Add offset option for fepois

[bradhackinen]

This code adds an offset option to fepois regressions so that the function can be used to model differences in outcome rates.

• In terms of the UI, I followed the functionality of the weights parameter in feols regressions. So the offset argument takes a string that must refer to a column name in the data. If no argument is supplied, the _offset vector is set to 0. This required modifying model_matrix_fixest to take an optional offset argument.
• Changes to the estimation code are very simple. The offset is added to Z before demeaning and WLS, and then re-added to eta before computing the deviance.

I tested the code manually on the included poisson test data, a separation example (01.csv), and data from a current research project. The output the matches the R implementation of fixest (at least to the decimal places visible in the output) in with and without FE.

[codecov]

Codecov Report

Attention: Patch coverage is 80.00000% with 5 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
pyfixest/estimation/model_matrix_fixest_.py	50.00%	3 Missing ⚠️
pyfixest/estimation/estimation.py	66.66%	1 Missing ⚠️
pyfixest/estimation/feols_.py	88.88%	1 Missing ⚠️

Flag	Coverage Δ
core-tests	80.82% <80.00%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
pyfixest/estimation/FixestMulti_.py	82.58% <100.00%> (+0.15%)	⬆️
pyfixest/estimation/fepois_.py	87.61% <100.00%> (+0.11%)	⬆️
pyfixest/estimation/estimation.py	93.65% <66.66%> (-0.71%)	⬇️
pyfixest/estimation/feols_.py	83.85% <88.88%> (+0.04%)	⬆️
pyfixest/estimation/model_matrix_fixest_.py	84.04% <50.00%> (-1.29%)	⬇️

[s3alfisc]

pre-commit.ci autofix

[s3alfisc]

Hi @bradhackinen , looks like the unit tests I added fail, but almost pass, which is of course the worst that could happen as it isn't clear if the implementation is slightly off or the tolerance just too low 😅 . I will try to debug this in the next days, but might not find time this weekend as I am traveling.

[bradhackinen]

No worries @s3alfisc. Thanks for writing the tests. Let me know if there is something I can do to help resolve the problems. I can get rpy2 set up on my system next week if necessary.