estimateG function

Estimate Treatment or Missingness Mechanism

An internal function called by the tmle function to obtain an estimate of conditional treatment assignment probabiliites $P(A=1|W)$, and conditional probabilites for missingness, $P(Delta=1|A,W)$. The estimate can be based on user-supplied values, a user-supplied regression formula, or a data-adaptive super learner fit. If the SuperLearner package is not available, and there are no user-specifications, estimation is carried out using main terms regression with glm. These main terms-based estimates may yield poor results.

estimateG(d, g1W, gform, SL.library, id, V, verbose, message, 
        outcome="A", newdata=d, discreteSL, obsWeights)

Arguments

• d: dataframe with binary dependent variable in the first column, predictors in remaining columns
• g1W: vector of values for $P(A=1|W)$, $P(Z=1|A,W)$, or $P(Delta=1|Z,A,W)$
• gform: regression formula of the form A~W1, (dependent variable is one of $A,Z,D$) if specified this overrides the call to SuperLearner
• SL.library: vector of prediction algorithms used by SuperLearner, default value is (SL.glm , tmle.SL.dbarts.k.5 , SL.gam )
• id: subject identifier
• V: Number of cross validation folds for Super Learning
• verbose: status messages printed if set to TRUE
• message: text specifies whether treatment or missingness mechanism is being estimated
• outcome: A, D, Z to indicate which quantity is being estimated.
• newdata: optional dataset to be used for prediction after fitting on d.
• discreteSL: If true, returns discrete SL estimates, otherwise ensemble estimates. Ignored when SL is not used.
• obsWeights: sampling weights

Returns

• g1W: a vector containing values for $P(A=1|W)$, matrix for $P(Z=1|A,W)$, evaluated at A=0, A=1, or matrix $P(Delta=1|Z,A,W))$ evaluated at (0,0), (0,1), (1,0), (1,1)

• coef: coefficients for each term in the working model used for estimation if glm was used

• type: estimation procedure

Author(s)

Susan Gruber

See Also

tmle, estimateQ, calcParameters, tmleMSM, calcSigma