Reduced chisquare can assume negative value without warning

[gotten]

Related to issue #65 . Reported by Richelle van Capelleveen.

No warning is given when the amount of datapoints is so low the degrees of freedom can assume a negative value. This triggers a larger discussion on the usefulness of the reduced chisquare in this regime. Is there a statistical metric that can handle this regime?

[wbalmer]

We could throw a userwarning in that case. More generally, we could have a toggle between reporting/using the chisq value itself (or another least-squares/goodness of fit metric) and the reduced chisq? I agree it should be a priority to have a good metric for the common use case of 2-3 points in the stationary case

[gotten]

I agree with the warning and having a toggle. At least to solve the immediate issue.

Need to think more about a general solution.

[tomasstolker]

Perhaps a Delta BIC/AIC can be calculated, comparing the maximum likelihood from the (non-)stationary case with a MLE from a fixed position or linear movement, in case of negligible and small orbital motion, respectively.