[R] zero values in LHS and RHS

[Gregor Gorjanc]

zubin <binabina <at> bellsouth.net> writes:

> 
> Hello!
> 
> I have a data set where i need to predict hotel stays in a 3 month 
> period by customer.  I have stays for each customer in the 3 month 
> period and the previous 3 month period + other variables.  these stays 
> contain an integer that ranges from 0 to 10.  classic poisson distribution.
> 
> my question is around zeros.  usually we could take logs on the stays 
> and run an OLS model i think an enhancement would be to fit a poisson 
> GLM model.  i am confused about handling of zeros.  i will have zero 
> values on the RHS and potentially on the LHS.  It seems like the poisson 
> GLM model can handle zeros on the LHS and RHS.   i guess my question is 
> fitting a poisson GLM, are zeros allowed?  If so, how is this estimated 
> with a log link function?
> 

I am not sure about the structure of LHS and RHS matrices, but yes, 
Poisson model can handle zeroes. If it turns out that there are to many 
zeroes as expected, you might try ZIP model.

Log link is the default for Poisson and it links the expected value of 
Poisson distribution and not the values itself.

Gregor