I have counts of infections reported in all states in the United States, and I want to test whether the rate at which infections are reported in each state is different. What is the appropriate test for this?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
You can formulate this as an generalized linear model with log link function, also known as poisson regression. The response variable will be the observed counts of infections, then you should include as an offset the logarithm of exposure. Exposure could be population of state, or some relevant subpopulation, or even persondays. You could use a poisson model, or, if overdispersion is an issue (which it probably is), negative binomial regression. Then the residual deviance from this model can be used for hypothesis testing.
For some more details on poisson regression see Goodness of fit and which model to choose linear regression or Poisson
answered Aug 8, 2017 at 19:09