T-test

Also known as: Student's t-test, t statistic, t-statistic

A $t$-test compares an estimate to a null value using a $t$ reference distribution. For simple OLS with Normal iid errors, under $H_0:\beta=c$,

\[t=\frac{\hat\beta-c}{\mathrm{s.e.}(\hat\beta)}\sim t_{n-2}.\]
Reject two-sided $H_0$ at level $\alpha$ if $ t >t_{n-2,1-\alpha/2}$. Equivalent to checking whether $c$ lies outside the confidence interval. Large-sample $z$ tests replace $t$ when the CLT applies and df is large.

← Back to CS Glossary