Frome DESCRIPTION: tmle implements targeted maximum likelihood estimation, first described in van der Laan and Rubin, 2006 (Targeted Maximum Likelihood Learning, The International Journal of biostatistics, 2(1), 1006. This implementation calculates the adjusted marginal difference in mean outcome associated with a binary point treatment, for continuous or binary outcomes. Relative risk and odds ratio estimates are also reported for binary outcomes. Missingness in the outcome is allowed, but not in treatment assignment or baseline covariate values. Effect estimation stratified by a binary mediating variable is also available. The population mean is calculated when there is missingness, and no variation in the treatment assignment. An ID argument can be used to identify repeated measures. Default settings call SuperLearner to estimate the Q and g portions of the likelihood, unless values or a user-supplied regression function are passed in as arguments.

From comment in tmle.R: Targeted Maximum Likelihood Estimation for non-parametric estimation of the marginal effect of a binary point treatment, adjusting for treatment (g) and missingness (g.Delta) mechanisms Parameters include: additive treatment effect: E_W[E(Y|A=1,W) - E(Y|A=0,W)] and, for binary outcomes, relative risk(RR) and odds ratio(RR) mu1 = E_W[E(Y|A=1,W)], mu0 = E_W[E(Y|A=0,W)] RR = mu1/mu0 OR = mu1/(1-mu1)/(mu0 / (1-mu0)) Controlled direct effect estimation available for optional binary Z intermediate variable, (P(DeltaY|Z,A,W, Delta=1)P(Delta|Z,A,W)P(Z|A,W)P(A|W)P(W)) EY1 parameter estimated when there is missinginess and no treatment assignment author: Susan Gruber, [email protected] date: September 25, 2010 revised: December 20, 2010

Models or estimated values for Q, g, g.Z, g.Delta can be user-supplied or estimated using super learner. Cross-validated inital Q can be obtained by specifying cvQinit=TRUE