The bound carries \(1/g_{\min}\), so it is
\(2R/(n g_{\min})\) and NOT \(O(1/n)\). Truncating the
propensity score is part of the privacy guarantee.
Usage
morie_ate_sensitivity(n, g_min, y_range = 1)
Arguments
- n
Number of observations.
- g_min
Propensity truncation bound.
- y_range
Range of the outcome.
Value
A list with the sensitivity and a few related quantities.
Examples
morie_ate_sensitivity(n = 1000, g_min = 0.05)$sensitivity # 2 / (1000 * 0.05)
#> [1] 0.04
