Doubly Robust Estimator

[nparent1]

PR to add support for the doubly robust estimator, with an example notebook. This PR addresses issue #407.

References:

Michele Jonsson Funk, Daniel Westreich, Chris Wiesen, Til Stürmer, M. Alan Brookhart, Marie Davidian, Doubly Robust Estimation of Causa Effects, American Journal of Epidemiology, Volume 173, Issue 1 April 2011, Pages 761-767, https://doi.org/10.1093/aje/kwq439

[nparent1]

tagging @amit-sharma or @bloebp for review!

[nparent1]

Note the 'regression_estimator' argument accepts a RegressionEstimator, whereas the 'propensity_score_model' argument accepts a propensity score model directly.

This is because it seemed more natural to let the user pass in a propensity score model directly, rather than a PropensityScoreEstimator, given most of the arguments, and all of the subclasses of the PropensityScoreEstimator class would be unneeded here (there would be no reason to pass in a PropensityScoreMatchingEstimator vs a PropensityScoreWeightingEstimator; the underlying propensity score model is working the same in all subclasses). Thus, it seemed more straightforward to just have the user pass in directly here the arguments that would be sufficient to handle the propensity score stage.

On the other hand, it seemed fair to use an argument of type RegressionEstimator for the regression stage, since a user may want to use a linear model, a glm, or some other future sub-class of the RegressionEstimator. In other words, the subclasses of RegressionEstimator may actually have their own constructors, that we wouldn't want to generalize here.

[nparent1]

Hello! Gentle bump that this PR is ready for review. Let me know if there is someone else I should tag!