Variable with high number of categories I would like some help with my analysis. I have an outcome variable that represents average number of errors somebody experiences per hour. It has about 26% zeros. 85% of values are smaller than 1. There are some outliers going up to >700. 
My explanatory variables are only categorical variables. One has 3 categories, one has 2 categories, one has six categories and one has 91 categories (area within a country). 
My question is whether anyone has some tips analyzing this dataset. I want to know the effects of the variables and potential interaction effects on the outcome. 
I would like to know as well if there are some locations that experience significantly high numbers of errors. 
Can anybody help me going in the right direction? I am just starting in statistics. 
 A: There are at least two issues here:
1) The distribution of the dependent variable and
2) The number of categories in one of the independent variables
Regarding 1) Since the variable is the average number of errors, it is essentially continuous, albeit very skewed. OLS regression makes no assumptions about the shape of the dependent variable, only about the residuals. But it is likely that they will be non-normal and you may need to transform the DV.
Regarding 2) There is no special reason you cannot have a variable with so many categories, provided that you have enough data. You will want to have enough cases in every region. You may need to combine regions.
A: I would suggest the following:
1) Inspect the assumption of a Poisson variable
2) Inspect overdispersion
3) Inspect excessive zeros
If 1) is not rejected, then proceed with Poisson regression in a generalized linear model.
If 1) is rejected then
If 2) is true and 3) is not, then proceed with Negative Binomial model.
If 2) is false and 3) is true, then proceed with Zero-Inflated Poisson model.
If both 2) and 3) are true, then proceed with Zero-Inflated Negative Binomial model.
