When we have $n$ observations, the Student's t-distribution has $k=(n-1)$ degrees of freedom. The [[Random Variable]] follow the form:
$
t_{n-1} \sim \frac{Z}{\sqrt {(V/k)}}, \quad \text{where}
\begin{cases}
Z \sim \mathcal N(0,1)\\[2pt]
V \sim \chi_k^2 \\[2pt]
Z \perp V
\end{cases}
$
**Assumptions:**
- $Z$: A standard [[Gaussian Distribution|Gaussian]] r.v. $Z \sim \mathcal N(0,1)$ that represents the deviation of a test statistic (e.g. sample mean difference).
- $V$: A [[Chi-Square Distribution|Chi-Square]] r.v. $V \sim \chi_k^2$ with $k$ degrees of freedom. It represents the number of independent pieces used to estimate variability.
- $Z \perp V$: According to [[T-Test#Cochran’s Theorem|Cochran’s Theorem]] the sample variance $S_n^2$ is independent of the sample mean $\bar X_n$, denoted as $S_n^2 \perp \bar X_n$.
![[student-t-distribution.png|center|400]]
When $k \to \infty$ we see that $V \over k$converges to $\mathbb E[Z_1^2]$. For a standard Gaussian, this is equal to its variance $(=1)$. Thus we can concluded that we large $k$, the more will the t-distribution look like a $\mathcal N(0,1)$.
$
\frac{V}{k} = \frac{Z_1^2+\dots+Z_k^2}{k} \xrightarrow[n\to \infty]{\mathbf P}\mathbb E[Z_1^2]=1
$