causalweight (version 1.1.1)

lateweight: Local average treatment effect estimation based on inverse probability weighting

Description

Instrumental variable-based evaluation of local average treatment effects using weighting by the inverse of the instrument propensity score.

Usage

lateweight(
  y,
  d,
  z,
  x,
  LATT = FALSE,
  trim = 0.05,
  logit = FALSE,
  boot = 1999,
  cluster = NULL
)

Value

A lateweight object contains 10 components, effect, se.effect, pval.effect, first, se.first, pval.first, ITT, se.ITT, pval.ITT, and ntrimmed:

effect: local average treatment effect (LATE) among compliers if LATT=FALSE or the local average treatment effect on treated compliers (LATT) if LATT=TRUE.

se.effect: bootstrap-based standard error of the effect.

pval.effect: p-value of the effect.

first: first stage estimate of the complier share if LATT=FALSE or the first stage estimate among treated if LATT=TRUE.

se.first: bootstrap-based standard error of the first stage effect.

pval.first: p-value of the first stage effect.

ITT: intention to treat effect (ITT) of z on y if LATT=FALSE or the ITT among treated if LATT=TRUE.

se.ITT: bootstrap-based standard error of the ITT.

pval.ITT: p-value of the ITT.

ntrimmed: number of discarded (trimmed) observations due to extreme propensity score values.

Arguments

y

Dependent variable, must not contain missings.

d

Treatment, must be binary (either 1 or 0), must not contain missings.

z

Instrument for the endogenous treatment, must be binary (either 1 or 0), must not contain missings.

x

Confounders of the instrument and outcome, must not contain missings.

LATT

If FALSE, the local average treatment effect (LATE) among compliers (whose treatment reacts to the instrument) is estimated. If TRUE, the local average treatment effect on the treated compliers (LATT) is estimated. Default is FALSE.

trim

Trimming rule for discarding observations with extreme propensity scores. If LATT=FALSE, observations with Pr(Z=1|X)<trim or Pr(Z=1|X)>(1-trim) are dropped. If LATT=TRUE, observations with Pr(Z=1|X)>(1-trim) are dropped. Default is 0.05.

logit

If FALSE, probit regression is used for propensity score estimation. If TRUE, logit regression is used. Default is FALSE.

boot

Number of bootstrap replications for estimating standard errors. Default is 1999.

cluster

A cluster ID for block or cluster bootstrapping when units are clustered rather than iid. Must be numerical. Default is NULL (standard bootstrap without clustering).

Details

Estimation of local average treatment effects of a binary endogenous treatment based on a binary instrument that is conditionally valid, implying that all confounders of the instrument and the outcome are observed. Units are weighted by the inverse of their conditional instrument propensities given the observed confounders, which are estimated by probit or logit regression. Standard errors are obtained by bootstrapping the effect.

References

Frölich, M. (2007): "Nonparametric IV estimation of local average treatment effects with covariates", Journal of Econometrics, 139, 35-75.

Examples

# A little example with simulated data (10000 observations)
if (FALSE) {
n=10000
u=rnorm(n)
x=rnorm(n)
z=(0.25*x+rnorm(n)>0)*1
d=(z+0.25*x+0.25*u+rnorm(n)>0.5)*1
y=0.5*d+0.25*x+u
# The true LATE is equal to 0.5
output=lateweight(y=y,d=d,z=z,x=x,trim=0.05,LATT=FALSE,logit=TRUE,boot=19)
cat("LATE: ",round(c(output$effect),3),", standard error: ",
             round(c(output$se.effect),3), ", p-value: ",
             round(c(output$pval.effect),3))
output$ntrimmed}