I don't know what the
convergence code 3 from bobyqa: bobyqa -- a trust region step failed to reduce qData: Data3 error message means. Should I be concerned about this error? See below for example:
I using backwards elimination to get the minimally significant lmer model.
My model comes up with an error that says:
boundary (singular) fit: see ?isSingular
I have done some research and I believe this error to be fine as individual child should be included as a random effect in the lmer model.
I compare two models
cca_p <= lmer ( A ~ B + C + D + E + (1|child), data = Data) # remove E (as least significant main effect from model) cca_p1 <= lmer ( A ~ B + C + D + (1|child), data = Data) # compare the two models to determine if E is significantly important to the model. anova (cca_p, cca_p1, test = "Chisq")
I then receive this output but do not understand what the convergence code 3 from bobyqa error message means:
Googling suggests that re-leveling my data alters this. Sure enough if I relevel some of my data the error will go away but reappear later on in the modelling.
Secondly the error will vanish depending on the order I remove the explanatory variables but I should really remove the least significant ones first rather than remove them in an order that doesn't give me the error message.
I then tried the following to determine which optimizer I should be using if not bobyqa:
and also allfit
I am assuming that it is something to do with the singular fit but I'm not sure. Should I be concerned about this or is it okay? I can backwards eliminate with this error and get a final minimal model from it if I ignore this but I am not sure if I should be ignoring it.
Any advice greatly appreciated.