LATEtest: LATEtest

In farbmacher/LATEtest: Local Instrument Validity Tests with Causal Forests

Description

Main function: Prepares data, calls help functions, stores results. See the LATEtest website on https://github.com/farbmacher/LATEtest for more information, documentation and examples. Please report any bugs to [e-mail]

Usage

LATEtest(data, covars, huge = FALSE, tree_fraction = 0.5,
  minsize = 100, cp = 0, slice = c("equidistant", "equisized"),
  subsets = 4, alpha = 0.05, seed = 10101)

Arguments

data	provides dataset: outcome must be labelled Y, treatment D and instrument Z
covars	provides names of covariables in data (e.g., "Xvar1" "Xvar2" "Xvar3")
huge	if set to TRUE, model for orthogonalization learned on random subset of data (size defined in tree_fraction)
tree_fraction	fraction of the data used to build each tree of causal forest (and if huge==T, also used for regression forests); default=0.5
minsize	causal forest insists on at least "minsize" treated and "minsize" control observations per leaf, and pruned tree insists on at least 2*"minsize" observations in the additional trees to identify the promising subgroups
cp	sets complexity parameter which rpart uses to fit the tree before pruning; default=0
slice	choose "equisized" for equisized subsets (quantile-based) of the outcome or "equidistant" for equidistant subsets (grid-based) of the outcome
subsets	number of subsets used to discretize the outcome
alpha	nominal significance level of the test
seed	set random seed number

Value

list of the pruned trees, test results

Examples

# Generate data:
library("LATEtest")
library("rpart.plot")

n = 3000; p = 3; rho=0.3
u <- rnorm(n)
v <- rho * u + sqrt(1 - rho^2) * rnorm(n)
X <- matrix(rnorm(n * p), n, p)
colnames(X) <- paste("Xvar", 1:p, sep="")
Z <- rbinom(n, size = 1, prob = 0.5)
D<-as.numeric(0.2 * Z + v > 0)
# local violation of the exclusion restriction:
gamma <- as.numeric(ifelse(X[, 2] < -1, 1.25, 0))
Y <- as.vector(D + gamma * Z + u)
data <- as.data.frame(cbind(Y,D,Z,X))

# Perform test:
covars = paste0(colnames(data)[4:ncol(data)])
test <- LATEtest(data = data, covars = covars, subsets = 4, alpha = 0.05)
test

# Draw plot of pruned tree that led to local violation of LATE assumptions:
maxtree <- eval(parse(text = paste("test$treelist$tree_", test$maxTtree$label, test$maxTtree$J,sep = "")))
rpart.plot(maxtree, extra = 101, box.palette = "GyRd", shadow.col = "gray", nn = TRUE, roundint = FALSE)

farbmacher/LATEtest documentation built on Nov. 20, 2020, 10:13 a.m.