ipw_shift {haldensify}

IPW Estimates of the Causal Effects of Stochatic Shift Interventions

Description

IPW Estimates of the Causal Effects of Stochatic Shift Interventions

Usage

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
)

Arguments

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".

Examples

# 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"
)