bootsens.os: Sensitivity analysis for observational studies

In qingyuanzhao/bootsens: Bootstrapping sensitivity analysis

Description

Sensitivity analysis for observational studies

Usage

bootsens.os(
  A,
  X,
  Y,
  gamma = 0,
  alpha = 0.05,
  estimand = c("ate", "att"),
  reg.adjust = FALSE,
  parallel = FALSE,
  B = 1000,
  warm.start = FALSE
)

extrema.os(
  A,
  X,
  Y,
  gamma = 0,
  estimand = c("ate", "att"),
  reg.adjust = FALSE,
  start = NULL
)

Arguments

A	Indicator of missingness
X	A matrix of covariates
Y	Outcome
gamma	Sensitivity parameter (log odds ratio)
alpha	Significance level
estimand	Either "ate" (average treatment effect) or "att" (average treatment effect on the treated)
reg.adjust	Should regression adjustment (augmented IPW) be used?
parallel	Should parallel computing be used?
B	Number of Bootstrap resamples.
warm.start	Warm start the refitting of propensity score model (doesn't seem to help).
start	Starting values for the propensity score model to warm start glm.

Value

A (1 - alpha) confidence interval.

Extrema (an interval).

Functions

• extrema.os: Obtain extrema of IPW estimator for observational studies

Examples

## Fish consumption and blood mercury dataset
require(CrossScreening)
data(nhanes.fish)
A <- as.numeric(nhanes.fish$fish.level == "high")
X <- nhanes.fish[, c("gender", "age", "income", "income.missing",
"race", "education", "smoking.ever", "smoking.now")]
X$race <- factor(X$race)
X1 <- model.matrix(~ . - 1, X)
Y <- log2(nhanes.fish$o.LBXTHG)

## Assuming no unmeasure confounder (i.e. gamma = 0 or Gamma = e^0 = 1)
extrema.os(A, X1, Y) # point estimate
bootsens.os(A, X1, Y, parallel = FALSE) # confidence interval

## Sensitivity analysis (gamma = 1, i.e. Gamma = e^1)
extrema.os(A, X1, Y, gamma = 1) # point estimate
bootsens.os(A, X1, Y, gamma = 1, parallel = FALSE) # confidence interval

## Sensitivity analysis using regression adjustment (gamma = 1, i.e. Gamma = e^1)
extrema.os(A, X1, Y, gamma = 1, reg.adjust = TRUE) # point estimate
bootsens.os(A, X1, Y, gamma = 1, reg.adjust = TRUE, parallel = FALSE) # confidence interval

## Reproduce the results for matching
d <- nhanes.log2diff()$o.LBXTHG
require(sensitivitymw)
senmwCI(d, gamma = exp(1), alpha = 0.1, one.sided = FALSE)

qingyuanzhao/bootsens documentation built on Jan. 17, 2021, 7:40 p.m.