Maximum Likelihood Estimation on a standardized normal variable

95 Views Asked by At

I have a random variable $Z$ ~ $N(\mu,\sigma)$ for which I can compute a likelihood using the normal pdf with mean = mu and standard deviation = sigma.

I could also standardize $Z$ into $Z^*$ by doing $Z^* = (Z-\mu)/\sigma$, for which I can then also compute a likelihood using the normal pdf with mean = 0 and std = 1.

The results are not the same. That is, $normpdf(Z,\mu,\sigma)$ is not equal to $normpdf(Z^*,0,1)$. My understanding is that this is normal, but please confirm.

My question is: would a Maxmimum Likelihood Estimation of the parameters yield the same results for both even though the (log) likelihood is not the same?

1

There are 1 best solutions below

0
On BEST ANSWER

Ignoring the MLE part of the question, the change in density function for a location-scale change is not difficult, though with a normal distribution you have to be careful not to confuse the variance and standard deviation

$\hat Z$ is the maximum likelihood estimator in the original case if and only if $\hat Z^* =(\hat Z-\mu)/\sigma $ is the maximum likelihood estimator in the standardised case. This property is called functional equivalence and works for any bijective function