11  Compare

Three broad class of methods

All weighted propensity score versions.

  1. No super learner used
  • Only investigator specified variables (no proxies)
  • Kitchen sink (no variable selection)
  • hdPS
  • LASSO (recurrent variables being associated with the outcome), or
  • hybrid (first hdPS, and then LASSO)
  1. Super learner versions of the above
  • 2 set of super learner used (3 learners vs 4)
  1. TMLE versions of the above
  • Also includes a double cross-fit version (very time consuming)