Custom degrees of freedom argument for the PowerDivergenceTest

[BERENZ]

Is it possible to add a custom parameter for the degrees of freedom in the PowerDivergenceTest function which will be an adjustment to the df for the p-value? This may be useful for the goodness of fit tests ($G^2, \chi^2$) as the current implementation does not account for the number of parameres of the reference distribution. I think that implementation proposed in scipy.stats.power_divergence seems be the right way to do so (ddof parameter). Now, the user needs to remember to correct p-value after the test.

[nalimilan]

Makes sense. Could you provide other references about this adjustment? Would you be willing to make a pull request?

[BERENZ]

Sure, I will make a pull request to the end of this week.

[kunzaatko]

Hello @BERENZ. May I ask about the state of this issue? It looks like you already have a commit in your fork for this. It there something else missing that needs to be done before the upstreaming? I can help with something if necessary.

[sprague252]

I am also interested in this functionality. I imagine it would work like the ddof parameter in scipy.stats.chisquare. See the documentation at https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.chisquare.html. In that function the ddof parameter gives the decrease in degrees of freedom from the default value. Of course, another possibility would be to add a dof parameter for the user to supply the number of degrees of freedom directly.

What can I do to help make this happen?