Nonparametric estimation of decomposition term for causal mediation analysis
with stochastic interventions

medshift(W, A, Z, Y, delta, g_lrnrs = sl3::Lrnr_glm_fast$new(family =
stats::binomial()), e_lrnrs = sl3::Lrnr_glm_fast$new(family =
stats::binomial()), m_lrnrs = sl3::Lrnr_glm_fast$new(),
phi_lrnrs = sl3::Lrnr_glm_fast$new(), estimator = c("onestep",
"tmle", "substitution", "reweighted"), estimator_args = list(cv_folds =
10, max_iter = 10000, step_size = 1e-06))

## Arguments

W |
A `matrix` , `data.frame` , or similar corresponding to a
set of baseline covariates. |

A |
A `numeric` vector corresponding to a treatment variable. The
parameter of interest is defined as a location shift of this quantity. |

Z |
A `numeric` vector, `matrix` , `data.frame` , or
similar corresponding to a set of mediators (on the causal pathway between
the intervention A and the outcome Y). |

Y |
A `numeric` vector corresponding to an outcome variable. |

delta |
A `numeric` value indicating the degree of shift in the
intervention to be used in defining the causal quantity of interest. In the
case of binary interventions, this takes the form of an incremental
propensity score shift, acting as a multiplier of the probability with which
a given observational unit receives the intervention (EH Kennedy, 2018,
JASA; <doi:10.1080/01621459.2017.1422737>). |

g_lrnrs |
A `Stack` object, or other learner class (inheriting from
`Lrnr_base` ), containing a single or set of instantiated learners from
the `sl3` package, to be used in fitting a model for the propensity
score, i.e., g = P(A | W). |

e_lrnrs |
A `Stack` object, or other learner class (inheriting from
`Lrnr_base` ), containing a single or set of instantiated learners from
the `sl3` package, to be used in fitting a cleverly parameterized
propensity score that includes the mediators, i.e., e = P(A | Z, W). |

m_lrnrs |
A `Stack` object, or other learner class (inheriting from
`Lrnr_base` ), containing a single or set of instantiated learners from
the `sl3` package, to be used in fitting the outcome regression, i.e.,
m(A, Z, W). |

phi_lrnrs |
A `Stack` object, or other learner class (inheriting
from `Lrnr_base` ), containing a single or set of instantiated learners
from the `sl3` package, to be used in fitting a reduced regression
useful for computing the efficient one-step estimator, i.e., phi(W) =
E[m(A = 1, Z, W) - m(A = 0, Z, W) | W). |

estimator |
The desired estimator of the natural direct effect to be
computed. Currently, choices are limited to a substitution estimator, a
re-weighted estimator, and an efficient one-step estimator. The interested
user should consider consulting Díaz & Hejazi (2019+) for a comparative
investigation of each of these estimators. |

estimator_args |
A `list` of extra arguments to be passed (via
`...` ) to the function call for the specified estimator. The default
is so chosen as to allow the number of folds used in computing the AIPW
estimator to be easily tweaked. Refer to the documentation for functions
`est_onestep` , `est_ipw` , and
`est_substitution` for details on what other arguments may be
specified through this mechanism. For the option `"tmle"` , there is
heavy reliance on the architecture provided by the `tmle3` package. |