# drinf.tmle: drinf.tmle In benkeser/drinf: Computes doubly-robust estimators and doubly-robust confidence intervals

## Description

This function computes the time-varying covariate-adjusted mean of an outcome under a specified treatment assignment using targeted minimum loss-based estimation.

## Usage

 ``` 1 2 3 4 5 6 7 8 9 10``` ```drinf.tmle(L0, L1, L2, A0, A1, abar = c(1, 1), stratify = TRUE, SL.Q = NULL, SL.g = NULL, SL.Qr = NULL, SL.gr = NULL, glm.Q = NULL, glm.g = NULL, guard = c("Q", "g"), universal = FALSE, universalStepSize = 1e-04, printFreq = 50, flucOrd = c("targetQ2", "targetQ1", "targetg1", "targetg0"), return.models = FALSE, maxIter = 20, tolIF = 1/(length(L2)), tolg = 1e-08, tolQ = 1e-08, verbose = TRUE, SL.Q.options = list(family = gaussian()), SL.g.options = list(family = binomial()), glm.Q.options = list(family = gaussian()), return.ltmle = TRUE, only.ltmle = FALSE, return.naive = TRUE, cvFolds = 1, ...) ```

## Arguments

 `L0` A `data.frame` featuring covariates measured at baseline. `L1` A `data.frame` featuring time-varying covariates measured at the first timepoint. `L2` A `vector` outcome of interest `A0` A `vector` treatment delivered at baseline. `A1` A `vector` treatment deliver after `L1` is measured. `abar` A `vector` of length 2 indicating the treatment assignment that is of interest. `stratify` A `boolean` indicating whether to pool across treatment nodes or to estimate outcome regression separately in each category. Should be kept `TRUE` until I have more time to think about how to pool across treatment arms? `SL.Q` A `vector` or `list` specifying the SuperLearner library to be used to estimate the outcome regressions at each time point. See `SuperLearner` package for details. `SL.g` A `vector` or `list` specifying the SuperLearner library to be used to estimate the conditional probability of treatment at each time point. See `SuperLearner` package for details. `SL.Qr` A `vector` or `list` specifying the SuperLearner library to be used to estimate the reduced-dimension regression to protect against misspecification of the outcome regressions. See `SuperLearner` package for details. `SL.gr` A `vector` or `list` specifying the SuperLearner library to be used to estimate the reduced-dimension regression to protect against misspecification of the conditional treatment probabilities. See `SuperLearner` package for details. `glm.Q` A `character` specifying the right-hand side of the `glm` formula used to estimate the outcome regressions at each time point. Only used if `SL.Q = NULL`. `glm.g` A `character` specifying the right-hand side of the `glm` formula used to estimate the conditional probability of treatment at each time point. Only used if `SL.g = NULL`. `guard` A `vector` of `characters`, either `"Q"`, `"g"`, both, or neither (`NULL`). Indicates whether to guard against misspecification of outcome or treatment regressions or both. Currently only works with `c("Q","g")`. `universal` A `boolean` indicating whether to perform TMLE step using locally least favorable parametric submodels (if `FALSE`) or universally least favorable submodels (if `TRUE`) `universalStepSize` A `numeric` indicating the step size for the recursive calculation of universally least favorable submodel. Default is `0.005`. `return.models` A `boolean` indicating whether the models for Q, g, Qr, and gr should be returned with the output. `maxIter` A `numeric` indicating the maximum number of TMLE iterations before stopping. `tolIF` A `numeric` stopping criteria for the TMLE updates based on the empirical average of the estimated influence curve. `tolg` A `numeric` indicating the truncation level for conditional treatment probabilities. `tolQ` A `numeric` indicating the truncation level for transformed outcome regressions. `verbose` A `boolean` indicating whether messages should be printed to indicate progress. `SL.Q.options` A `list` of additional arguments passed to `SuperLearner` for outcome regression fits. `SL.g.options` A `list` of additional arguments passed to `SuperLearner` for condtional treatment probability fits. `glm.Q.options` A `list` of additional arguments passed to `glm` for outcome regression fits. `return.ltmle` A `boolean` indicating whether to compute the LTMLE estimate using a similar iterative updating scheme. `only.ltmle` Only return ltmle (for bootstrapping) `return.naive` A `boolean` indicating whether to return the naive plug-in estimate. `...` Other arguments (not currently used)

## Value

TO DO: Add return values

## Examples

 `1` ```TO DO : Add Examples ```

benkeser/drinf documentation built on May 12, 2019, 11:59 a.m.