I am trying to run a model selection anlysis (with MuMIn and AICc) based on a set of mixed effects models with a binomial distribution family. My data looks like this (sorry for not posting it more elegantly)
And the saturated model (out of 38 combinations) is this:
model1 <-glmer( occ~Habitat*NDVI_centre+Dist_Paths+Dist_Power_Lines+Dist_relocation + (1 | Individual) , family=binomial , data=g.hab, control=glmerControl(optimizer="bobyqa"),nAGQ=10)
Some of the models run smoothly but about half of the models give me this error:
Error in pwrssUpdate(pp, resp, tol = tolPwrss, GQmat = GHrule(0L), compDev = compDev, : (maxstephalfit) PIRLS step-halvings failed to reduce deviance in pwrssUpdate
I wasn't able to find a solution for this and will appreciate any guidance.
hopefully useful info :
- all variable were centered
- the same parameters (indp var) were used in a model with a different binomial response variable and worked fine.
- the dataset is large (44000 observations)
In some more details:
The full dataset can be found here
The following models showed no error (just a couple examples):
`model38<-glmer( occ~Habitat + (1 | Individual) , family=binomial , data=g.hab, control=glmerControl(optimizer="bobyqa"),nAGQ=10)
model33 <-glmer( occ~Habitat+Dist_Paths + (1 | Individual) , family=binomial , data=g.hab, control=glmerControl(optimizer="bobyqa"),nAGQ=10)
model8 <-glmer( occ~Habitat+Dist_Paths+Dist_Power_Lines+NDVI_centre + (1 | Individual) , family=binomial , data=g.hab, control=glmerControl(optimizer="bobyqa"),nAGQ=10)
` But these showed the above error :
model3 <-glmer( occ~Dist_Paths+Dist_Power_Lines+Dist_relocation+NDVI_centre + (1 | Individual) , family=binomial , data=g.hab, control=glmerControl(optimizer="bobyqa"),nAGQ=10)
model1 <-glmer( occ~Habitat*NDVI_centre+Dist_Paths+Dist_Power_Lines+Dist_relocation + (1 | Individual) , family=binomial , data=g.hab, control=glmerControl(optimizer="bobyqa"),nAGQ=10)
`
Thank you very much