Zero-inflated and correlated count data

[Mehdi_R]

Hello everyone,

I have a huge longitudinal data, with 45000 individuals and about 5 visits. The outcome is count with so many zeros. The only way I know to take into account the excess zeros and the correlation, is to use random effect models in either PROC MCMC or PROC NLMIXED. However, both of these procedures will take forever because number of patients is huge number. I was wondering if anyone has seen a macro which uses GEE method and simultaneously takes excess zeros into account. I think such a method will be much quicker to analyse this dataset...

Thank you very much in advance.

[RyanSimmons]

Have you done diagnostics on the degree of zero-inflation to verify whether or not you need zero-inflated model to begin with? For example, in many cases a negative binomial model (which can be easily fit in GENMOD with a repeated statement) will fit zero-inflated data quite well (in fact, anecdotally speaking, in my experience a ZIP or ZINB model rarely offers any practical advantages over a negative binomial model in the case where you have zero-inflation and overdispersion).

[plf515]

Have you  tried PROC COUNTREG? 

[Mehdi_R]

I haven't tried it, but I think this procedure does not adjust for correlation. 

[Mehdi_R]

Thank you for the reply.

Actually it seems you're right. I compared a zero-inflated random effect model with just a random effect. It seems fit becomes better by taking into account the excess zero, but the estimation of parameters in the main models are quiet similar.