Hi, I an trying to analyze an experiment with count data that I am having great difficulty finding a model that will converge that has all the appropriate random variables within the model. The data structure is a summary count of events in an animal during a day. The summary counts are generated from a continuous stream recording of the data from multiple animals over 10 different days, which are unequally spaced. Day=(-1,0,1,7,14,23,24,25,26) The amount of the recording that is readable varies with each recording, as animal movement etc makes the recording unreadable at times. The data is originally reported by the instrument software on an hourly basis, but I have summarized it as a count on a daily basis for a couple reasons... the primary reason being that the # of events is very low and the overwhelming majority of hourly readings are 0. The outcome of interest is the # of events between trt groups (n=3). There are 2 blocks to the experiment. Animals are group housed and individually managed (EU=Animal). I've attached code below, but it doesn't work. Variables that cause problems are pen and block if the day component is the in the model. We want to know the effect of trt over time so day is crucial in the analysis. I can get trt|day into the model if block becomes fixed. As we would like the inference to apply to 'all blocks' I would like this to be random. Bottom line: The overdispersion in this dataset is will not allow me to fit a poisson or nb model (all tiand my understanding of GLIMMIX is that it won't do Zero inflated models. That leaves me with PROC NLMIXED, which I have no experience on how to set up the parameters for. Any advice for a new approach would be appreciated. lday= log(days) day is in decimal for. i.e. 97% or .97 of the day's recording was readable. I think this is the right form to put this in but I'm not sure. proc glimmix data=data plot=(residualpanel studentpanel); class trt animal block pen day; model Event=trt|day /solution dist=nb ddfm=kr offset=ldays; random block; random _residual_/subject=animal; random int /subject=pen; lsmeans trt|daymod; run;
... View more