---
title: "Benkard's (2000) headline 0.96 depreciation figure is the spillover-adjusted specification, not the simplest one"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://web.stanford.edu/~lanierb/research/Learning_and_Forgetting_AER.pdf"
source_author: "C. Lanier Benkard, 'Learning and Forgetting: The Dynamics of Aircraft Production' (American Economic Review 90(4):1034-1054)"
source_date: 2000
source_quote: "The estimate of δ rises significantly when the model is extended to allow for incomplete spillovers... Leaving out incomplete spillovers from the model thus leads to a downward bias in the experience depreciation parameter."
source_tier: 1
audit_status: "verified-verbatim — read directly from the author's own Stanford faculty-page PDF, tls: verified, OCR-extracted"
provenance: "Promotion from 10-inbox/raw/2026-07-24-does-benkard-2000-report-the-organizational-forgetting-depreciation.md, 2026-07-24"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-24-does-benkard-2000-report-the-organizational-forgetting-depreciation.md"
date_created: "2026-07-24T00:00:00.000Z"
tags: ["economics","learning-curve","organizational-forgetting","aircraft-manufacturing","benkard","model-specification"]
---


Benkard's Table 3 reports two general-learning-model regressions relevant to the [[claim-organizational-forgetting-can-reverse-the-learning-curve|0.96/61% depreciation figure]]. Regression 9 ("OF only," [[entity-organizational-forgetting|organizational forgetting]] with complete spillovers assumed between [[entity-lockheed-l-1011-tristar|L-1011]] model variants, λ=1) estimates δ = 0.952. Regression 10, which adds an incomplete-spillovers parameter estimated at λ≈0.70, estimates the higher δ = 0.960 — the figure Benkard uses for his 61%-annual-survival interpretation. He explains the direction of the shift: "The estimate of δ rises significantly when the model is extended to allow for incomplete spillovers... Leaving out incomplete spillovers from the model thus leads to a downward bias in the experience depreciation parameter."

The headline number is therefore specifically tied to a modeling choice — treating knowledge transfer between the L-1011's -1/-100/-200 and -500 variants as partial rather than complete — and not to the plainest version of the organizational-forgetting specification. The difference (0.952 vs. 0.960) is modest in absolute terms but is the mechanism by which the "which regression is being cited" question changes the answer.

> [!note] Seek's commentary:
> A 0.008 gap between two rows of the same table isn't the kind of thing a summary survives intact — it's exactly the kind of thing that gets rounded away between the paper and the paraphrase, and rounded-away is not the same as wrong, but it is the seam where a wrong number would hide if one were hiding. I'm noting which regression the vault's headline figure actually comes from now, so that if a future capture finds someone citing 0.952 instead of 0.960 for the "same" claim, it's a known fork in the road and not a contradiction to chase. — Seek