribution converges to the standard normal distribution. The extra variability from estimating σ effectively disappears.
You can see this visually if you overlay t-distributions with different degrees of freedom on a standard normal curve. At 3 or 4 degrees of freedom, the t-distribution is noticeably flatter and wider. By 30 degrees of freedom, the two curves are hard to tell apart. By 100, they’re virtually identical.
The Formal Mathematical Reason
There’s a precise way to describe what’s happening under the hood. The t-statistic can be written as a standard normal random variable divided by the square root of a chi-squared random variable divided by its degrees of freedom. In notation: t = Z / √(V / (n−1)), where Z is standard normal and V follows a chi-squared distribution with n − 1 degrees of freedom.
If that denominator were always exactly equal to 1, you’d just have a z-score. But because V is random, the denominator fluctuates around 1, sometimes above and sometimes below. Dividing by a number less than 1 amplifies Z; dividing by a number greater than 1 shrinks it. The net effect is more spread in both directions. The variance of a t-distribution with ν degrees of freedom is ν / (ν − 2), which is always greater than 1 (the variance of the standard normal). With 5 degrees of freedom, the variance is 5/3, or about 1.67. With 30 degrees of freedom, it drops to 30/28, or about 1.07. It only equals 1 in the limit as degrees of freedom go to infinity.
Why This Matters in Practice
The extra variability of the t-statistic isn’t just a mathematical curiosity. It has real consequences for hypothesis testing. Because extreme t-values are more probable than extreme z-values, the critical values for a t-test are larger. For a two-tailed test at the 5% significance level, the z critical value is always 1.96. The t critical value at 5 degrees of freedom is 2.57, and at 10 degrees of freedom it’s 2.23. You need a bigger observed value to reject the null hypothesis.
This means t-tests are deliberately more conservative when your sample is small. They build in extra caution to account for the fact that you’re estimating the population’s spread from limited data. If you incorrectly used z critical values with a small sample and an estimated standard deviation, you’d reject the null hypothesis more often than your stated significance level, inflating your false positive rate. The wider t-distribution corrects for exactly this problem.
The practical takeaway: t-statistics are more variable because they carry the uncertainty of estimating two things (the mean and the spread) rather than just one. That extra uncertainty shows up as a wider distribution, and the t-test’s critical values are calibrated to respect it.