Multi-level (nested) correlation structures via geepack package

You may get lucky, but generally such package-specific questions don't get
responses here. There are about 20000 packages after all. You might do
better posting on the r-sig-mixed-models list or by asking the package
maintainer (?maintainer) whether there is some sort of support list for the
package.

Bert Gunter

"The trouble with having an open mind is that people keep coming along and
sticking things into it."
-- Opus (aka Berkeley Breathed in his "Bloom County" comic strip )


On Sun, Jul 12, 2020 at 9:59 AM Phat Chau <phat.chau at mail.utoronto.ca>
wrote:

Hello,

I have a multi-level, cohort dataset with three levels: repeat measures of
a response (level 1), that are collected from individual participants
(level 2) who are students within a school (level 3). I would like to do a
generalized estimating equation (GEE) analysis of this clustered data, but
to do so I need to specify ?nested? correlation structures (e.g.
exchangeable, compound symmetric, Toeplitz) to account for the
within-individual and within-cluster correlations.

Here is a reference paper that describes a nested exchangeable correlation
structure and nested compound symmetry:
doi:10.1111/j.1541-0420.2009.01374.x.

The geepack is available in R to do GEE analyses, but it seems to me that
it only allows the user to specify a correlation structure via the
geepack(??corstr = ?) option which only accounts for the within-individual
correlations (that arise from repeated measures). Would it be possible to
specify the nested correlation structures that I refer to here to also
account for the within-cluster correlations using this package?

Thank you,

Edward


        [[alternative HTML version deleted]]

______________________________________________
R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.

Hi

Your choice of package should partly depend on the type of dependent 
variable or Y that you are going to be dealing with
categorical/ordinal data may involve different packages than continuous 
or binary data see multgee for one.
The number of samples can also make a difference GEE with the "correct 
model" should normally have no problems with numbers 30-40; 25 or less 
would normally require corrections and a diffence package.

The doi for multgee  paper is 10.1111/biom.12054 and Touloumis paper in 
Journal of Statistical Software

For longitudinal data there is the following doi:
10.2307/2531248
and
10.1097/EDE.0b013e3181caeb90
10.1093/biomet/90.1.29
10.1007/s00362-017-0881-0
10.1002/sim.2368

a search for gee in the list of available packages should show you the 
alternatives.

As a check of the result do the statistics on another package. I 
remember doing a simple gee with an example
from a book using 4 different packages 2 of which gave poor or 
unreasonable answers

Regards

Duncan

Duncan Mackay
Department of Agronomy and Soil Science
University of New England
ARMIDALE NSW 2351



------ Original Message ------
From: "Phat Chau" <phat.chau at mail.utoronto.ca>
To: "r-help at R-project.org" <r-help at R-project.org>; "sorenh at math.aau.dk" 
<sorenh at math.aau.dk>
Sent: Sunday, 12 Jul, 2020 At 11:52 PM
Subject: Re: [R]  Multi-level (nested) correlation structures via 
geepack package

Hello,

I have a multi-level, cohort dataset with three levels: repeat measures 
of a response (level 1), that are collected from individual participants 
(level 2) who are students within a school (level 3). I would like to do 
a generalized estimating equation (GEE) analysis of this clustered data, 
but to do so I need to specify ?nested? correlation structures (e.g. 
exchangeable, compound symmetric, Toeplitz) to account for the 
within-individual and within-cluster correlations.

Here is a reference paper that describes a nested exchangeable 
correlation structure and nested compound symmetry: 
doi:10.1111/j.1541-0420.2009.01374.x.

The geepack is available in R to do GEE analyses, but it seems to me 
that it only allows the user to specify a correlation structure via the 
geepack(??corstr = ?) option which only accounts for the 
within-individual correlations (that arise from repeated measures). 
Would it be possible to specify the nested correlation structures that I 
refer to here to also account for the within-cluster correlations using 
this package?

Thank you,

Edward



______________________________________________
R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.

Hi Duncan, 

Thank you for your message. All of those citations and suggestions are wonderful and highly relevant for GEEs. My concern is that I need multi-level/nested correlation structures, whereas those handle the specification of covariance matrices at the level of repeated observations within-subject. I think the paper attached contains the theory behind such nested correlation structures, but the coding for it is not as apparent.

Edward

Hi
    
    Your choice of package should partly depend on the type of dependent 
    variable or Y that you are going to be dealing with
    categorical/ordinal data may involve different packages than continuous 
    or binary data see multgee for one.
    The number of samples can also make a difference GEE with the "correct 
    model" should normally have no problems with numbers 30-40; 25 or less 
    would normally require corrections and a diffence package.
    
    The doi for multgee  paper is 10.1111/biom.12054 and Touloumis paper in 
    Journal of Statistical Software
    
    For longitudinal data there is the following doi:
    10.2307/2531248
    and
    10.1097/EDE.0b013e3181caeb90
    10.1093/biomet/90.1.29
    10.1007/s00362-017-0881-0
    10.1002/sim.2368
    
    a search for gee in the list of available packages should show you the 
    alternatives.
    
    As a check of the result do the statistics on another package. I 
    remember doing a simple gee with an example
    from a book using 4 different packages 2 of which gave poor or 
    unreasonable answers
    
    Regards
    
    Duncan
    
    Duncan Mackay
    Department of Agronomy and Soil Science
    University of New England
    ARMIDALE NSW 2351
    
    
    
    ------ Original Message ------
    From: "Phat Chau" <phat.chau at mail.utoronto.ca>
    To: "r-help at R-project.org" <r-help at R-project.org>; "sorenh at math.aau.dk" 
    <sorenh at math.aau.dk>
    Sent: Sunday, 12 Jul, 2020 At 11:52 PM
    Subject: Re: [R]  Multi-level (nested) correlation structures via 
    geepack package
    
    Hello,
    
    I have a multi-level, cohort dataset with three levels: repeat measures 
    of a response (level 1), that are collected from individual participants 
    (level 2) who are students within a school (level 3). I would like to do 
    a generalized estimating equation (GEE) analysis of this clustered data, 
    but to do so I need to specify ?nested? correlation structures (e.g. 
    exchangeable, compound symmetric, Toeplitz) to account for the 
    within-individual and within-cluster correlations.
    
    Here is a reference paper that describes a nested exchangeable 
    correlation structure and nested compound symmetry: 
    doi:10.1111/j.1541-0420.2009.01374.x.
    
    The geepack is available in R to do GEE analyses, but it seems to me 
    that it only allows the user to specify a correlation structure via the 
    geepack(??corstr = ?) option which only accounts for the 
    within-individual correlations (that arise from repeated measures). 
    Would it be possible to specify the nested correlation structures that I 
    refer to here to also account for the within-cluster correlations using 
    this package?
    
    Thank you,
    
    Edward
    
    
    
    ______________________________________________
    R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
    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.