Mixed effect model, Hurdle function
This isn't really a "mixed effect model" in the standard terminology (there's no random effect). Nevertheless ...
On 16-11-17 11:29 PM, Pin chanratana wrote:
Hi everyone here, My name's Ratana, and I'm a Msc. student studying conservation ecology. I'm new to the using of mixed effects model. I try to fit the mixed effect models include an offset term (which is trapnight) of my camera-trap data by using hurdle function. The following are the model that I fit. m1 <- hurdle(GI ~ depth+Ele+dRoad+offset(log(TN))|depth+Ele+offset(log(TN)), data=ndata2, dist="poisson", zero.dist="poisson") m2 <- hurdle(GI ~ depth+Ele+dRoad+offset(TN2)|depth+Ele+offset(TN2), data=ndata2, dist="poisson", zero.dist="poisson")
I'm unfamiliar with models that use the censored Poisson for their hurdle model (binomial is more standard in my experience), but whatever.
GI: Giant ibis depth: depth of waterholes Ele: Elevation dRoad: distance to Road TN: Trap-night TN2: standardize or scale of Trap-night But, I could not fit model m1 and there are error message: Error in glm.fit(X, Y, family = poisson(), weights = weights, offset = offsetx) : NA/NaN/Inf in 'y' Model m2 is work fine and the result look reasonable.
Your first model looks more correct/standard; log(exposure) is the standard offset in a Poisson count model. (Not as clear what to use as an offset for the hurdle; I would actually say that a log-exposure offset with a complementary log-log link would actually make the most sense for a binomial, but I haven't thought about how that would go together with a censored Poisson ...) Is it possible that you have some observations in your data with TN=0? That would cause the first model to fail. (It wouldn't really make sense, but I've seen observational data like this where the person taking the data rounded down to zero trap-nights.) Could you please send follow-up questions, if any, to r-sig-ecology at r-project.org ?
I would like to ask if anyone know, is the way I do with model 2 is correct by standardize the Trap-night that's use for offset in the fitting model? If it's ok to do so, is there any reference (that I can cite) regarding to this kind of process ? Regards, Ratana [[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models