Josef Perktold
Josef Perktold
I don't find a docstring that includes description and defintion summary info. current point: I was not sure whether kurtosis is fisher or pearson, i.e. 0 or 3 in normal...
getting started with a wishlist for outlier robust multivariate location and scatter estimators #3220 size (overall scaling) - MCD in scikit-learn, not good with high contamination and large k_vars -...
(I found several articles that I printed at around 2016) - robust, RLM is only for univariate endog - MultivariateLS has been added in #8919 - a first set of...
see #9064 #3528 The main references that I used for this are in first comment of #9064. currently mainly tools to convert random variable and cov between 3 representations (to...
For implementation it looks like we could reuse sandwich covariance. For RLM: H1 is the analog of OLS "nonrobust" covariance I haven't figured out if H3 is HC, it has...
(another PR to consolidate my robust regression code) I added it for now to miscmodels because the models subclass GenericLikelihoodModel. this includes a Huber-Dutter model class, see related #3258 (i.e....
Reading Li and Valliant https://github.com/statsmodels/statsmodels/issues/9067#issuecomment-1823299955 They have a case where the model implied cov_params for WLS params is cov(beta) = sigma^2 A^{-1} X' W^2 X A^{-1} where A = X'...
(starting a separate issue for this) #2805 implements frequency weights for GLM, but it doesn't work for robust covariances one issue is restrictive assumption for nobs, k_params, see #2827 but...
related: margins with weights #2696 weighted sum/total #2851 how do we predict a weighted average over the population? similar to POM (potential outcome mean) triggered by a WLS survey weights...
**edit** see #4732 for updated information and implementation (a random find) Hosmer,Lemeshow, and May in chapter 8 refer to a influence function definition of Cooks Distance and use score residuals....