Common-support and positivity diagnostic — morie_causal_overlap

The propensity overlap between arms: the common support interval, the fraction of units outside it, the fraction with extreme scores, and the overlap coefficient \(\sum_b \min(p_{1b}, p_{0b})\).

Usage

morie_causal_overlap_diagnostic(ps, treat, bins = 20, eps = 0.05)

Arguments

ps: propensity scores.
treat: 0/1 treatment indicator.
bins: histogram bins on \([0, 1]\).
eps: threshold defining an extreme score.

Value

list with common_support, n_outside, prop_extreme, overlap_coefficient, hist_treated, hist_control.

Details

Positivity is an assumption about the POPULATION and cannot be verified from a sample. An empty region may be structurally impossible or merely unobserved, and no diagnostic distinguishes those. What this function can do is show you that the estimator is extrapolating; it will produce a number either way.

References

Crump, R. K. et al. (2009). Dealing with limited overlap in estimation of average treatment effects. Biometrika, 96(1), 187-199.

Examples

set.seed(1)
ps <- plogis(rnorm(300))
morie_causal_overlap_diagnostic(ps, rbinom(300, 1, ps))$overlap_coefficient
#> [1] 0.633317