C-TMLE for Continuous Tuning
A TMLE of a causal quantity of interest first constructs an initial estimator of the relevant part of the likelihood of the data and then updates this initial estimator along a least favorable parametric model that uses the initial estimator as an off-set. The least favorable parametric model typically depends on an orthogonal nuisance parameter such as the treatment and censoring mechanism. This nuisance parameter is not needed to evaluate the target parameter, and, in fact, is orthogonal to the target parameter in the sense that a maximum likelihood estimator would completely ignore this nuisance parameter, or, at least, its scores are orthogonal to the scores of the relevant part of the likelihood.
- S. Gruber, M.J. van der Laan, A targeted maximum likelihood estimator of a causal effect on a bounded continuous outcome. Int. J. Biostat. 6(1), Article 26 (2010b)Google Scholar
- M.J. van der Laan, Targeted estimation of nuisance parameters to obtain valid statistical inference. Int. J. Biostat. 10(1), 29–57 (2014b)Google Scholar
- M.J. van der Laan, S. Gruber, Collaborative double robust penalized targeted maximum likelihood estimation. Int. J. Biostat. 6(1), Article 17 (2010)Google Scholar
- H. Wang, S. Rose, M.J. van der Laan, Finding quantitative trait loci genes with collaborative targeted maximum likelihood learning. Stat. Probab. Lett. 81(7), 792–796 (2011a)Google Scholar