search

tlverse / tmle3shift

🎯 :game_die: Targeted Learning of the Causal Effects of Stochastic Interventions
https://tlverse.org/tmle3shift
GNU General Public License v3.0
16 stars 1 forks source link

Parameter with Boundary Guards for Shifting #5

Closed nhejazi closed 6 years ago

nhejazi commented 6 years ago

This PR includes new shift functions that define a parameter that is the counterfactual outcome under a particular user-specified shift of the intervention, incorporating a data-adaptive approach to shifting. That is, the counterfactual value of the intervention is set to be

  1. the observed intervention value (or natural value of the treatment) plus the specified value of the shift, if there is support in the data for this new value, conditional on baseline covariates; or,
  2. the highest (or lowest, for a negative shift value) value of the intervention for which there is support in the data, conditional on baseline covariates.

This represents a potential improvement to the types of shift functions considered in previous work.