I can't reproduce by hand Mathematica's PearsonChiSquareTest[dataset1,dataset2] on two datasets of discrete rand variables, with lengths N1 and N2, N1 much less than N2. (I just want to understand what it's doing).

I assume Mathematica determines B bins (using B=Ceiling[2*N1^0.4] - note the N1.), applies this to dataset2, and uses these to bin the datasets. The chisqr (as per NumC recipes and elsewhere) is not the sum over $i$ of:


where $u$ and $v$ are the bin frequencies for the 2 datasets, and K=Sqrt[N2/N1]. Using this I get chisqr's scattered near, but below what M finds. Btw, bin delimiters are from Quantile[dataset2, Range[0.,1,1/B]].

Unfortunately, the HypothesisTestData property doesn't contain enough info to really see what's going on. Maybe Mathematica chooses bins differently? or uses a correction to the formula above?

Thanks for any insights.



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.