Before–After–Elsewhere Evaluation¶

Quasi-experimental evaluation — instantiates Migration-Resistant Hazard Control

Measures the target outcome before and after at the intervention site and — the defining addition — at the places the hazard could have moved to, so a local win cannot pass as reduction until 'elsewhere' clears too.

A Before–After–Elsewhere Evaluation measures the target outcome at the intervention site before and after the control and — the defining addition — measures the same outcome at the places the hazard could have moved to over the same period. Its distinguishing claim in this archetype is that a local improvement cannot count as genuine risk reduction until the "elsewhere" arm shows the hazard did not simply relocate. Where the modelling mechanisms predict displacement and the monitors watch for it, this is the evaluation design that tests for it: same metric, same window, measured with equal rigour at the source and at the candidate destinations, and held open long enough for a lagged migration to appear.

Example¶

A police department floods a six-block hot spot with foot patrols and, three months later, reports burglaries in the zone down by roughly a third. A Before–After–Elsewhere Evaluation refuses to stop there. It measures burglaries in the hot spot before and after — the before/after arms — but also, over the identical window, in the surrounding blocks the offenders could have shifted to, and it counts them with the same rules and effort used inside the zone, so a quieter beat outside cannot be an artefact of looking less hard. Because displacement often lags while offenders find new ground, the evaluation holds the elsewhere measurement open for months rather than weeks. Two patterns become distinguishable only because the "elsewhere" arm exists: if the surrounding blocks held steady, the drop is real reduction; if their burglaries rose by nearly what the hot spot lost, the patrols mostly moved the crime. The same design can catch the happier case — a fall outside the zone — which is a genuine diffusion of benefit, not displacement.

How it works¶

Add the "elsewhere" arm. Measuring candidate destinations, not just the treated site, is the entire point; a before/after with no elsewhere cannot tell reduction from relocation.
Enforce measurement symmetry. The destination is measured with the same metric, definitions, and effort as the source, so displacement cannot hide behind weaker observation.
Hold the window open for the lag. Migration often arrives after a delay; the evaluation runs long enough for a slow reroute to surface rather than closing on the early, flattering read.
Read displacement and diffusion together. A rise elsewhere is displacement; a fall elsewhere is diffusion of benefit — both are visible only with the extra arm.

Tuning parameters¶

Choice of "elsewhere" — which sites serve as the displacement arm. Pick sites the hazard cannot actually move to and the test is rigged to find nothing.
Measurement window — how long after the intervention you keep counting. Too short and a lagged migration is scored as success.
Symmetry of effort — how equally source and destination are observed. Asymmetric effort manufactures a false "no displacement."
Displacement radius — how wide a net of destinations you scan. Too narrow and the hazard slips to an unmeasured site.
Outcome metric and unit — what is counted and at what grain. A metric that only exists inside the treated site cannot detect movement out of it.

When it helps, and when it misleads¶

Its strength is that the "elsewhere" arm is the single feature separating genuine reduction from displacement, and it does so with measurement rather than argument. It also catches the honest good news — diffusion of benefits — that a treated-site-only study cannot see.

Its failure mode is that it is only as good as its comparison. A contaminated or ill-chosen "elsewhere," a window closed before the lagged migration arrives, or displacement to a destination nobody measured will each read as clean success. Its classic misuses are choosing an "elsewhere" that cannot show displacement and stopping the clock early. The discipline is to fix the comparison sites and window before the results are in, to keep observation symmetric, and to scan several plausible destinations — the logic of a Before-After-Control-Impact design, whose whole power comes from the untreated comparison arm.^[1]

How it implements the components¶

displacement_monitor — the "elsewhere" arm is a direct measurement of whether the hazard relocated rather than fell.
measurement_symmetry_rule — the requirement that destination and source be measured with equal rigour is built into the design, not left to goodwill.
migration_lag_window — holding the evaluation open long enough for a delayed migration to appear is one of its explicit parameters.

It does not build the map of where "elsewhere" is — that comes from Causal Loop Diagram and Fault Tree Analysis — nor does it assemble the local-vs-system consequence picture, which is Whole-System Impact Map.

Instantiates: Migration-Resistant Hazard Control — the evaluation is the empirical test that a control reduced the hazard rather than moving it.
Consumes: Causal Loop Diagram and Fault Tree Analysis — the migration field and transfer paths that say which destinations the "elsewhere" arm must measure.
Sibling mechanisms: Whole-System Impact Map · Causal Loop Diagram · Fault Tree Analysis · Mass Balance · Hazard Analysis · Adaptive Circumvention Red Team · Agent-Based Experiment or Simulation · System-Wide Net-Risk Dashboard · Boundary Expansion Review · Cross-Boundary Hazard Ledger · Cross-Jurisdiction Incident Review · Intervention Displacement Stress Test · Migration Sentinel Network · Pressure-Absorption Redesign Workshop · Source-Reduction or Safe-Dissipation Plan

Notes¶

A null result in the "elsewhere" arm only means "no displacement" if the arm was measured widely and long enough; an unmeasured destination or a window closed before the lag reads identically to a true null. Treat a clean elsewhere-result as evidence proportional to how many destinations were scanned and how long the clock ran — and resist over-claiming the favourable case (diffusion of benefits) on the same logic.

References¶

[1] Before-After-Control-Impact (BACI) designs — and their econometric cousin, difference-in-differences — estimate an intervention's effect by comparing the change at the treated site against the change at an untreated comparison over the same period. Adding an "elsewhere" arm turns that comparison into a displacement test: the comparison sites are the places the hazard might have moved to. ↩