Non-visual mirror Northstar Home · no customer data
The Store That May Not Have Caused It
Did a fictional store change total value, shift channel, or coincide with a wider trend?
Narrative sequence
- Catchment hypothesis — Fictional customer locations define a catchment hypothesis, not geographic truth.
- Comparison histories — Stable cohorts align into treated and comparison histories.
- Pre-trend — Only comparisons with credible pre-trends earn support.
- Event study — The before/after gap becomes a difference-in-differences estimate with uncertainty.
- Store heterogeneity — One pooled headline separates into heterogeneous store estimates.
- Sensitivity — Placebo dates, catchment changes and leave-one-out ranges attack the conclusion.
- Decision groups — Stores finish as supported, weak or unsettled.
| Evidence | Value | Status / meaning |
|---|---|---|
| Pre-trend | PASS | selected comparison |
| Pooled estimate | -12 index | scenario |
| Leave-one-out | -5 to -17 | sensitivity |
| Placebo date | 0 to -3 | counter-case |
| Stores supported | 2 | decision |
| Stores unsettled | 3 | decision |
Trivial baseline
Naive before/after comparison: attribute the entire post-opening change to the store.
Sensitivity
Test pre-trends, placebo dates, alternate catchments and leave-one-store-out ranges; report the detectable effect bound.
Strongest counter-case
Stores may open where demand was already changing; placebo dates and alternate comparisons attack the result.
Verdict
DOWNGRADED: pooled estimate is unstable. SURVIVES: two store-specific signals; three remain unsettled.
Synthetic example. Deterministic fictional records; not customer data or achieved customer results.