Let $X_{1},X_{2},....,X_{n}$ be independent $N(0,1)$ random variables and set $\bar{X}=\frac{1}{n}\sum_{i=1}^{n}X_{i}$ show that $\bigg(\bar{X},\frac{1}{n}\sum_{i=1}^{n}(X_{i}-\bar{X})^{2}\bigg)$ and $\bigg({X}/\sqrt{n},\frac{1}{n}\sum_{i=1}^{n-1}(X_{i})^{2}\bigg)$ have the same distribution. This shows that the sample mean and sample variance are independent in this model.

Thoughts I tried showing that they have the same characteristic function but couldn't get it to work that way. I think maybe I could do something involving the CLT but I'm not sure how.


Using the Gram-Schmidt process, construct an orthonormal basis of ${\mathbb R}^n$ containing the unit vector ${\bf a}:=\frac1{\sqrt n}(1,\ldots,1)^T$. Convert this basis into an orthogonal matrix $A$ with $\bf a$ as its last row. That is, we have found an $n\times n$ matrix $A$ with $AA^T=A^TA=I$ and $A_{ni}=\frac1{\sqrt n}$ for each $i$.

Pack the given variables $X_1,\ldots,X_n$ into a column vector $X$ and define a new vector of random variables $Z:=(Z_1,\ldots,Z_n)^T$ by $Z=AX$. We observe:

  1. The variables $Z_1,\ldots,Z_n$ are also independent standard normal variables, since they are a linear combination of the $X$'s, with mean vector zero and covariance matrix $$\operatorname{Cov}(Z)=E(ZZ^T)=E(AXX^TA^T)=AE(XX^T)A^T=AA^T=I.$$
  2. The final variable $Z_n$ equals $\sqrt n\bar X$, since $$Z_n=(AX)_n=\sum_iA_{ni}X_i=\sum\frac1{\sqrt n} X_i=\frac1{\sqrt n}n\bar X.$$
  3. The sum $\sum_{i=1}^nZ_i^2$ equals $\sum_{i=1}^nX_i^2$, since $$\sum_{i=1}^nZ_i^2=Z^TZ=(AX)^T(AX)=X^TA^TAX=X^TX=\sum_{i=1}^nX_i^2.$$
  4. The variable $\sum_{i=1}^{n-1}Z_i^2$ equals $\sum_{i=1}^nZ_i^2-Z_n^2=\sum_{i=1}^nX_i^2-n(\bar X)^2$ by (2) and (3). This last can be rewritten as $\sum_{i=1}^n(X_i-\bar X)^2$.

We conclude that the pair $(\sqrt n\bar X,\sum_{i=1}^n(X_i-\bar X)^2)$ is identical to the pair $(Z_n,\sum_{i=1}^{n-1}Z_i^2)$, which by (1) has the same distribution as $(X_n,\sum_{i=1}^{n-1}X_i^2)$.

This result can be generalized: Let $X_1,\ldots,X_n$ be iid, each with standard normal distribution. If $\alpha_1,\ldots,\alpha_n$ are constants with $\sum\alpha_i^2=1$, then $\sum_{i=1}^nX_i^2-\left(\sum_{i=1}^n\alpha_iX_i\right)^2$ has chi-squared distribution with $n-1$ degrees of freedom and is independent of $\sum\alpha_iX_i$, which has standard normal distribution.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.