power.sen: Power of sensitivity analysis

In CrossScreening: Cross-Screening in Observational Studies that Test Many Hypotheses

Description

Power of sensitivity analysis

Usage

power.sen(mu.F = 1/2, sigma.F = sqrt(1/3), d = NULL, mm = c(2, 2, 2),
  gamma = 1, alpha = 0.05, I = 100, approx.method = c("changing.alpha",
  "fixed.alpha"), score.method = c("approximate", "exact"))

Arguments

mu.F	mean of the signed rank statistic
sigma.F	standard deviation of the signed rank statistic
d	empirical data used to estimate mu.F and sigma.F by jackknife
mm	test statistic, either a vector of length 3 or a matrix of three rows where each column corresponds to a U-statistic. Default is the (approximate) Wilcoxon's signed rank test.
gamma	target sensitivity level
alpha	target significance level
I	sample size
approx.method	which approximation method to use?
score.method	either approximate score or exact score

Details

If approx.method is "fixed.alpha", then the significance level alpha is considered fixed and the corresponding quantile negligible. Otherwise we also use the alpha-quantile in the approximation formula. For more detail, see the reference.

Value

power of the sensitivity analysis, possibly a vector if mm has multiple columns.

References

Qingyuan Zhao. On sensitivity value of pair-matched observational studies. arXiv 1702.03442, https://arxiv.org/abs/1702.03442.

Examples

power.sen(d = rnorm(100) + 0.5, I = 200, gamma = 2)

## The following code reproduces an example of power analysis in Zhao (2017)
power.sen(0.76, sqrt(0.26), gamma = 2.5, I = 200)
power.sen(0.76, sqrt(0.26), gamma = 2.5, I = 200, approx.method = "fixed.alpha")

CrossScreening documentation built on May 2, 2019, 5:15 a.m.