get_ifvals: Estimation of influence functions for the bounds on ATE

In matteobonvini/sensitivitypuc: Sensitivity Analysis via the Proportion of Unmeasured Confounding (PUC)

Description

get_ifvals is the main function for getting the influence function values used to construct the estimator of either the lower and upper bound curves.

Usage

get_ifvals(n, neps, ndelta, upper, nu, tau, g, nsplits, quant, nobs_fold)

Arguments

n	sample size.
upper	boolean for whether the upper or the lower bound curve needs to be computed.
nu	a list of size nsplits, where element j is a (num obs in split j)x1 matrix containing the influence function values for EE(Y|A = 1, X) - E(Y|A = 0, X) computed using regressions fns estimated using all obs except those in fold j and evaluated at obs in fold j.
tau	a list of size nsplits, where element j is a (num obs in split j)x1 matrix containing the influence function values for Eg(eta) computed using regressions fns estimated using all obs except those in fold j and evaluated at obs in fold j.
eps	vector of arbitrary length specifying the values for the proportion of confounding where the lower and upper bounds curves are evaluated.
ghatmat	a list of size nsplits, where element j is a (num obs in split j)xlength(delta) matrix containing the values of g(eta) for lower bound if upper = FALSE, for the upper bound otherwise, computed using regressions fns estimated using all obs except those in fold j and evaluated at obs in fold j.
delta	vector of delta values. length(delta) should match ncol(ghatmat).

Value

A list containing:

ifvals	a n x length(eps) x length(delta) array containing the influence functions (varphi_l or varphi_u in manuscript) evaluated at the observed X as a function of epsilon and delta;
lambda	a list of size nsplits, where element j contains a nxlength(eps)xlength(delta) array containing the indicator ghatmat <= q (if upper = FALSE) or ghatmat > q (if upper = TRUE), where q is eps-quantile or 1-eps quantile of ghatmat and, computed using regression functions estimated using all folds but j and evaluated using obs at fold j.
quant	a list of size nsplits, where element j is a (num eps) x (num delta) matrix containing estimates of eps-quantile (if upper = FALSE) or (1-eps) quantiles (if upper = TRUE) of ghatmat, where the regression functions are computed and evaluated using all obs except those in fold j. See get_quant_g.
lambdaq	a list of size nsplits containing the product lambda * quant.

See Also

if_gamma, if_tau.

matteobonvini/sensitivitypuc documentation built on Dec. 9, 2020, 2:24 a.m.