search

romainkp / stremr

Streamlined Estimation for Static, Dynamic and Stochastic Treatment Regimes in Longitudinal Data
MIT License
38 stars 8 forks source link

stratifyQ_by_rule #41

Open marycombs opened 6 years ago

marycombs commented 6 years ago

The documentation in fit_GCOMP() for stratifyQ_by_rule says,

Set to 'TRUE' for stratifying the fit of Q (the outcome model) by rule-followers only.

Does this mean that non-rule-followers are excluded at each time point from the hazard estimate? I don't see separate "stratified" estimates.

Thanks,

osofr commented 6 years ago

Nothing to do with the hazard.

The outcome model is the fit of the expectation E[Y|\bar{L},\bar{A}] (over many time-points).

If the observed \bar{A} is not consistent with following the rule (dynamic or static), the above regression will be only fit among the observations who are actually following the rule.

Claro?

marycombs commented 6 years ago

So is this most applicable for a target parameter like a data-adaptive parameter? When you intervene on treatment according to a rule, isn't everyone following that rule at each time point under the intervened treatment? Under what circumstance would you calculated a treatment-specific mean but only over individuals who actually followed that treatment?