AN-013: Buyer-by-class placebo battery¶
Economic intuition
The original placebo asks whether never-litigated items show the same urgent-price pattern. This version makes the placebo harder to pass: it keeps only never-litigated items bought by the same buyers in the same item classes where litigated purchases also appear elsewhere in the panel. The result stays flat. The urgent pattern does not reappear even in overlapping procurement environments where a generic buyer, class, or platform story would be more plausible.
Question¶
Does the never-litigated placebo remain null when restricted to procurement environments that overlap with litigated procurement?
Design¶
- Sample: never-litigated items with both ordinary and administrative urgent observations. The matched subset keeps buyer-by-class cells that also contain litigated purchases elsewhere in the panel.
- Specification: item, year, and PBU fixed effects, with standard errors clustered by PBU.
- Outcomes: log negotiated price, log reference price, and log number of bidders.
- Role: specificity check, not a stand-alone identification design.
Results¶
| Placebo specification | Coefficient | SE |
|---|---|---|
| All never-litigated negotiated price | -0.020 | 0.032 |
| Matched buyer-by-class negotiated price | -0.046 | 0.053 |
| Matched negotiated price + quantity control | -0.020 | 0.070 |
| Matched reference price | -0.071 | 0.056 |
| Matched number of bidders | -0.032 | 0.043 |
The matched placebo contains 763 never-litigated items across 99 buyer-by-class
cells. Output:
v10-causal-mechanism/output/tables/tab_h7_placebo_battery.tex.
Interpretation¶
Confidence: green for the stated placebo question. The tighter placebo returns economically small, statistically null coefficients across negotiated price, quantity-controlled negotiated price, reference price, and bidder count inside buyer-by-class environments where litigated purchases also appear. This substantially weakens generic platform, buyer-class, and time-trend explanations for the urgent-procurement pattern. The dynamic evidence remains diagnostic, but this placebo battery is strong for the specificity claim it is designed to test.
Follow-ups¶
- Combine this placebo battery with classifier-threshold sensitivity if purchase-regime confidence scores are exposed in a future public replication file.
- The event-study path should remain diagnostic unless a future design supplies stronger pre-period robustness.