Pre-treatment falsification (placebo) regression — morie_causal_falsification

Regresses a PRE-treatment outcome on the treatment indicator. Since treatment cannot have caused it, a non-zero coefficient is evidence the design is broken.

Usage

morie_causal_falsification_test(y_pre, treat, X_baseline = NULL)

Arguments

y_pre: pre-treatment outcome.
treat: 0/1 treatment indicator.
X_baseline: optional baseline covariates.

Value

list with estimate, se, z, p_value, passed, min_detectable_effect.

Details

The asymmetry is the point and is routinely misread. FAILING is informative – it shows the groups already differed. PASSING is not: a null here has only the power the sample gives it, and rules out effects larger than about min_detectable_effect and nothing smaller. It does not establish parallel trends.

References

Angrist, J. D. and Pischke, J.-S. (2009). Mostly Harmless Econometrics, Sec. 5.2. Princeton University Press.

Examples

set.seed(1)
morie_causal_falsification_test(rnorm(200), rbinom(200, 1, 0.5))$passed
#> [1] TRUE