# Rule of thumb - number of predictors - Poisson regression rates

I am interested in estimating a Poisson regression for mortality rates, with number of deaths as the dependent variable and log(population size) as the offset. I have 50 observations (states).

I am aware of rules of thumbs in various regression models to figure out the maximum number of predictors. For example: Cox model: at least 5 events per predictor, logistic regression: at least 5 observations where Y=1 per predictor etc...

Is there a similar rule of thumb for Poisson models?

• ¿Why do you need "a rule of thumb"? Those are mainly, anyway, just meant as some yardstick people can use until they get experience! Just do some simulations in R and find out four yourself ... Anyway, in Poisson regression the dependent variable $Y$ is a count, so a numeric variable, so as a start you could use the same rules you would apply for ordinary linear regression... The main difference between those two models is not in the distributional assumption (Normal versus Poisson), but in the structure of the expectation --- linear versus multiplicative. – kjetil b halvorsen Sep 25 '12 at 22:29