MCMCglmm multinomial model results

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

MCMCglmm multinomial model results

Michelle Kline
Hi,

Thanks in advance for any help on this question. I'm running multinomial
models using the MCMCglmm package. The models have 5 outcome variables
(each with count data), and an additional two random effects built into the
models. The issue is that when I use the following code, the summary only
gives me results for four of the outcome variables.

Here is the code for my model:

m3.random <- MCMCglmm(cbind(Opp_teacher , Dir_teacher, Enh_teacher,
SocTol_teacher, Eval_teacher) ~ trait -1,
               random = ~ us(trait):other + us(trait):focal,
               rcov = ~ us(trait):units,
               prior = list(
                 R = list(fix=1, V=0.5 * (I + J), n = 4),
                 G = list(
                   G1 = list(V = diag(4), n = 4),
                   G2 = list(V = diag(4), n = 4))),
               burnin = burn,
               nitt = iter,
               family = "multinomial5",
               data = data,
               pr=TRUE,
               pl=TRUE,
               DIC = TRUE,
               verbose = FALSE)

And the summary of the main effects:

post.mean  l-95% CI  u-95% CI eff.samp        pMCMC
traitOpp_teacher    -3.828752 -4.616731 -3.067424 184.4305 5.263158e-05
traitDir_teacher    -3.400481 -4.041069 -2.813063 259.1084 5.263158e-05
traitEnh_teacher    -1.779129 -2.197415 -1.366496 624.9759 5.263158e-05
traitSocTol_teacher -2.852684 -3.429799 -2.332909 468.7098 5.263158e-05


It is not an issue of the suppressing the intercept, since I'm already
doing that (see the -1 term. When I remove that term, the model solutions
includes an intercept and only 3 additional main effects).

The model does throw the following error, but after searching previous
messages on this list, I've concluded that this error message doesn't have
to do with  my current problem. Just in case: " observations with zero
weight not used for calculating dispersion"

I have also posted a similar question on stackoverflow about a week ago,
but with no response, so I thought I would try here. Link in case people
want to gain reputation points for a
response: https://stackoverflow.com/questions/49309027/missing-term-in-mcmcglmm-multinomial-model-results-not-in-intercept-issue
<https://stackoverflow.com/questions/49309027/missing-term-in-mcmcglmm-multinomial-model-results-not-in-intercept-issue>

And of course I've checked various other sources including the course
notes, but can't make sense of why the 5th term is dropped from the model.
Any help is much appreciated.

Best,

Michelle

--
Michelle A. Kline, PhD

Assistant Professor
Department of Psychology
Simon Fraser University

        [[alternative HTML version deleted]]

______________________________________________
[hidden email] 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.
Reply | Threaded
Open this post in threaded view
|

Re: MCMCglmm multinomial model results

David Winsemius

> On Mar 22, 2018, at 1:31 PM, Michelle Kline <[hidden email]> wrote:
>
> Hi,
>
> Thanks in advance for any help on this question. I'm running multinomial
> models using the MCMCglmm package. The models have 5 outcome variables
> (each with count data), and an additional two random effects built into the
> models. The issue is that when I use the following code, the summary only
> gives me results for four of the outcome variables.
>
> Here is the code for my model:
>
> m3.random <- MCMCglmm(cbind(Opp_teacher , Dir_teacher, Enh_teacher,
> SocTol_teacher, Eval_teacher) ~ trait -1,
>               random = ~ us(trait):other + us(trait):focal,
>               rcov = ~ us(trait):units,
>               prior = list(
>                 R = list(fix=1, V=0.5 * (I + J), n = 4),
>                 G = list(
>                   G1 = list(V = diag(4), n = 4),
>                   G2 = list(V = diag(4), n = 4))),
>               burnin = burn,
>               nitt = iter,
>               family = "multinomial5",
>               data = data,

We have no way to debug this without the data. Perhaps you should contact the maintainer and in your message attach the data?

 maintainer('MCMCglmm')
[1] "Jarrod Hadfield <[hidden email]>"


An equally effective approach would be to post (again with data that reproduces the error)  on the R-SIG-mixed-models mailing list since Hadfield is a regular contributor on that list. (To me it suggests not an error since you got output but rather a warning. Generally warnings and errors are properly labeled so you may not have included the full output.)

--
David.

>               pr=TRUE,
>               pl=TRUE,
>               DIC = TRUE,
>               verbose = FALSE)
>
> And the summary of the main effects:
>
> post.mean  l-95% CI  u-95% CI eff.samp        pMCMC
> traitOpp_teacher    -3.828752 -4.616731 -3.067424 184.4305 5.263158e-05
> traitDir_teacher    -3.400481 -4.041069 -2.813063 259.1084 5.263158e-05
> traitEnh_teacher    -1.779129 -2.197415 -1.366496 624.9759 5.263158e-05
> traitSocTol_teacher -2.852684 -3.429799 -2.332909 468.7098 5.263158e-05
>
>
> It is not an issue of the suppressing the intercept, since I'm already
> doing that (see the -1 term. When I remove that term, the model solutions
> includes an intercept and only 3 additional main effects).
>
> The model does throw the following error, but after searching previous
> messages on this list, I've concluded that this error message doesn't have
> to do with  my current problem. Just in case: " observations with zero
> weight not used for calculating dispersion"
>
> I have also posted a similar question on stackoverflow about a week ago,
> but with no response, so I thought I would try here. Link in case people
> want to gain reputation points for a
> response: https://stackoverflow.com/questions/49309027/missing-term-in-mcmcglmm-multinomial-model-results-not-in-intercept-issue
> <https://stackoverflow.com/questions/49309027/missing-term-in-mcmcglmm-multinomial-model-results-not-in-intercept-issue>
>
> And of course I've checked various other sources including the course
> notes, but can't make sense of why the 5th term is dropped from the model.
> Any help is much appreciated.
>
> Best,
>
> Michelle
>
> --
> Michelle A. Kline, PhD
>
> Assistant Professor
> Department of Psychology
> Simon Fraser University
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> [hidden email] 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.

David Winsemius
Alameda, CA, USA

'Any technology distinguishable from magic is insufficiently advanced.'   -Gehm's Corollary to Clarke's Third Law

______________________________________________
[hidden email] 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.
Reply | Threaded
Open this post in threaded view
|

Re: MCMCglmm multinomial model results

Michelle Kline
Hi David,

Thanks for your comment. I haven't posted the data because they are
unpublished and include human subjects so there are issues with sharing on
a list serv, but I thought perhaps someone had encountered a similar
problem and would already know the answer.

I will reconsider whether my University's ethics approval would allow me to
post the data and update the question if I think it is allowable.

Michelle

On Fri, Mar 23, 2018, 10:13 AM David Winsemius <[hidden email]>
wrote:

>
> > On Mar 22, 2018, at 1:31 PM, Michelle Kline <
> [hidden email]> wrote:
> >
> > Hi,
> >
> > Thanks in advance for any help on this question. I'm running multinomial
> > models using the MCMCglmm package. The models have 5 outcome variables
> > (each with count data), and an additional two random effects built into
> the
> > models. The issue is that when I use the following code, the summary only
> > gives me results for four of the outcome variables.
> >
> > Here is the code for my model:
> >
> > m3.random <- MCMCglmm(cbind(Opp_teacher , Dir_teacher, Enh_teacher,
> > SocTol_teacher, Eval_teacher) ~ trait -1,
> >               random = ~ us(trait):other + us(trait):focal,
> >               rcov = ~ us(trait):units,
> >               prior = list(
> >                 R = list(fix=1, V=0.5 * (I + J), n = 4),
> >                 G = list(
> >                   G1 = list(V = diag(4), n = 4),
> >                   G2 = list(V = diag(4), n = 4))),
> >               burnin = burn,
> >               nitt = iter,
> >               family = "multinomial5",
> >               data = data,
>
> We have no way to debug this without the data. Perhaps you should contact
> the maintainer and in your message attach the data?
>
>  maintainer('MCMCglmm')
> [1] "Jarrod Hadfield <[hidden email]>"
>
>
> An equally effective approach would be to post (again with data that
> reproduces the error)  on the R-SIG-mixed-models mailing list since
> Hadfield is a regular contributor on that list. (To me it suggests not an
> error since you got output but rather a warning. Generally warnings and
> errors are properly labeled so you may not have included the full output.)
>
> --
> David.
> >               pr=TRUE,
> >               pl=TRUE,
> >               DIC = TRUE,
> >               verbose = FALSE)
> >
> > And the summary of the main effects:
> >
> > post.mean  l-95% CI  u-95% CI eff.samp        pMCMC
> > traitOpp_teacher    -3.828752 -4.616731 -3.067424 184.4305 5.263158e-05
> > traitDir_teacher    -3.400481 -4.041069 -2.813063 259.1084 5.263158e-05
> > traitEnh_teacher    -1.779129 -2.197415 -1.366496 624.9759 5.263158e-05
> > traitSocTol_teacher -2.852684 -3.429799 -2.332909 468.7098 5.263158e-05
> >
> >
> > It is not an issue of the suppressing the intercept, since I'm already
> > doing that (see the -1 term. When I remove that term, the model solutions
> > includes an intercept and only 3 additional main effects).
> >
> > The model does throw the following error, but after searching previous
> > messages on this list, I've concluded that this error message doesn't
> have
> > to do with  my current problem. Just in case: " observations with zero
> > weight not used for calculating dispersion"
> >
> > I have also posted a similar question on stackoverflow about a week ago,
> > but with no response, so I thought I would try here. Link in case people
> > want to gain reputation points for a
> > response:
> https://stackoverflow.com/questions/49309027/missing-term-in-mcmcglmm-multinomial-model-results-not-in-intercept-issue
> > <
> https://stackoverflow.com/questions/49309027/missing-term-in-mcmcglmm-multinomial-model-results-not-in-intercept-issue
> >
> >
> > And of course I've checked various other sources including the course
> > notes, but can't make sense of why the 5th term is dropped from the
> model.
> > Any help is much appreciated.
> >
> > Best,
> >
> > Michelle
> >
> > --
> > Michelle A. Kline, PhD
> >
> > Assistant Professor
> > Department of Psychology
> > Simon Fraser University
> >
> >       [[alternative HTML version deleted]]
> >
> > ______________________________________________
> > [hidden email] 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.
>
> David Winsemius
> Alameda, CA, USA
>
> 'Any technology distinguishable from magic is insufficiently advanced.'
>  -Gehm's Corollary to Clarke's Third Law
>
>
>
>
>
>

        [[alternative HTML version deleted]]

______________________________________________
[hidden email] 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.
Reply | Threaded
Open this post in threaded view
|

Re: MCMCglmm multinomial model results

Bert Gunter-2
Does not the sum of probabilities (on the untransformed scale) = 1, whence
only 4 outcome categories to predict?

Cheers,
Bert



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 Sat, Mar 24, 2018 at 11:15 AM, Michelle Kline <
[hidden email]> wrote:

> Hi David,
>
> Thanks for your comment. I haven't posted the data because they are
> unpublished and include human subjects so there are issues with sharing on
> a list serv, but I thought perhaps someone had encountered a similar
> problem and would already know the answer.
>
> I will reconsider whether my University's ethics approval would allow me to
> post the data and update the question if I think it is allowable.
>
> Michelle
>
> On Fri, Mar 23, 2018, 10:13 AM David Winsemius <[hidden email]>
> wrote:
>
> >
> > > On Mar 22, 2018, at 1:31 PM, Michelle Kline <
> > [hidden email]> wrote:
> > >
> > > Hi,
> > >
> > > Thanks in advance for any help on this question. I'm running
> multinomial
> > > models using the MCMCglmm package. The models have 5 outcome variables
> > > (each with count data), and an additional two random effects built into
> > the
> > > models. The issue is that when I use the following code, the summary
> only
> > > gives me results for four of the outcome variables.
> > >
> > > Here is the code for my model:
> > >
> > > m3.random <- MCMCglmm(cbind(Opp_teacher , Dir_teacher, Enh_teacher,
> > > SocTol_teacher, Eval_teacher) ~ trait -1,
> > >               random = ~ us(trait):other + us(trait):focal,
> > >               rcov = ~ us(trait):units,
> > >               prior = list(
> > >                 R = list(fix=1, V=0.5 * (I + J), n = 4),
> > >                 G = list(
> > >                   G1 = list(V = diag(4), n = 4),
> > >                   G2 = list(V = diag(4), n = 4))),
> > >               burnin = burn,
> > >               nitt = iter,
> > >               family = "multinomial5",
> > >               data = data,
> >
> > We have no way to debug this without the data. Perhaps you should contact
> > the maintainer and in your message attach the data?
> >
> >  maintainer('MCMCglmm')
> > [1] "Jarrod Hadfield <[hidden email]>"
> >
> >
> > An equally effective approach would be to post (again with data that
> > reproduces the error)  on the R-SIG-mixed-models mailing list since
> > Hadfield is a regular contributor on that list. (To me it suggests not an
> > error since you got output but rather a warning. Generally warnings and
> > errors are properly labeled so you may not have included the full
> output.)
> >
> > --
> > David.
> > >               pr=TRUE,
> > >               pl=TRUE,
> > >               DIC = TRUE,
> > >               verbose = FALSE)
> > >
> > > And the summary of the main effects:
> > >
> > > post.mean  l-95% CI  u-95% CI eff.samp        pMCMC
> > > traitOpp_teacher    -3.828752 -4.616731 -3.067424 184.4305 5.263158e-05
> > > traitDir_teacher    -3.400481 -4.041069 -2.813063 259.1084 5.263158e-05
> > > traitEnh_teacher    -1.779129 -2.197415 -1.366496 624.9759 5.263158e-05
> > > traitSocTol_teacher -2.852684 -3.429799 -2.332909 468.7098 5.263158e-05
> > >
> > >
> > > It is not an issue of the suppressing the intercept, since I'm already
> > > doing that (see the -1 term. When I remove that term, the model
> solutions
> > > includes an intercept and only 3 additional main effects).
> > >
> > > The model does throw the following error, but after searching previous
> > > messages on this list, I've concluded that this error message doesn't
> > have
> > > to do with  my current problem. Just in case: " observations with zero
> > > weight not used for calculating dispersion"
> > >
> > > I have also posted a similar question on stackoverflow about a week
> ago,
> > > but with no response, so I thought I would try here. Link in case
> people
> > > want to gain reputation points for a
> > > response:
> > https://stackoverflow.com/questions/49309027/missing-
> term-in-mcmcglmm-multinomial-model-results-not-in-intercept-issue
> > > <
> > https://stackoverflow.com/questions/49309027/missing-
> term-in-mcmcglmm-multinomial-model-results-not-in-intercept-issue
> > >
> > >
> > > And of course I've checked various other sources including the course
> > > notes, but can't make sense of why the 5th term is dropped from the
> > model.
> > > Any help is much appreciated.
> > >
> > > Best,
> > >
> > > Michelle
> > >
> > > --
> > > Michelle A. Kline, PhD
> > >
> > > Assistant Professor
> > > Department of Psychology
> > > Simon Fraser University
> > >
> > >       [[alternative HTML version deleted]]
> > >
> > > ______________________________________________
> > > [hidden email] 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.
> >
> > David Winsemius
> > Alameda, CA, USA
> >
> > 'Any technology distinguishable from magic is insufficiently advanced.'
> >  -Gehm's Corollary to Clarke's Third Law
> >
> >
> >
> >
> >
> >
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> [hidden email] 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.
>

        [[alternative HTML version deleted]]

______________________________________________
[hidden email] 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.