Given two samples: $X_1,X_2,\ldots,X_m$ from a normal population with mean $\mu_1$ and variance $\sigma^2$ and $Y_1,Y_2,\ldots,Y_n$ from a different normal population with mean $\mu_2$ but the same variance $\sigma^2$, find the maximum likelihood estimator for the common variance $\sigma^2$, $\mu_1$ and $\mu_2$.

The solution to this problem gives the likelihood function as $$L(\sigma^2)=(2\pi\sigma^2)^{\frac{-(m+n)}{2}}\exp\left(-\frac{1}{2\sigma^2}\left [\sum_{i=1}^m (x_i-\mu_1)^2+\sum_{i=1}^n (y_i-\mu_2)^2\right]\right)$$

And say $$\hat{u_1}=\bar{x}\text{ and } \hat{u_2}=\bar{y}$$ Then if $\ell$ is the log-likelihood function, then by setting $$\frac{\partial \ell}{\partial \sigma^2}=0$$ we have $$\hat{\sigma^2}=\frac{1}{m+n} \left(\displaystyle\sum_{i=1}^n (x_i-\mu_1)^2 + \displaystyle\sum_{j=1}^m (y_j-\mu_2)^2\right)$$ Then check $$\frac{\partial^2 \ell}{\partial (\sigma^2)^2}|_{\hat{\sigma^2}}<0$$ We say we attainded maximum at $\hat{\sigma^2}, \hat{u_1}, \hat{u_2}$.

My question is why we don't use the $3 \times 3$ Hessian Matrix to check whether we attained the maximum value? I actually checked the determinant of this Hessian matrix, it is zero. Why we can simply say that $\hat{u_1}$ and $\hat{u_2}$ is the MLE here?


We should indeed check the Hessian to be sure.

The hessian matrix evaluated those values that I computed is of the form of $$\begin{bmatrix} -\frac{n}{\hat{\sigma^2}} & 0 & 0\\ 0 & -\frac{m}{\hat{\sigma^2}} & 0 \\ 0 & 0 & 2\frac{(m+n)}{\hat{\sigma^2}}- 4 \sum_{i=1}^n \frac{(x_i-\hat{\mu}_1)^2}{\hat{\sigma^4}}- 4 \sum_{j=1}^m \frac{(y_j-\hat{\mu}_2)^2}{\hat{\sigma^4}}\end{bmatrix}.$$

We claim that this matrix is negative definite.

Observe that $\hat{\sigma^2} < 2 \hat{\sigma^2}$ if $\hat{\sigma} > 0$.

Hence, $\hat{\sigma^2} < \frac4{2(m+n)}\left( \sum_{i=1}^n (x_i-\hat{\mu}_1)^2+ \sum_{j=1}^m(y_j-\hat{\mu}_2)^2\right)$

$$\frac{\hat{\sigma^4}}{\hat{\sigma^2}} < \frac4{2(m+n)}\left( \sum_{i=1}^n (x_i-\hat{\mu}_1)^2+ \sum_{j=1}^m(y_j-\hat{\mu}_2)^2\right)$$

$$\frac{2(m+n)}{\hat{\sigma^2}} < \frac4{\hat{\sigma^4}}\left( \sum_{i=1}^n (x_i-\hat{\mu}_1)^2+ \sum_{j=1}^m(y_j-\hat{\mu}_2)^2\right)$$

and hence the $(3,3)$-entry of the Hessian matrix is negative.

| cite | improve this answer | |

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.