Update wls types tests

[s3alfisc]

What this PR does:

1. It slightly adjusts the default small sample corrections for heteroskedastic errors. It implements the following logic:

            adj1 = _N / (_N - 1)
            adj2 = (_N - 1) / (_N - _k)
            adj3 = _N / (_N - _k)

            adj1, adj2, adj3 = (np.array([x]) for x in [adj1, adj2, adj3])

            if _ssc_dict["adj"] and _ssc_dict["cluster_adj"]:
                self._ssc = adj3
            elif _ssc_dict["adj"] and not _ssc_dict["cluster_adj"]:
                self._ssc = adj2
            elif not _ssc_dict["adj"] and _ssc_dict["cluster_adj"]:
                self._ssc = adj1
            elif not _ssc_dict["adj"] and not _ssc_dict["cluster_adj"]:
                self._ssc = np.array([1.0])

which is exactly what r-fixest implements. See this issue in the fixest repo for details. Unit tests are updated accordingly.

2. Adds a vignette that compares results from r-fixest with pyfixest. See here.

3. Re-organizes some code logic in the Fepois method: self._X and self._Y are not multiplied with the IWLS weights. Now the logic is identical between the Feols, Feiv, and Fepois classes. Updates the CRV1 vcov method accordingly.

4. Slightly refactors the vcov_hetero() method, for easier readability.

Too many only slightly related changes in this PR imho. Sorry about that!

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB