10  Sensitivity analysis

11 Sensitivity analysis

didgpu ships with three sensitivity-analysis wrappers that take any fitted didgpu / didgpu_cs / didgpu_fect object and return the same robust-confidence-set output the underlying methodology papers do.

11.1 HonestDiD (relative magnitudes, smoothness)

hd <- didgpu_honest_did(fit, M_bar = c(0, 0.5, 1, 2))
hd
plot(hd)

Wraps [@rambachan2023more]’s HonestDiD package. The “smoothness” (\(M\)) and “relative magnitudes” (\(\bar{M}\)) restrictions both work unchanged; the FLCI variant is also available.

11.2 Rambachan-Roth equivalence (passes-at-delta)

eq <- didgpu_equivalence(fit, delta = 0.05, alpha = 0.05)
eq

The “equivalence-style” test reports passes_at_delta — whether the joint test rejects the null that every effect is within \(\pm\delta\) of zero.

11.3 Freyaldenhoven-Hansen-Shapiro pre-trend correction

fhs <- didgpu_freyaldenhoven(fit, ...)

Wraps [@freyaldenhoven2019pretrends] for trend-stable confidence sets.

Tip

A worked walkthrough — taking one fit through all three sensitivity-analysis methods and reading the output — is in progress.

11.4 References