Quasi Poisson for glmm
Many thanks to you both, Thierry and Ben for your kind responses, which I find really helpful. Regards Faith
On Thu, 3 Dec 2020 8:31 pm Ben Bolker, <bbolker at gmail.com> wrote:
I agree with Thierry that the binomial is a good start.
If you do find that there is overdispersion in your binomial model,
there are (at least) three possible approaches (see
http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html#overdispersion ):
* beta-binomial model
* observation-level random effects in a binomial model
* quasi-binomial
The last one is not available in glmmTMB, but the GLMM FAQ
http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html shows you how to
get quasi-likelihood results if you want.
cheers
Ben Bolker
On 12/3/20 10:55 AM, Thierry Onkelinx via R-sig-mixed-models wrote:
Dear Faith, I'd recommend starting with a full model with binomial distribution. What you perceive as overdispersion in the response is often modelled by the covariates. Best regards, ir. Thierry Onkelinx Statisticus / Statistician Vlaamse Overheid / Government of Flanders INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE
AND
FOREST Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance thierry.onkelinx at inbo.be Havenlaan 88 bus 73, 1000 Brussel www.inbo.be
///////////////////////////////////////////////////////////////////////////////////////////
To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to
say
what the experiment died of. ~ Sir Ronald Aylmer Fisher The plural of anecdote is not data. ~ Roger Brinner The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of
data.
~ John Tukey
///////////////////////////////////////////////////////////////////////////////////////////
<https://www.inbo.be> Op do 3 dec. 2020 om 16:17 schreef Ebhodaghe Faith <
ebhodaghefaith at gmail.com
:
Dear Thierry, The proportions are on number of individuals infected by a parasite divided by total number of individuals examined. Thanks Faith On Thu, 3 Dec 2020, 4:48 p.m. Thierry Onkelinx, <
thierry.onkelinx at inbo.be>
wrote:
Dear Faith, I missed to see you have a proportion response. The negative binomial
is
a (better) alternative for the quasi Poisson. But they assume count
data.
What kind of proportions do you have? Is it based on a number of successes for a number of trials (binomial, beta binomial)? Or a
continuous
value between 0 and 1 (beta)? Best regards, ir. Thierry Onkelinx Statisticus / Statistician Vlaamse Overheid / Government of Flanders INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND FOREST Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance thierry.onkelinx at inbo.be Havenlaan 88 bus 73, 1000 Brussel www.inbo.be
///////////////////////////////////////////////////////////////////////////////////////////
To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able
to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher The plural of anecdote is not data. ~ Roger Brinner The combination of some data and an aching desire for an answer does
not
ensure that a reasonable answer can be extracted from a given body of
data.
~ John Tukey
///////////////////////////////////////////////////////////////////////////////////////////
<https://www.inbo.be> Op do 3 dec. 2020 om 13:14 schreef Ebhodaghe Faith < ebhodaghefaith at gmail.com>:
Thanks, Thierry. But could you please refer me to an article preferably in the
biological
sciences where a negative binomial distribution was used to model an over-dispersed multilevel proportion response variable? Thanks for your kind assistance. Regards Faith On Thu, 3 Dec 2020, 1:32 p.m. Thierry Onkelinx, < thierry.onkelinx at inbo.be> wrote:
Dear Faith, You can use a negative binomial distribution. Best regards, ir. Thierry Onkelinx Statisticus / Statistician Vlaamse Overheid / Government of Flanders INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR
NATURE
AND FOREST Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance thierry.onkelinx at inbo.be Havenlaan 88 bus 73, 1000 Brussel www.inbo.be
///////////////////////////////////////////////////////////////////////////////////////////
To call in the statistician after the experiment is done may be no
more
than asking him to perform a post-mortem examination: he may be able
to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher The plural of anecdote is not data. ~ Roger Brinner The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given
body of
data. ~ John Tukey
///////////////////////////////////////////////////////////////////////////////////////////
<https://www.inbo.be> Op do 3 dec. 2020 om 11:24 schreef Ebhodaghe Faith < ebhodaghefaith at gmail.com>:
Hi All. I have a dataset for wish I intend to model an over-dispersed proportion response variable with hierarchical structure. I tried using the
Quasi
Poisson family, but available packages including glmmTMB do not
allow
this.
What do you advice?
Thanks in advance for your kind response.
Faith Ebhodaghe
Nairobi, Kenya
[[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
[[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models