[R-sig-ME] Question on random effects glm interpretation

[Emmanuel Curis]

Dear Ben,

Thank you for these clarifications, and recalling in mind the problem
of independance between individuals in such contexts. We will going
back to the writing of these models to see exactly how to answer the
question using each of them...

Best regards,

On Fri, Nov 13, 2015 at 04:09:00PM -0500, Ben Bolker wrote:
« On 15-11-13 11:32 AM, Emmanuel Curis wrote:
« > Thanks for these answers.
« > 
« > After thinking a little bit, I still have a concern with the
« > survival approach.  What is in interest in the data is not really
« > the time before patients are infected (at least, as far as I
« > understood the practicionners problem), but the proportion of
« > patients that are infected at a given moment (and, at the end, does
« > this proportion change with some covariates, including period in
« > the year). I'm not clear how the survival model can give this
« > information (but I'm not familiar with survival analysis), but I
« > thought it was more oriented toward modeling time-to-event?
« 
«   Well, these are really just different representations of the same
« information.  In a canned statistical formulation (e.g. GLMM *or*
« survival analysis), it's easier to deal with the constraint that
« individuals are represented by a string of 0s followed by a string of
« 1s in the time-to-event framework.
« 
«   Given a cohort of individuals, one could construct predictions for
« the expected number infected at a given time from the results of a
« survival analysis (although might be easiest to do via stochastic
« simulation).
« 
«   The assumption of independence of the individuals also breaks down
« for infectious disease, unless the population is large enough that you
« can assume randomly sampled individuals/neglect the effect of
« infection status of some individuals on others' probability of
« becoming infected ...
« 
« > Also, going back to the original question, for my own understanding
« > of these models: is it correct to say that for a binomial GLMM to
« > apply, with patient as the (only) random effect, then for a given
« > patient the different Bernoulli variables must be independant,
« > identically distributed as in a usual logistic regression? Or is
« > this hierarchical approach too simplist?
« 
«   I think that's correct, extending your conditions slightly to say
« that the variables are conditionally iid given both the identity of
« the individual (latent variable) and any applicable
« (time/patient-specific) fixed effects.

-- 
                                Emmanuel CURIS
                                emmanuel.curis at parisdescartes.fr

Page WWW: http://emmanuel.curis.online.fr/index.html