[R-sig-ME] ZIPoisson MCMCglmm: potential non-independence of random effect repeat number and response variable

Jarrod Hadfield j.hadfield at ed.ac.uk
Thu Jan 19 10:43:48 CET 2017


Mixed models deal with unbalanced data so having a variable number of 
observations per animal is not a problem. However, if the number of 
observations is correlated with the outcome you have to remember that 
the model is 'correcting' for this. For example, imagine 20 individuals 
have 2 observations and an expected outcome of 1, and 20 individuals 
have 6 observations with an expected outcome of 2, the intercept will be 
(on average) 1.5 not 1.75 (the average across observations).



On 18/01/2017 12:14, Rebecca Hooper wrote:
> Dear List,
> I am looking at the annual reproductive success of individuals relative to
> their dispersal behaviour using ZIPoisson MCMCglmms.
> The data is structured in such a way that individuals have one row per year
> lived (so if they live to 5yo they have 5 rows).
> Annual reproductive success (ARS) is the response variable and is zero
> inflated (85% 0s).
> Dispersal score per year is the predictor variable, and Individual ID is
> the random effect.
> ARS may not be independent of longevity, and thus the number of rows
> individuals have in the data may not be independent of the response
> variable (e.g. individuals that live longer, with more rows in the data,
> may have decreased reproductive success per year relative to those that
> live less long).
> I don't know how this non-independence between the response variable and
> the number of repeats of the random effect might effect the model results,
> and would be very grateful for some insight into whether this is likely to
> cause problems.
> Many thanks,
> Beki
> 	[[alternative HTML version deleted]]
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.

More information about the R-sig-mixed-models mailing list