`PyFixest` Support

[s3alfisc]

Hi all,

The package looks really great!

We're building pyfixest , which is just like R's fixest, but in Python, and I'd like to add support for sensitivity analyses.

How difficult would it be to add support for other classes than statsmodels OLS? Would it be possible for me to set up basic support for pyfixest? And if yes, how would you a PR to look like?

Best, Alex

[s3alfisc]

It seems to be that we could get very far by simply adjusting the model_helper function?

[carloscinelli]

The model helper would likely be sufficient. As of now you would need to pass classical standard errors, support for robust standard errors is coming soon (in practice, for sensitivity analysis, the choice of standard errors is a second order concern as the bias dominates).

[s3alfisc]

Great! I'll then prepare a PR. Thanks @carloscinelli!

[carloscinelli]

Great! I'll then prepare a PR. Thanks @carloscinelli!

On further thought, there is also the bounds using observed covariates, which needs to fit a new regression model for the treatment, see here:
https://github.com/s3alfisc/PySensemakr/blob/2bf5cfd0f17936b9dfe5090f03ede53387fe66a3/sensemakr/sensitivity_bounds.py#L147