ic.data.examp: Function for calculating the influence function used for the...

In amp: Statistical Test for the Multivariate Point Null Hypotheses

Function for calculating the influence function used for the real data example.

Description

Function for calculating the influence function used for the real data example.

Usage

ic.data.examp(obs_data, what = "both", control = NULL)

Arguments

obs_data	the observed data. The first column should be the outcome.
what	the desired return value. Should be one of "ic" (influence curve), "est" (estimate), or "both".
control	any other control parameters to be passed to the estimator.

Value

If what is

- "est", then return the estimated parameter.

- "ic", then return the estimated IC of the parameter estimate.

- "both", then return both the parameter estimate and corresponding estimated IC.

Examples

expit <- function(x) exp(x) / (1 + exp(x))
ws <- matrix(rnorm(3000), ncol = 3)
probs <- expit(ws  %*% c(-1, 0, 2))
y <- rbinom(n = nrow(probs), size = 1, prob = probs[, 1])
wts <-   abs(rnorm(length(y))) + 1
wts <- length(wts) * wts / sum(wts)
cats <- rep(1:10, 100)
obs_dat <- cbind(y, "cat" = cats, "wt" = wts, ws)
est_ic <- ic.data.examp(obs_dat, what = "both")
my_est <- est_ic$est
my_ic <- est_ic$ic / nrow(ws)
var_mat <- t(my_ic) %*% my_ic
sqrt(diag(var_mat))
for(cov_idx in 1:ncol(ws)){
 print(summary(stats::glm(y ~ ws[, cov_idx], weights = obs_dat[, "wt"],
                    family = binomial))$coefficients[2, 1:2])
}

amp documentation built on April 6, 2022, 9:06 a.m.