# Modeling fixed- effects without interaction terms

I'm trying to model a linear mixed effects model but whenever I tried to model interaction terms my model gives error about convergence. Although I tried to increase the number of iterations as possible solution I could not achieve to model interactions either. What are your suggestions about this?

With only main effects, my code works fine but ı need to interpret the interaction terms as well. Also I 'm not quite sure is it appropriate to model the fixed effects with only main effects, without interaction?

Any help would be greatly appreciated.

• Could you post some code and/or the output? – Mark White Apr 26 '17 at 21:16
• How many samples and independent variables do you have? Did you initially include all first order interactions? – Michael Chernick May 20 '17 at 1:48
• I have 3 independent variables, age, gender and unit (unit has 6 categories). I only include significant first order interactions to the model. – eee May 25 '17 at 16:39

Sorry for the late response. Here are the R codes that I've used:

# LME FORMULATION

lme.model2=lme(trsf~time*age+gender+unit, random = ~time+I(time^2)|id ,
control='ctrl',na.action = na.omit,data = crp)


# JOINT MODELING FORMULATION

ctrl <- list(iter.EM=200,tol3=1e-09,numeriDeriv="cd",eps.Hes=1e-04)
jm.model2.uns<- jointModel(lme.model2,surv, timeVar = "time",control='ctrl')
Error in optim(thetas, opt.survWB, gr.survWB, method = "BFGS", control = list(maxit = if (it <  :
non-finite value supplied by optim


Whenever I include the interaction term, joint model function gives the error but if I use only the main effects I don't get this kind of error.