tmle: Targeted minimum loss based estimator (TMLE)

In idiazst/causalquantile: Estimate quantile treatment effects (Causal Quantile)

Description

Targeted minimum loss based estimator (TMLE)

Usage

tmle(y, t, Q, g, q)

Arguments

y	Vector with outcome values.
t	Vector with binary treatment indicator.
Q	Conditional outcome distribution estimate. This should come in the for of an n x p matrix, where each column represents a conditional quantile. (See Kang & Schafer example)
g	Propensity score.
q	Quantile to be computed (e.g., q = 0.5 for the median.)

Value

A point estimate

Examples

n.quant <- 500
formT <- T ~ X1+X2+X3+X4
formY <- Y ~ X1+X2+X3+X4
data <- KSdatagen(1000)
X <- data$X
Y <- data$Y
T <- data$T
fitT <- glm(formT, data = data.frame(T=T, X), family = binomial)
fitY1 <- lm(formY, data = data.frame(Y=Y, T=T, X), subset = T == 1)
fitY0 <- lm(formY, data = data.frame(Y=Y, T=T, X), subset = T == 0)
median1 <- predict(fitY1, newdata = data.frame(T=1, X))
median0 <- predict(fitY0, newdata = data.frame(T=0, X))
Q1 <- sapply(seq(1/n.quant, 1 - 1/n.quant, 1/n.quant),
            function(q)qnorm(q, mean = median1, sd = summary(fitY1)$sigma))
Q0 <- sapply(seq(1/n.quant, 1 - 1/n.quant, 1/n.quant),
            function(q)qnorm(q, mean = median0, sd = summary(fitY0)$sigma))
g1 <- trim(predict(fitT, type = 'response'))
ame <- tmle(Y, T, Q1, g1, 0.5) - tmle(Y, 1 - T, Q0, 1 - g1, 0.5)

idiazst/causalquantile documentation built on May 18, 2019, 2:32 a.m.