As the title:

I have a discrete distribution with a small number of categories. I want to test if it is compatible with a parameterized distribution. If I were to do a chi-squared or G-test, I should account for the fact that I'm estimating the parameter when computing the p-value from the test statistic.

I don't want to use these tests because I've been warned that they are not appropriate when categories have fewer than around 5 counts observed or expected, and my counts are nowhere near uniform, so categories are routinely small. But because I do have few categories, the multinomial test is completely feasible in terms of computation time. However, I'm comparing against a family of parameterized distributions, and all the presentations have been against a pre-specified distribution.

Is the multinomial test still appropriate here? Is correcting for estimating the parameter necessary or even possible?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.