Hello, everyone, I've been using the proc glimmix approach for quite some time now to account for random effects while exploring my response of interest, but now I have a new variable, which is zero inflated, and I want to take it into account as a response. There are 76 samples with values different than zero and 238 with 0 as value (Bd_load). I also have the binomial option of the variable (Bd_presence), which replaces the values above 0 by 1, but does not change the ratio of 0/ non zero. That's my model: PROC GLIMMIX DATA=Bddata plots=all;
CLASS lake;
MODEL observed_spp = Bd_load julian_date altitude longitude human_impact / link=log s dist=negbin DDFM=SATTERTH;
random lake;
RUN; Where observed_spp is the number of species observed in each sample, julian_date/ altitude/ longitude/ human_impact are quantitative variables without any zero and Bd_load indicates my zero inflated vector. My question is: do I need to treat this variable in a special way so it can be correctly taken into account, or it can be considered like that, since it is an explanatory variable and not my response? I tried to change the order and ran the following model PROC GLIMMIX DATA=Bddata plots=all;
CLASS lake;
MODEL Bd_load = observed_spp julian_date altitude longitude human_impact / link=log s dist=negbin DDFM=SATTERTH;
random lake;
RUN; to see if the model was correctly structured for zero inflated data, but it did not converge using the variable as the response and negative binomial as distribution. Does someone know what I could change to correctly consider my variable? Many thanks in advance, Adriana
... View more