So, I'm attempting to run a negative binomial in Generalized Linear Models. The outcome variable is homicide rate per 100,000. The analysis must be conducted with this outcome variable (rather than the raw count) because that is the convention (and I'd rather not deal with critiques of using something else). However, I get the error below.
"One or more cases were found with dependent variable data values that are non-integer or less than zero. These values are invalid for the negative binomial probability distribution, and the cases are not used in the analysis.
There are no valid cases for the log link function.
Execution of this command stops."
My thought is of course it is a non-integer it is homicide rate per 100,000 people. I have to admit that I'm perplexed on how to resolve this issue so that I can run the analysis. Am I missing something here. The data is over-dispersed with mean = 3.55 and variance = 9.17. Any thoughts on how to resolve this? Is it acceptable to round the values to integers? Is there a better way? Thanks!