[R-sig-ME] Suggestions on how to correct a misapprehension?

[Martin Maechler]

>>>>> Jeroen van Paridon via R-sig-mixed-models 
>>>>>     on Wed, 14 Dec 2022 02:11:46 +0000 writes:

    > Hi Ben, It seems like Python's statsmodels might not be
    > using EM either, which would mean that exactly none of the
    > four stats packages listed in the original Wikipedia
    > article actually use EM.

That's not correct, either:

nlme  clearly uses *BOTH*  EM (initial steps) *and*
Newton-Raphson (==> non-sense in the SAS docu you cited).

Note the reference on lme's help page,

  Lindstrom, M.J. and Bates, D.M. (1988) "Newton-Raphson and EM
  Algorithms for Linear Mixed-Effects Models for Repeated-Measures
  Data", Journal of the American Statistical Association, 83, 1014--1022. 

which even has both algorithms in the title.


    > My edits were a mix of Phillip Alday's thoughts on the
    > matter and mine; if I can work out exactly what
    > statsmodels is doing I'll make further changes. (And I'm
    > open to suggestions about anything else that might be good
    > to add!

I've also amended Wikipedia pages (in my name) since 2006, as I
now looked up...

The current entry about lme4 / Mixed Models is still "grossly" misleading,
in my view.  Notably as it basically relates everything to
Henderson's MM equations -- which were phantastic when stated
63 years ago.

The main idea of Doug Bates' approach (in lme4, MixedModels) --
which Doug explained nicely -- namely that you can profile out much of the
parameter space (beta, sigma) via PLS while only needing to optimize
(the PLS resulting profiled likelihood) over the var-cov
parameters (theta) is still not all mentioned. 

Martin


    > Cheers,

    > JP ________________________________ From:
    > R-sig-mixed-models
    > <r-sig-mixed-models-bounces using r-project.org> on behalf of
    > Ben Bolker <bbolker using gmail.com> Sent: Tuesday, December 13,
    > 2022 7:38:16 PM To: r-sig-mixed-models using r-project.org
    > <r-sig-mixed-models using r-project.org> Subject: Re: [R-sig-ME]
    > Suggestions on how to correct a misapprehension?

    >    FWIW I can see that JP van Paridon has already gotten
    > us most of the way there.

    > FWIW I don't think SAS uses EM either, at least as far as
    > I can tell:

    > https://documentation.sas.com/doc/en/statcdc/14.2/statug/statug_mixed_details58.htm

    > PROC MIXED uses a ridge-stabilized Newton-Raphson
    > algorithm to optimize either a full (ML) or residual
    > (REML) likelihood function. The Newton-Raphson algorithm
    > is preferred to the EM algorithm (Lindstrom and Bates
    > 1988). PROC MIXED profiles the likelihood with respect to
    > the fixed effects and also with respect to the residual
    > variance whenever it appears reasonable to do so. The
    > residual profiling can be avoided by using the NOPROFILE
    > option of the PROC MIXED statement. PROC MIXED uses the
    > MIVQUE0 method (Rao 1972; Giesbrecht 1989) to compute
    > initial values.

    > On 2022-12-13 8:25 p.m., Stuart Luppescu wrote:
    >> On 12/14/22 02:39, Douglas Bates wrote:
    >>> I have never gone through the process of proposing an
    >>> edit in a Wikipedia article.  As I understand it I would
    >>> need to create a login etc.  Would anyone who does have
    >>> such a login be willing to propose an edit and save me
    >>> the steps?
    >> 
    >> I was in this situation recently. I found AN ERROR in
    >> Wikipedia and resolved to fix it, but the Wikipedia
    >> admins wouldn't let me. I had been logged into my VPN on
    >> DigitalOcean, and they disable all attempts to edit
    >> articles from DigitalOcean. I had to log out of the VPN,
    >> and erase my Wikipedia cookies before I could correct the
    >> entry. In any case, I think I have a Wikipedia login, but
    >> I couldn't remember it so I just did the edit as
    >> "guest". If you want to be able to talk about your edit
    >> with other people post hoc, I think you might need a
    >> login. Otherwise, just do it without a login.
    >> 

    > _______________________________________________
    > R-sig-mixed-models using r-project.org mailing list
    > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

    > 	[[alternative HTML version deleted]]

    > _______________________________________________
    > R-sig-mixed-models using r-project.org mailing list
    > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models