A difference-in-differences within each of two groups, then the
difference between them.
Usage
morie_causal_did_three_way(y, treated, post, group)
Arguments
- y
outcome.
- treated
0/1 treated indicator.
- post
0/1 post-period indicator.
- group
0/1 eligible-group indicator.
Value
list with ddd, did_eligible, did_placebo,
cell_means, se, z, p_value.
Details
The point is to relax parallel trends: DiD needs the trend to be the
same across arms, while DDD only needs the VIOLATION of parallel
trends to be common to both groups. That is weaker, not absent –
inspect did_placebo, because a large value means the third
difference is doing heavy lifting and the identifying assumption has
moved rather than gone away.
References
Gruber, J. (1994). The incidence of mandated maternity
benefits. American Economic Review, 84(3), 622-641.
Examples
set.seed(1)
n <- 400
g <- rbinom(n, 1, 0.5)
tt <- rbinom(n, 1, 0.5)
p <- rbinom(n, 1, 0.5)
y <- 1 + g + tt + p + 2 * g * tt * p + rnorm(n)
round(morie_causal_did_three_way(y, tt, p, g)$ddd, 2)
#> [1] 1.93
