IPW.meandifference: Estimating the treatment effect from a partially nested...

In IPWpn: Inverse-Propensity-Weighting for Partially Nested Designs

Description

IPW.meandifference is used to estimate the treatment effect from a partially nested design (where the treatment arm has multiple treatment clusters, and the control arm has no clusters) with the inverse-propensity-weighted mean difference estimation approach adapted to the partially nested design feature.

Usage

IPW.meandifference(
  Y,
  Trt,
  clus,
  Lyz,
  Ly
)

Arguments

Y	An outcome variable
Trt	Treatment assignment indicator (1 for treatment and 0 for control). The treatment arm has multiple treatment clusters, and the control arm has no clusters.
clus	Observed treatment cluster assignment. clus = 0 for the control arm
Lyz	A matrix containing pre-treatment covariates. Lyz can contain two types of covariates. The first type of covariates affect both the treatment assignment and outcome directly. The second type of covariates do not affect the outcome in a given treatment cluster directly, but affect both the treatment assignment and treatment cluster assignment.
Ly	A matrix containing the pre-treatment covariates that affect both the treatment assignment and outcome directly. Ly is the first type of covariates contained in Lyz.

Details

The sandwisch type standard error estimation does not account for the clustering in the treatment arm.

Value

IPW.meandifference returns a list "ipw10.md" containing the following components:

meanDiff	the treatment effect estimate, i.e., the mean difference between the treatment and control arm.
se_sw	the sandwich-type standard error estimate of the treatment effect estimate.
z.wald	the Wald statistic (i.e., meanDiff/se_sw).

Examples

data(dat_obs)
head(dat_obs)
IPW.meandifference(Y=dat_obs$Y,
                   Trt=dat_obs$Trt,
                   clus=dat_obs$clus,
                   Lyz=dat_obs[,c('Ly','Lz')],
                   Ly=dat_obs[,c('Ly')] )

IPWpn documentation built on April 13, 2021, 5:08 p.m.