What are the methodological concerns with regards to using Poisson distributed count variables as an independent variable in the regression analysis?