modified Poisson regression and the prevalence of the outcome is below 10%

[Sonia Zam]

I am trying to estimate which demographic variables are associated with smoking and alcohol drinking. The prevalence of smoking and alcohol drinking in the study sample (cross sectional study) are 15% and 2%, respectively. I am using modified Poisson regression to estimate the prevalence ratio. Based on the literature modified Poisson regression is recommended if the prevalence of the outcome >10%. Is this model still appropriate for to estimate the prevalence ratio to identify which demographic variables are associated with alcohol drinking, where the prevalence is 2%?

Many thanks for your help.

[Sonia Zam]

any advice, please

[Richard Williams]

Sonia, I don't know if it will help, but I would cite your source for the modified Poisson Method as well as the Stata code you are using.

Allison discusses rare events in logistic regression. He says a key concern is how large your sample is. Perhaps the same is true with your modified Poisson.

http://statisticalhorizons.com/logis...or-rare-events