R help - May 2010 - How can I fit a fixed-effect linear model or generalized linear model with method="ml"?

Hi,

I want to fit a linear model (without any random effect) with method
"ml". I
tried to use "glm"  I found that there is no option for "ml"
or "reml" and
the default one is "reml". THen I tried to use "lme" but it
requires a
random effect. How can I fix this problem?

Of course, it's not necessary to be "glm" or "lme", I am
just looking for
commands which allow me to fit a linear model (without any random effect)
with "ml" method. And later I may also need to fit generalized linear
models
(probit, gamma and negtive binomial) with method "ml". Any
suggestions?

Thanks.

	[[alternative HTML version deleted]]

On Sun, 30 May 2010, Sherrie Jin wrote:
> Hi,
>
> I want to fit a linear model (without any random effect) with method
"ml". I
> tried to use "glm"  I found that there is no option for
"ml" or "reml" and
> the default one is "reml". THen I tried to use "lme"
but it requires a
> random effect. How can I fix this problem?
>
> Of course, it's not necessary to be "glm" or "lme",
I am just looking for
> commands which allow me to fit a linear model (without any random effect)
> with "ml" method. And later I may also need to fit generalized
linear models
> (probit, gamma and negtive binomial) with method "ml". Any
suggestions?
glm() always performs Maximum Likelihood estimation and thus has no 
method="ml" argument.
Z
> Thanks.
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

Why are you not using the lm() function? OLS is ML with no random effects
________________________________________
From: r-help-bounces at r-project.org [r-help-bounces at r-project.org] On
Behalf Of Sherrie Jin [jingoosy at gmail.com]
Sent: Sunday, May 30, 2010 5:05 AM
To: r-help at r-project.org
Subject: [R] How can I fit a fixed-effect linear model or generalized linear   
model with method="ml"?

Hi,

I want to fit a linear model (without any random effect) with method
"ml". I
tried to use "glm"  I found that there is no option for "ml"
or "reml" and
the default one is "reml". THen I tried to use "lme" but it
requires a
random effect. How can I fix this problem?

Of course, it's not necessary to be "glm" or "lme", I am
just looking for
commands which allow me to fit a linear model (without any random effect)
with "ml" method. And later I may also need to fit generalized linear
models
(probit, gamma and negtive binomial) with method "ml". Any
suggestions?

Thanks.

        [[alternative HTML version deleted]]

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

Thanks for your reply.
I should have made my question clearer. Yes, I want to estimate standard
errors with the method "ml" rather than "reml". According to
your
suggestion, I looked at the source code for "summary.glm". Now I know
it's
"summary.glm" rather than "glm"  itself estimates the
standard errors. So it
must be "summary.glm" that uses "reml" as a default to
compute standard
errors.

But when I read the source code for "summary.glm", I'm still
confused. How
can I modify the code to re-compute the standard errors with method
"reml"?
I'm pretty new to R and haven't write any code by myself. Thanks.

On Sun, May 30, 2010 at 8:01 AM, Prof Brian Ripley
<ripley@stats.ox.ac.uk>wrote:
> On Sun, 30 May 2010, Achim Zeileis wrote:
>
>  On Sun, 30 May 2010, Sherrie Jin wrote:
>>
>>  Hi,
>>>
>>> I want to fit a linear model (without any random effect) with
method
>>> "ml". I
>>> tried to use "glm"  I found that there is no option for
"ml" or "reml"
>>> and
>>> the default one is "reml". THen I tried to use
"lme" but it requires a
>>> random effect. How can I fix this problem?
>>>
>>> Of course, it's not necessary to be "glm" or
"lme", I am just looking for
>>> commands which allow me to fit a linear model (without any random
effect)
>>> with "ml" method. And later I may also need to fit
generalized linear
>>> models
>>> (probit, gamma and negtive binomial) with method "ml".
Any suggestions?
>>>
>>
>> glm() always performs Maximum Likelihood estimation and thus has no
>> method="ml" argument.
>>
>
> ML of what?  For the coefficients, yes.  However, for parameters in the
> distribution estimated via the dispersion, not usually and not in the case
> of sigma^2 for the gaussian.
>
> Strictly, neither lm() not glm() estimate the dispersion, but their
> summary() methods do.  However, you can use your own estimate of dispersion
> and supply it to summary.glm (but not summary.lm): you can also use gls()
> from package nlme.  MASS (the book) discusses how to get MLE for a gamma
> glm, for example, and MASS::glm.nb fits a negative binomial glm with
> parameters estimated by ML.
>
> --
> Brian D. Ripley,                  ripley@stats.ox.ac.uk
> Professor of Applied Statistics, 
http://www.stats.ox.ac.uk/~ripley/<http://www.stats.ox.ac.uk/%7Eripley/>
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272866 (PA)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595
>
	[[alternative HTML version deleted]]