Time-dependent Negative binomial regression

Dear Amir,

Have a look at the lme4, glmmTMB or INLA packages. Note that if you need on
the fly transformations in the model you need to code them as
I(log(person.time)) instead of log(person.time). Personally, I prefer to
create a new variable in the data.frame and use that new variable in the
model.

Another thing is that you shouldn't use gender and baseline.age as random
effects. Either don't use them (as their effect is handled by the id random
effect) or add them as fixed effects.

library(lme4)
glmer.nb(event ~ offset(log_time) + treatment + gender + baseline.age +
(1|id), data = df)

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 8 jul. 2021 om 08:59 schreef <
Amirhossein.AmirhosseinTalebi at radboudumc.nl>:

Dear responders,


Recently I have processed and cleaned a data for the aim of application of
a negative binomial regression.

First, I tried to use the function glm.nb of package MASS in R and I had a
problem with ensuring that the model will realize the data are for one
unique participant (possible correlations in a group of observations).

Then, I realized that I can use glmmPQL of package MASS or glmer of
package lme4 and use the family negative binomial in it's family link.

The question is I would like to know in which part of the model I can
embed the offset (logarithm of the number of days of treatment) also how
should I insert the time-constant observations for an id (such as gender
and baseline age in the df)?

My latest attempt was:

(glmmPQL (event ~ treatment + offset (log(person.time)) ,
random= list (id=~1, gender=~1, baseline.age=~1),
family= negative.binomial (theta=1.75), data=df ))


which faced with a memory-related error (probably because of the wrong
code). data example:

df<-data.frame(id=rep(1:3,each=4),treatment=sample(c(0,1),12,replace = T),
event=sample(c(0,1),12,replace = T),
person.time=sample(c(15,31,30),12,replace = T),
age=rep(c(65,58,74),each=4),gender=rep(c("m","f","m"),each=4))


Thank you for your time and considerations,

Amir

De informatie in dit bericht is uitsluitend bestemd voor de geadresseerde.
Aan dit bericht en de bijlagen kunnen geen rechten worden ontleend. Heeft u
deze e-mail onbedoeld ontvangen? Dan verzoeken wij u het te vernietigen en
de afzender te informeren. Openbaar maken, kopi?ren en verspreiden van deze
e-mail of informatie uit deze e-mail is alleen toegestaan met voorafgaande
schriftelijke toestemming van de afzender. Het Radboudumc staat
geregistreerd bij de Kamer van Koophandel in het handelsregister onder
nummer 80262783.

The content of this message is intended solely for the addressee. No
rights can be derived from this message or its attachments. If you are not
the intended recipient, we kindly request you to delete the message and
inform the sender. It is strictly prohibited to disclose, copy or
distribute this email or the information inside it, without a written
consent from the sender. Radboud university medical center is registered
with the Dutch Chamber of Commerce trade register with number 80262783.

         [[alternative HTML version deleted]]

	[[alternative HTML version deleted]]