s-news
[Top] [All Lists]

Re: Interpreting glmmpql results.

To: "Thomas Jagger" <tjagger@blarg.net>
Subject: Re: Interpreting glmmpql results.
From: "Michael O'Connell" <moconnell@insightful.com>
Date: Tue, 17 May 2005 12:37:48 -0400
Cc: <s-news@lists.biostat.wustl.edu>
Thread-index: AcVXFoRn4lZIeEJdSzaRxwl3iWCgnwAAHVLwAAFseiAA+A6cHQ==
Thread-topic: [S] Interpreting glmmpql results.
Thomas
 
thanks for the clarification! One correction is that I believe glmmPQL was
developed by Brian Ripley and Bill Venables; as part of the fabulous MASS
library - Software and datasets to support `Modern Applied Statistics with
S', fourth edition, by W. N. Venables and B. D. Ripley. Springer, 2002, ISBN
0-387-95457-0.
 
Im looking forward to an interesting webinar on generalized linear mixed
models this week
http://www.insightful.com/news_events/webcasts/2005/05novartis/default.asp
 
Michael


        -----Original Message----- 
        From: Thomas Jagger [mailto:tjagger@blarg.net] 
        Sent: Thu 5/12/2005 3:12 PM 
        To: Michael O'Connell 
        Cc: s-news@lists.biostat.wustl.edu 
        Subject: RE: [S] Interpreting glmmpql results.
        
        

        For those who do not have Splus 7,
        glmmPQL is available with the MASS library. Here is a comparable
result
        using both:
        
        glme is from the  correlated data library and
        glmmPQL is from the MASS library.
        
        Both glmmPQL and glme are from Jose Pinheiro, as I understand, with
glmmPQL
        generated from iterative calls to lme.
        Thus, they should and do on this one example, produce consistent
results.
        
        However, Splus Version 7.0 glme has additional methods such as
LAPLACE and
        AGQUAD for likelihood approximations. However, in this case the
Residual
        variance is 1.0.
        
        summary(glme(y ~ trt + I(week > 2), random = ~ 1 |
        ID,family=binomial,data=bacteria,method="PQL"))
        Generalized linear mixed-effects model fit by PQL
        Family: Binomial with Logit link
          Data: bacteria
              AIC      BIC    logLik
          1113.68 1134.042 -550.8402
        
        Random effects:
         Formula:  ~ 1 | ID
                (Intercept)  Residual
        StdDev:    1.409806 0.7801296
        
        Fixed effects: y ~ trt + I(week > 2)
                        Value Std.Error  DF   t-value p-value
        (Intercept)  1.940794 0.2810486 169  6.905548  <.0001
               trt1 -0.623590 0.3219447  47 -1.936949  0.0588
               trt2 -0.043518 0.1911304  47 -0.227687  0.8209
        I(week > 2) -0.803483 0.1791946 169 -4.483858  <.0001
        
        Deviance Within-Group Residuals:
               Min        Q1      Med        Q3      Max
         -2.391251 0.1730478 0.380906 0.5270572 1.447784
        
        Number of Observations: 220
        Number of Groups: 50
        > summary(glmmPQL(y ~ trt + I(week > 2), random = ~ 1 |
        ID,family=binomial,data=bacteria))
        iteration 1
        iteration 2
        iteration 3
        iteration 4
        iteration 5
        iteration 6
        Linear mixed-effects model fit by maximum likelihood
         Data: bacteria
               AIC      BIC    logLik
          1113.622 1133.984 -550.8111
        
        Random effects:
         Formula:  ~ 1 | ID
                (Intercept)  Residual
        StdDev:    1.410632 0.7800514
        
        Variance function:
         Structure: fixed weights
         Formula:  ~ invwt
        
        Fixed effects: y ~ trt + I(week > 2)
                        Value Std.Error  DF   t-value p-value
        (Intercept)  1.941156 0.2811167 169  6.905160  <.0001
               trt1 -0.623677 0.3220311  47 -1.936699  0.0588
               trt2 -0.043550 0.1911861  47 -0.227787  0.8208
        I(week > 2) -0.803628 0.1791689 169 -4.485307  <.0001
        
        Standardized Within-Group Residuals:
               Min        Q1       Med        Q3      Max
         -5.198533 0.1572341 0.3513082 0.4949478 1.744882
        
        Number of Observations: 220
        Number of Groups: 50
        >
        
        -----Original Message-----
        From: s-news-owner@lists.biostat.wustl.edu
        [mailto:s-news-owner@lists.biostat.wustl.edu] On Behalf Of Michael
O'Connell
        Sent: Thursday, May 12, 2005 11:44 AM
        To: Clegg, Philip; s-news@lists.biostat.wustl.edu
        Subject: Re: [S] Interpreting glmmpql results.
        
        1) I dont think glmmpql is a function that ships with S-PLUS 7. Is
this from
        a user-contributed package? I strongly recommend that you use the new
        correlatedData library in S-PLUS 7. There is help and doc in the
library
        directory eg in:
        C:\Program Files\Insightful\splus70\library\correlatedData
        as a basic install
        
        The correlatedData library includes PQL ( glme() ); and exact
(quadrature)
        methods for binomial and Poisson regression. PQL generally works
well,
        contrary to what some folks say :-) See Breslow and Clayton (1993)
and
        Wolfinger and O'Connell (1993) for details. The latter paper has a
more
        intuitive derivation and explanation even if I do say so :-)... And
PQL is
        implemented in the correlatedData library in a way that improves upon
an
        earlier SAS MACRO implementation...
        
        Mixed models with non-Gaussian responses is a very interesting,
important
        and
        expansive area! Please join us at next weeks (free) webinar on this
topic,
        given by Jose Pinheiro of Novartis who was deeply involved in the
        development
        of glme() and correlatedData:
        
http://www.insightful.com/news_events/webcasts/2005/05novartis/default.asp
        
        2) In general, to set a factor variable in S-PLUS, set
options()$contrast to
        contrast.treatment or something. Look at the help on contrast and/or
search
        the list archive.  
        
        3) Lets get on with our statistical software discussions on the list
- it is
        a great user-contributed S-PLUS/stats list and everyone is making
very
        valuable contributions. So lets keep it coming. In particular, if
anyone
        develops any interesting applications of glme() and the
correlatedData
        library please let me know - I have a goal to pull together a
collection of
        applications in this area over the coming year.
        
        Good luck !
        
        Michael
        
        Michael O'Connell, Ph.D.
        Director, Life Science Solutions   W: (800) 569-0123 x454
        Insightful Corporation                              
        moconnell@insightful.com           www.insightful.com
        
        
        -----Original Message-----
        From: Clegg, Philip [mailto:P.Clegg@leedsmet.ac.uk]
        Sent: Thursday, May 12, 2005 11:20 AM
        To: 's-news@lists.biostat.wustl.edu'
        Subject: [S] Interpreting glmmpql results.
        
        
        
        I see that in the recent period, there have been many postings about
list
        etiquette,  the appropriateness of questions asked, and the need to
avoid
        trivial questions. I recently asked  about why glmmpql seems to treat
two
        level factors  as  continuous variables but received no replies. I
hope this
        does not mean my question is trivial. I am also concerned that I have
not
        asked the question in the right way. But what is the right way to ask
the
        question.
        It is difficult to formulate questions in appropriate statistical
language
        if your own are of expertise is an a non-mathematical or statistical
        subject. I  want to use glmmpql as a tool to find out if there is a
spatial
        component to suicide. I doubt that I will ever understand how the
tool was
        constructed - just as I have no real understanding of what lies
underneath
        the bonnet/hood of my car.
        
        Best wishes,
        Phil
        
        
        
        -----Original Message-----
        From: Clegg, Philip [mailto:P.Clegg@leedsmet.ac.uk]
        Sent: 10 May 2005 12:38
        To: 's-news@lists.biostat.wustl.edu'
        Subject: [S] Numerical codes in GlMMPQL
        
        
        I am attempting to fit a glmmpql model on suicide case history data
using
        postcode as a spatial random effect. Most of the variables I wish to
test
        are in binary form with codes 0, 1 and -1 for missing data. In the
following
        example, the p-value for the intercept is highly significant
indicating that
        there is a spatial postcode effect. The relationship between Gender
and
        'notreatz' (whether or not cases had treatment for mental illness in
the
        last year) is also significant. However, my problem is that gender
does not
        appear to have been treated as a factor but as a continuous variable.
I
        guess the problem is that S-Plus does not recognise my variables as
        categorical. I have tried using as.factor but that has not helped.
Has
        anyone any advice please? Best wishes, Phil
        
        
        > summary(glmmPQL(notreatz ~ genderz, family = binomial, random =  ~
1 |
                POSTCODE))
        iteration 1
        Linear mixed-effects model fit by maximum likelihood
         Data: NULL
               AIC      BIC    logLik
          1026.248 1040.103 -509.1238
        
        Random effects:
         Formula:  ~ 1 | POSTCODE
                (Intercept)  Residual
        StdDev:  0.01600788 0.9999703
        
        Variance function:
         Structure: fixed weights
         Formula:  ~ invwt
        Fixed effects: notreatz ~ genderz
                        Value Std.Error  DF  t-value p-value
        (Intercept) 0.7145204 0.1805515 229 3.957432  0.0001
            genderz 0.5382434 0.1805512   5 2.981112  0.0308
        
        Standardized Within-Group Residuals:
               Min        Q1       Med        Q3       Max
         -1.870798 -1.092102 0.5345133 0.9156031 0.9156726
        
        Number of Observations: 236
        Number of Groups: 230
        
        
        
        
        To view the terms under which this email is distributed, please go to
        http://disclaimer.leedsmet.ac.uk/email.htm
        --------------------------------------------------------------------
        This message was distributed by s-news@lists.biostat.wustl.edu.  To
        unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu
with the
        BODY of the message:  unsubscribe s-news
        
        
        To view the terms under which this email is distributed, please go to
        http://disclaimer.leedsmet.ac.uk/email.htm
        --------------------------------------------------------------------
        This message was distributed by s-news@lists.biostat.wustl.edu.  To
        unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu
with
        the BODY of the message:  unsubscribe s-news
        --------------------------------------------------------------------
        This message was distributed by s-news@lists.biostat.wustl.edu.  To
        unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu
with
        the BODY of the message:  unsubscribe s-news
        
        

<Prev in Thread] Current Thread [Next in Thread>