bpCausal: Bayesian Causal inference with TSCS data

View source: R/blasso_default.R

bpCausalR Documentation

Bayesian Causal inference with TSCS data

Description

Bayesian Causal inference with TSCS data

Usage

bpCausal(data, index, Yname, Dname, Xname, Zname, Aname, 
               re = "unit", ar1 = TRUE, r, 
               niter = 10000, burn = 5000,
               xlasso, zlasso, alasso, flasso,
               a1 = 0.001, a2 = 0.001, b1 = 0.001, b2 = 0.001,
               c1 = 0.001, c2 = 0.001, p1 = 0.001, p2 = 0.001)  

Arguments

data

a data frame (must be with a dichotomous treatment but balanced is not reqiored).

index

a two-element string vector specifying the unit (group) and time indicators. Must be of length 2.

Yname

outcome.

Dname

treatment.

Xname

covariates that have constant effects.

Zname

covariates that have unit-level random effects.

Aname

covariates that have time-level random effects.

re

a string indicating whether additive unit or time fixed effects will be imposed. Must be one of the following, "none", "unit", "time", or "two-way". The default is "unit".

ar1

a logical flag indicating whether time-varying parameters and factors are AR(1) process or independent across periods.

r

an integer specifying the (upper bound of) number of factors.

niter

an integer specifying the iterations of MCMC run. Default value is 10000.

burn

an integer specifying the length of burn in iterations of MCMC run. Default value is 5000.

xlasso

a logical flag indicating whether to assign constant coefficients hierachical shrinkage priors.

zlasso

a logical flag indicating whether to assign unit-level random coefficients hierachical shrinkage priors.

alasso

a logical flag indicating whether to assign time-level random coefficients hierachical shrinkage priors.

flasso

a logical flag indicating whether to assign factor loadings hierachical shrinkage priors for factor selection.

a1

shape parameter for the Gamma prior for constant coefficients shrinkage.

a2

rate parameter for the Gamma prior for constant coefficients shrinkage.

b1

shape parameter for the Gamma prior for unit-level random coefficients shrinkage.

b2

rate parameter for the Gamma prior for unit-level random coefficients shrinkage.

c1

shape parameter for the Gamma prior for time-level random coefficients shrinkage.

c2

rate parameter for the Gamma prior for unit-level random coefficients shrinkage.

p1

shape parameter for the Gamma prior for factor loadings shrinkage.

p2

rate parameter for the Gamma prior for factor loadings shrinkage.

Details

bpCausal implements counterfactual estimators in TSCS data analysis. It simulates counterfactual outcomes under control for observations under treatment based on posterior predictive distributions.

Value

yct

trace of simulated counterfactual outcomes under control for each observation under treated.

sigma2

trace of simulated variance for the error term.

beta

trace of constant coefficients.

alpha

trace of unit-level random coefficients.

xi

trace of time-level random coefficients.

gamma

trace of factor loadings.

f

trace of factors.

pxi

trace of ar1 parameter for time-level random coefficients.

pf

trace of ar1 parameter for factors.

wa

trace of square roots for prior variance of unit-level random coefficients.

wxi

trace of square roots for prior variance of time-level random coefficients.

wg

trace of square roots for prior variance of factor loadings.

sb2

trace of prior variance of constant coefficients.

swa2

trace of prior variance of square roots for prior variance of unit-level random coefficients.

swxi2

trace of prior variance of square roots for prior variance of time-level random coefficients.

swg2

trace of prior variance of square roots for prior variance of factor loadings.

raw.id.tr

a vector of names of treated units.

id.tr

a vector of integer indicating which units are under treated.

time.tr

a vector of periods (see index) for treated units.

tr.unit.pos

a vector of integer indicating which observations in the dataset correspond to a treated unit.

rela.time.tr

a vector of period relative to the occurrence of treatment for each observation that corresponds to a treated unit.

References

A Bayesian Alternative to Synthetic Control for Comparative Case Studies. Pang et. al (2021).

Examples

library(bpCausal)
data(bpCausal)
out <- bpCausal(data = simdata, index = c("id", "time"), 
               Yname = "Y", Dname = "D", 
               Xname = c("X1", "X2"), Zname = NULL, Aname = c("X1", "X2"), 
               re = "time", r = 10, niter = 20000)

liulch/bpCausal documentation built on Jan. 16, 2024, 10:59 p.m.