I am carrying out analysis of a large seedlings dynamics dataset (20258 seedlings from 7467 plots) which I am unable to share for privacy reasons, as it is not my dataset. In this analysis I am trying to determine how several explanatory variables (stem length, functional group, conspecific and heterospecific density, and the interactions between functional groups and density) affect the probability of survival for a given seedling - using the glmer() function in lme4. Before using the model, I scaled all continuous variables using the scale() function and made sure all non-continuous variables were listed as factors, and played around with optimizer settings and iteration number to get rid of some standard optimizer warnings.
The model is structured like this:
themod <- glmer(survival ~ stemlength + group + conspecific density + heterospecific density + group:conspecifics + group:heterospecifics +group:stemlength + (1|plot) + (1 + conspecifics + heterospecifics|species), family=binomial(link = "logit"), data = dataset, control=glmerControl(optimizer="bobyqa", optCtrl=list(maxfun=2e5)))
I am trying out ways of selecting these models beyond AIC values, dropterm analysis or R squared values. One residual vs fitted diagnostic plot i have found used for glmms is found at this link:
The author suggests that the plotted line should follow the line in a model that fits the data well.
I have used this to generate some plots for the model, with and without residual outliers (see figures, click link below):
My two questions are 1), are these plots useful these plots are with binomial data, such as seedling survival? and 2), if so, do my plots indicate that my model is of a sufficient fit to my data?
Thanks a lot in advance, and sorry if im being dense or missing something easy to google. Cheers!