Why don't we use restricted maximum likelihood to estimate parameters in non-mixed models?
$\begingroup$
$\endgroup$
1
-
3$\begingroup$ You use it: REML is for estimating variance parameters, and in non-mixed models there is only one such, $\sigma^2$. When dividing by $n-1$ to get a variance estimate, you are using REML. See stats.stackexchange.com/questions/48671/… $\endgroup$– kjetil b halvorsen ♦Commented May 4, 2015 at 17:02
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
For REML under fixed effects the variance estimator equivalent to $S^2=\frac{\sum_{i=1}^n (y_i - x_i \hat{\beta})^2}{n-p}$ where $\hat{\beta}$ is the OLS estimator, but as a bonus with OLS you also get to use Gauss-Markov Theorem for $\hat{\beta}$.
With REML you don't get any coefficient estimates.
