Message-ID: <loom.20140327T033920-442@post.gmane.org>
Date: 2014-03-27T02:45:18Z
From: Ben Bolker
Subject: compare maximum likelihood and restricted maximum likelihood
Adianto Pangaribuan <adianto_pangaribuan at ...> writes:
>
> I want to compare maximum likelihood and restricted maximum
> likelihood for linear mixed effects model.
> Can you give me suggest how to compare the two estimation?
> If I use AIC, BIC and value of log-likelihood, the
> best estimation is MLE. But, I think i am wrong with it.
This doesn't sound on the face of it like a sensible question.
ML and REML are two different approaches to estimation -- they're not
directly comparable. Depending on your background I don't know what
analogy will make sense, but "which fits better, least-squares regression
or logistic regression?" might be a similar question. They're
different criteria.
REML generally gives a less biased estimate of the variance components.
ML gives estimates that are comparable across models with different
fixed-effect components.
It's conceivable that a more extensive description of what you're
trying to do would convince us that this is indeed a sensible question
(and provide you with more useful feedback) -- although I doubt it.
sincerely
Ben Bolker