The per-unit diagonal of the causal curvature of bic_curvature()
(Definition 5.1 of Ospina, 2026), split by whether each unit's
influence runs mainly through the discrete or the continuous component
(Definition 5.2), and screened for disproportionate influence by the
2 * mean(ICIM) rule.
Arguments
- cv
A
"bic_curvature"object frombic_curvature().
Value
A data frame with one row per unit and columns ICIM,
ICIM_discrete, ICIM_continuous, type
("discrete"/"continuous"/"mixed") and influential (logical,
ICIM > 2 * mean(ICIM)).
Details
influential = TRUE flags a unit whose leverage on the fitted model is
more than twice the sample's typical leverage, a screening rule in the
spirit of the 2 * mean(hat value) rule for leverage in linear models.
It is not a claim that the unit has an unusually large individual
treatment effect, and a flagged unit can have a small or even
near-zero estimated effect; leverage and effect size answer different
questions. type says which component a flagged unit's leverage runs
through, which is useful for tracing an unexpected curvature share
back to specific observations (e.g. a handful of households with
extreme covariate profiles on the boundary side) rather than treating
the curvature as an unexplained aggregate number.