I´m trying to find a glmm, which is convergent with an interaction between my categorical and nummeric variable. My experiment consists of different plot types (A, B, C & D), which were mapped in the years 0, 1, 2 & 3.
Now I want to know how different the number of species were between the respective plot types and the years. Therefore I wanted to create a glmm with the function glmmTMB. Unfortunately, my ideal model with the precipitation and the temperature (of the year when the plot was mapped) as additional correlated fixed effects didn't work. As random effect I used the different sites of the plots.
I already tried different families, different combinations of random effects with different hierarchies and adding other fixed effects like the crop on the site, the adjacent habitat and more. In addition I tried to use glmer and lmer instead, but none of my models had a convergent solution without an error meassage with the interaction of my variables.
I also tried to include the year in the plot type variable (A_0, A_1, A_2, A_3, B_0,... ). With this my model worked and I also got the significances I needed. Is this possible to do or does this generate wrong significances or correlations?
My ideal model, which is not convergent so far, is:
glmmTMB(Number_of_species ~ Year_of_Implementation * Plot_type + Precipitation * Temperature + (1|site), family = binom1, data= example)
This gives me the error message:
50 warnings or more (21: In f(par, order = order, ...): value out of range in 'lgamma') I also got often the error: Warning messages: 1: In nlminb(start = par, objective = fn, gradient = gr, control = control$optCtrl) : NA/NaN function evaluation 2: In nlminb(start = par, objective = fn, gradient = gr, control = control$optCtrl) : NA/NaN function evaluation 3: In fitTMB(TMBStruc) : Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')
My aim is to get significances for a plot showing the development for each plot type in each year.
