12  Results

12.1 Bias

  • u represents those where no proxies are utilized
  • SL represents those where super learner was used with the following 4 candidate learners
    1. Logistic regression
    2. MARS (Multivariate Adaptive Regression Splines)
    3. LASSO
    4. XGBoost (Extreme Gradient Boosting)
  • TMLE represents those where TMLE was used
  • DC represents double cross-fit.

Same super learner used for SL and TMLE methods.

Tip

Clearly using proxies improve bias estimates

12.2 Bias (used proxies)

  • SL methods seem to have negligible improvements over non-SL methods in terms of bias.
  • TMLE methods winning in terms of bias.

12.3 MSE

  • TMLE methods winning in terms of MSE.

12.4 Relative Error

  • TMLE methods are have worse relative % error in Model SE estimation.
  • SL methods are winners.

12.5 Coverage

  • TMLE methods are have worse 95% coverage (below 85%).
  • SL methods are winners.
  • But some of these methods were biased, so hard to compare.

12.6 Bias eliminated coverage

  • TMLE methods are have worse 95% bias eliminated coverage (below 85%).