[DOC] Clarifying question about causalforest vs OrthoForest

Hi,

I have a doc question from the following page: https://econml.azurewebsites.net/spec/estimation/forest.html#causalforest-aka-forest-double-machine-learning

In the docs, I am also a bit confused on the following statement:

The Causal Forest has two main differences from the OrthoForest: first the nuisance estimates q and f are fitted based 
on a global objective and not locally for every target point. So typically they will not be minimizing some form of local 
mean squared error. Second the similarity metric that was potentially used to fit these estimates (e.g. if a 
RandomForest was used) is not coupled with the similarity metric used in the final effect estimation. This difference 
can potentially lead to an improvement in the estimation error of the OrthoForest as opposed to the Causal Forest. 
However, it does add significant computation cost, as a nuisance function needs to be estimated locally for each target 
prediction.

The first difference is worded as if tho the differences highlight what changes from OrthoForest -> CausalForest.

The second difference is worded almost as if the changes from CausalForest -> OrthoForest.

Second the similarity metric that was potentially used to fit these estimates (e.g. if a RandomForest was used) is not 
coupled with the similarity metric used in the final effect estimation. This difference can potentially lead to an 
improvement in the estimation error of the OrthoForest as opposed to the Causal Forest. However, it does add 
significant computation cost, as a nuisance function needs to be estimated locally for each target prediction.

There's a lot of grammar flow where it is difficult to follow what "this difference", or "it" means. For example, does the second difference statement mean that "the CausalForest does not couple the similarity metric used in the final estimation and this leads to a decrease in the estimation error as compared to OrthoForest"? By not coupling the similarity metric used in the final estimation, the CausalForest has increased computation cost? This doesn't make sense, so I'm a bit lost.

Is it possible to clarify this?

Start with the forest documentation page linked in the issue, especially the CausalForest and OrthoForest comparison. Trace the references to global versus local nuisance estimation and similarity metrics, then clarify the direction of each difference and the pronouns referring to it. Done means the comparison is grammatically clear and its claims are unambiguous.