Multiple Treated

In bsynth: Bayesian Synthetic Control

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  echo = TRUE,
  warning = FALSE,
  message = FALSE,
  out.width = '90%', 
  out.height = '600px',
  fig.align='center'
)
set.seed(123)

Background

In this vignette I show how {bsynth} can be used for causal estimation when you have multiple treated accounts. In particular, I use the synthetic data exaple from Xu, Yiqing, 2017. In this example, the treatment starts in period 21 and increases by one each period (e.g. the effect is 5 period 25 and 10 in period 30).

knitr::include_graphics(path = "https://ignacio.martinez.fyi/synthetic_control/gsynth.png")

Bayesian Synthetic Control with Covariates

library(bsynth)
ci_width <- 0.95
data(gsynth, package = "gsynth")
dplyr::glimpse(simdata)

outcome_data <- simdata %>% 
  dplyr::select(time, id, D, Y)

covariates <- simdata %>% 
  dplyr::select(time, id, X1, X2)

synth <-
  bsynth::bayesianSynth$new(
    data = outcome_data,
    time = time,
    id = id,
    treated = D,
    outcome = Y, 
    ci_width = ci_width,
    covariates = covariates
  )

synth$timeTiles +
  ggplot2::theme(text = ggplot2::element_text(size=6))

Fit

synth$fit()

Visualize the synthetic controls for each treated unit

synth$synthetic  

Visualize the treatment effect

synth$effectPlot(subset = c("Average"), facet = FALSE) + 
  ggplot2::scale_y_continuous(breaks=seq(-2,12,2))

bsynth documentation built on June 25, 2024, 5:08 p.m.