Placebo doubly-robust difference-in-differences — morie_placebo_dr

Runs morie_dr_overlap_weighted on the change between two PRE-treatment periods, where the true effect is zero by construction.

Usage

morie_placebo_dr_did(y_pre1, y_pre2, D, X, ...)

Arguments

y_pre1, y_pre2: outcomes in two pre-treatment periods.
D: 0/1 treatment (as eventually assigned).
X: covariates.
...: passed to morie_dr_overlap_weighted.

Value

list with placebo_effect, se, z, p_value, passed, min_detectable.

Details

The value of running the placebo through the SAME machinery as the headline estimate is that any bias in the estimator itself shows up here rather than being attributed to the treatment.

Failure is informative; passing is weak. A pre-trend smaller than the effect of interest is not evidence of parallel trends, only evidence that the sample could not detect a larger one – min_detectable says how large.

References

Roth, J. (2022). Pretest with caution: event-study estimates after testing for parallel trends. AER: Insights, 4(3), 305-322.

Examples

set.seed(1)
X <- matrix(rnorm(400), ncol = 2)
D <- rbinom(200, 1, plogis(X[, 1]))
morie_placebo_dr_did(rnorm(200), rnorm(200), D, X)$passed
#> [1] TRUE