scale=deviance option in proc genmod for zero-inflated poisson and zero-inflated negative binomial

[kc]

Hello,

I get a warning in SAS saying "WARNING: Scaling the covariance is not available for zero-inflated models." when I use the scale=d option in the model statement of proc genmod.

Any idea of how to handle overdispersion if scale option isn't available for these models? Using some other procedures to run a zip or zinb models, perhaps?

Ideas welcome - thanks in advance.

[SteveDenham · Accepted solution]

I had two ideas, but neither worked. First I thought to use PSCALE and use Pearson scaling rather than deviance scaling.  Then I tried calculating the deviance scale by hand, and feeding that in as a fixed scale.  Both gave the error you reported.

I am left with trying a mixture of two negative binomials or Poisson using PROC FMM.  There is an example for handling an overdispersed Poisson as a mixture of two binomial series here .  Perhaps this could be adapted for your use.

SteveDenham