Hi Team,
I’ve been working on two PRs recently that would add new features:
gh-22185 proposes the addition of an equal_var option to scipy.stats.tukey_hsd. When equal_var=True (default), the test assumes the variances of the populations underlying the samples are equal; this is the Tukey HSD or Tukey-Kramer test, depending on whether the groups have the same or different numbers of observations. When equal_var=False, the test does not make this assumption, and the standard error and degrees of freedom are adjusted accordingly; this is the Games-Howell test. In this context, equal_var is analogous to the existing equal_var parameter of ttest_ind.
gh-22644 proposes the addition of a method='winsor' (and possibly variations like method='winsor-round' or method=('winsor', np.round)) to the new scipy.stats.quantile. This notion of a quantile is particularly useful when winsorizing or trimming data. To replicate all the existing behavior of scipy.stats.mstats.mquantiles, we are also considering a “Blom” method that would allow the user to specify alpha/beta parameters as originally proposed by Gunnar Blom in “Statistical Estimates and Transformed Beta Variables” (1958).
Please join the discussions in the linked PRs!
Thanks!
Matt