Engineering, production, governance, cost, communication · insight-1291
Look for bunching just past a KPI target to catch gaming
- Evidence type
- practitioner heuristic
- Difficulty
- hard
- Roles
- Data analyst, Data scientist
- Industries
- cross-industry, telecom, public-sector
Summary: situation, action, reason, boundary
Situation
A branch bonus needs 95 percent of tickets resolved within the service target, and many branches report exactly 95 to 96 percent (hypothetical).
Action
Plot the unit-level metric around the cutoff for each reporting unit, test for a jump in density at the threshold, compare with the period before the target existed, and inspect records near the line before rewarding or acting on the metric.
Reason
When agents can manipulate a variable around a cutoff, its density is discontinuous there, which a density-continuity test is designed to detect.
Boundary
Bonus thresholds, SLA targets, regulatory limits and eligibility cutoffs measured per unit.
Symptom or risk
Values pile up just past the cutoff, which suggests reclassification, timing or selective closing rather than better service.
Exceptions
Round numbers and capacity limits also cause clustering; the pre-target comparison separates them.
Trade-offs
Needs unit-level data and enough units; a jump is a reason to investigate, not proof of intent.
Pitfalls
Testing only the aggregate mean, where gaming in both directions cancels.
Lessons and concepts
Claims and their sources
McCrary proposes testing continuity of the running variable's density at a cutoff, because manipulation of that variable can invalidate identification.