ipw_shift: IPW Estimates of the Causal Effects of Stochatic Shift...
In haldensify: Highly Adaptive Lasso Conditional Density Estimation

IPW Estimates of the Causal Effects of Stochatic Shift Interventions

ipw_shift(
  W,
  A,
  Y,
  delta,
  n_bins = make_bins(A, "hist"),
  cv_folds = 10L,
  lambda_seq,
  ...,
  bin_type = c("equal_range", "equal_mass"),
  trim_density = FALSE,
  undersmooth_type = c("dcar", "plateau", "gcv", "all"),
  bootstrap = FALSE,
  n_boot = 1000L
)

`W`	A `matrix`, `data.frame`, or similar containing a set of baseline covariates.
`A`	A `numeric` vector corresponding to a exposure variable. The parameter of interest is defined as a location shift of this quantity.
`Y`	A `numeric` vector of the observed outcomes.
`delta`	A `numeric` value indicating the shift in the exposure to be used in defining the target parameter. This is defined with respect to the scale of the exposure (A).
`n_bins`	A `numeric`, scalar or vector, indicating the number of bins into which the support of A is to be partitioned for constructing conditional density estimates.
`cv_folds`	A `numeric` giving the number of folds to be used for cross-validation. Note that this form of sample splitting is used for the selection of tuning parameters by empirical risk minimization, not for the estimation of nuisance parameters (i.e., to relax regularity conditions).
`lambda_seq`	A `numeric` sequence of the regularization parameter (L1 norm of HAL coefficients) to be used in fitting HAL models.
`...`	Additional arguments for model fitting to be passed directly to `haldensify`.
`bin_type`	A `character` indicating the strategy to be used in creating bins along the observed support of `A`. For bins of equal range, use `"equal_range"`; to ensure each bin has the same number of observations, use instead `"equal_mass"`. For more information, see documentation of `grid_type` in `haldensify`.
`trim_density`	A `logical` indicating whether estimates of the conditional density should be trimmed. Refer to the `trim` argument of the `predict` method of `haldensify` for details. The default is `FALSE` since propensity score truncation can lead to estimation bias.
`undersmooth_type`	A `character` indicating the selection strategy to be used in identifying an efficent IPW estimator. The choices include `"gcv"` for global cross-validation, `"dcar"` for solving the IPW representation of the EIF through, and `"plateau"` for an approach that balances changes in the parameter estimate and its mean squared error, based on Lepski's method. The option `"all"` produces results based on all three selection strategies, sharing redundant computation between each.
`bootstrap`	A `logical` indicating whether the estimator variance should be approximated using the nonparametric bootstrap. The default is `FALSE`, in which case the empirical variances of the IPW estimating function and the EIF are used for for estimator selection and for variance estimation, respectively. When set to `TRUE`, the bootstrap variance is used for both of these purposes instead. Note that the bootstrap is very computationally intensive and scales relatively poorly.
`n_boot`	A `numeric` giving the number of bootstrap re-samples to be used in computing the mean squared error as part of the plateau selector criterion. Ignored when `undersmooth_type` is not `"plateau"`.

# simulate data
n_obs <- 50
W1 <- rbinom(n_obs, 1, 0.6)
W2 <- rbinom(n_obs, 1, 0.2)
A <- rnorm(n_obs, (2 * W1 - W2 - W1 * W2), 2)
Y <- rbinom(n_obs, 1, plogis(3 * A + W1 + W2 - W1 * W2))

# fit the IPW estimator
est_ipw_shift <- ipw_shift(
  W = cbind(W1, W2), A = A, Y = Y,
  delta = 0.5, n_bins = 3L, cv_folds = 2L,
  lambda_seq = exp(seq(-1, -10, length = 100L)),
  # arguments passed to hal9001::fit_hal()
  max_degree = 1,
  # ...continue arguments for IPW
  undersmooth_type = "gcv"
)