[Feature Request]: Observation weights

[kamielkanfoudi]

[Feature Request]: Observation weights

[PhilippBach]

Dear @kamielkanfoudi

thanks for your feature request. We agree that observation weights are very important for applying DoubleML on real data sets. We'll add this feature to the list of future extensions!

Unfortunately, I'm not yet aware of a paper that goes through the weighting adjustments in DoubleML. We'd have to read a bit more on this. The complication that arises in DoubleML as compared, for example, to linear regression is that the weights show up in two separate tasks: First - and this is what I guess you probably have in mind - they have to be accounted for during the estimation of the (causal) regression parameters. Second, they might play a role for the classification and regression ML learners.... I'm not sure, if ignoring them in the 2. step might create issues.

We'll investigate this in further detail and also discuss it with some colleagues who are more familiar with that literature, but adding it to the package might take a little bit of time.

Once more, thank you and best,

Philipp