/synthdid

Synthetic difference in differences

Primary LanguageRBSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

synthdid: Synthetic Difference in Differences Estimation

Build Status

This package implements the synthetic difference in difference estimator (SDID) for the average treatment effect in panel data, as proposed in Arkhangelsky et al (2019). We observe matrices of outcomes Y and binary treatment indicators W that we think of as satisfying Yij = Lij + τij Wij + εij. Here τij is the effect of treatment on the unit i at time j, and we estimate the average effect of treatment when and where it happened: the average of τij over the observations with Wij=1. All treated units must begin treatment simultaneously, so W is a block matrix: Wij = 1 for i > N0 and j > T0 and zero otherwise, with N0 denoting the number of control units and T0 the number of observation times before onset of treatment. This applies, in particular, to the case of a single treated unit or treated period.

This package is currently in beta and the functionality and interface is subject to change.

Some helpful links for getting started:

Installation

The current development version can be installed from source using devtools.

devtools::install_github("synth-inference/synthdid")

Example

library(synthdid)

# Estimate the effect of California Proposition 99 on cigarette consumption
data('california_prop99')
setup = panel.matrices(california_prop99)
tau.hat = synthdid_estimate(setup$Y, setup$N0, setup$T0)
se = sqrt(vcov(tau.hat, method='placebo'))
sprintf('point estimate: %1.2f', tau.hat)
sprintf('95%% CI (%1.2f, %1.2f)', tau.hat - 1.96 * se, tau.hat + 1.96 * se)
plot(tau.hat)

References

Dmitry Arkhangelsky, Susan Athey, David A. Hirshberg, Guido W. Imbens, and Stefan Wager. Synthetic Difference in Differences, 2019. [arxiv]