3
$\begingroup$

Can I use the Fisher information matrix to derive estimators for variances of unknown parameters?

I know that for the Fisher information (non-matrix form) the variance of, say, $\theta$ is given by the inverse of $J(\theta)$ (the Fisher information function).

But what about for multiple (e.g. 2) variables and the Fisher information matrix form?

$\endgroup$
  • 2
    $\begingroup$ In case the fisher information $J(\theta)$ is a matrix the size $n \times n$ with $n > 1$ the variance of the parameters are still given by the inverse of the fisher information. i.e. $J(\theta)^{-1}$. However, inverting a matrix is slightly more tricky than inverting a scalar. You need to find the matrix $B$ whose matrix-product with $J(\theta)$ results in the identity matrix $I$. $\endgroup$ – Harald Thomson Feb 24 '16 at 10:17
3
$\begingroup$

The variances of the parameters (of a distribution that's compatible with the requirements of Fisher information) are found from the diagonal of the inverse $J(\theta)^{-1}$ of the Fisher information matrix.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Variances of what precisely? $\endgroup$ – Scortchi - Reinstate Monica Feb 24 '16 at 12:28
  • $\begingroup$ "Variables of a distribution" isn't a standard term: if you mean "parameters of a distribution", as suggested by the $\theta$ notation in your question, then these are unknown constants & don't have a variance. Fisher information relates to the variance of estimators, but how? & to which estimators? $\endgroup$ – Scortchi - Reinstate Monica Feb 24 '16 at 13:27
  • 2
    $\begingroup$ Not really: as I said, parameters don't have variances because they're not random variables. $\endgroup$ – Scortchi - Reinstate Monica Feb 24 '16 at 14:14

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.