# Feeder Networks as a Proxy for the Experience Curve: Non-Fatal Reportable-Event Feasibility — Research Memo **Date:** June 2026 **Status:** EXPLORATORY feasibility research. This memo informs the design of a *separate, fresh* pre-registered FOQA/ASAP precursor non-inferiority study. It does **not** re-open, re-score, or revise FRISA's locked net-ledger verdict on the 1,500-hour rule ("sign indeterminate, magnitude small, modest adverse lean"). Every quantitative claim carries an inline source; figures that cannot be verified against primary data are flagged **[UNVERIFIED]** or **[NOT PUBLICLY LINKABLE]**. --- ## Executive Summary **Is the feeder network a suitable proxy for studying the pilot experience curve? — QUALIFIED YES, but only for a *within-regional* design, not a regional-vs-major contrast.** US regional/feeder carriers (SkyWest, Republic, Endeavor, PSA, Envoy, Mesa, etc.) are genuinely the population where the least-experienced Part 121 first officers concentrate: majors hire seasoned pilots out of the regionals, so the experience floor effect, if it exists, is observable here and essentially nowhere else in Part 121 ([GAO-26-107856, 2026](https://www.gao.gov/products/gao-26-107856)). That part of the proxy logic is sound. The serious problem is **low-tail compression**. Post-2013, the ATP/R-ATP floor truncated exactly the low end of the experience distribution the study most wants to observe. The residual variation is real but narrow: R-ATP 750/1,000/1,250-hr entrants vs 1,500+-hr entrants, new-hire vs seasoned FOs, and time-in-type. Tellingly, in the most recent fatal regional event (PSA/American Eagle 5342, Jan 2025) the first officer had ~2,469 hr and the captain ~3,950 hr — **both far above the floor** ([2025 Potomac River mid-air collision](https://en.wikipedia.org/wiki/2025_Potomac_River_mid-air_collision)). The "regional vs major" contrast is **hopelessly confounded** (different fleets, route structures, exposure, pay, fatigue, scheduling). A *within-regional, experience-stratified* contrast is defensible; a *between-segment* contrast is not. **Is non-fatal reportable-event frequency high enough to give statistical power the fatal endpoint lacks? — YES, by orders of magnitude — but only if the right (mandatory or exposure-normalized) endpoints are used.** The fatal endpoint is near-zero: ~2 fatal Part 121 accidents in the 15 years 2009–2024 ([NTSB](https://www.ntsb.gov/safety/data/Pages/Part121AccidentSurvivability.aspx)), making the rule's marginal effect statistically unmeasurable — FRISA's stated limitation. By contrast, non-fatal precursors are abundant: ~29 Part 121 accidents/yr; ASRS received **106,087 reports in 2023, ~52% from air carrier flight crew** ([NASA ASRS briefing, 2024](https://ntrs.nasa.gov/api/citations/20240014226/downloads/ICASS%202024%20ASRS.pdf)); unstable approaches are common with only ~3% resulting in a go-around, go-arounds ~**1–3 per 1,000 approaches** ([Flight Safety Foundation Go-Around Study](https://flightsafety.org/wp-content/uploads/2017/03/Go-around-study_final.pdf)). Regionals supply a disproportionate share: **regional airlines are >40% of ASIAS participants and provide ~50% of the data analyzed** ([DOT IG ASIAS report, 2021](https://www.oig.dot.gov/sites/default/files/FAA%20ASIAS%20Final%20Report%20-%2003.10.2021.pdf)). **The catch:** the high-volume endpoints (ASAP, ASRS) are **voluntary** and therefore reporting-culture-confounded in a *direction that can flip the sign* — a better safety culture reports MORE and looks "worse." A valid design must lean on **FOQA exceedances and mandatory NTSB Part 830 / FAA AIDS incidents** (objective, exposure-linkable), reserve voluntary reports for sensitivity analysis with an explicit reporting-rate adjustment, normalize per departure/cycle (not per hour), and stratify **within** the regional population. **Bottom line:** The feeder network is the correct *population and denominator*. The non-fatal event count is more than sufficient for power. The proxy is suitable *if and only if* the design abandons the regional-vs-major framing for within-regional experience stratification on mandatory/objective endpoints. The binding constraint is not event scarcity — it is **(a) compressed experience variation** and **(b) the public unavailability of event-level pilot-experience linkage** (see §4). --- ## 1. Proxy Suitability **Concentration — confirmed.** Majors backfilled vacancies by hiring *from* regionals; regionals were capacity-constrained by insufficient supply of "both captains and new first officers" ([GAO-26-107856](https://www.gao.gov/products/gao-26-107856)). The least-experienced Part 121 FOs enter at regionals at/near the floor; the most experienced migrate up. If an experience-floor effect exists in Part 121, the feeder network is where it would be visible. **Strongest pillar of the proxy logic.** **Residual variation — real but narrow (low-tail compression).** The R-ATP carve-outs create a discrete low-end gradient: **750 hr** (military), **1,000 hr** (bachelor's aviation), **1,250 hr** (associate's), vs **1,500 hr** standard. New-hire vs seasoned FO, and **time-in-type** — a regulatorily meaningful axis: under **14 CFR 121.438**, a <100-hr-in-type SIC triggers PIC takeoff/landing requirements at special airports and in low-visibility/contaminated conditions ([eCFR 121.438](https://www.law.cornell.edu/cfr/text/14/121.438)). Both a usable stratum and a built-in confounder (§3). **Compression is severe.** PSA 5342's FO (~2,469 hr) and the Colgan crew (captain 3,379 hr, FO 2,244 hr) all *exceeded* the floor — the natural experiment is weak in the most policy-relevant region. **Regional-vs-major is not usable**; only within-regional stratification is defensible. **Deeper caution.** The Pilot Source Study series found *flight time alone does not predict regional training success* — instructor experience, accredited collegiate training, fewer prior failures, and younger age were stronger ([Pilot Source Study](https://www.pilotsourcestudy.org/)). Raw hours is a noisy proxy for "experience"; treat it as one ordinal axis among several. --- ## 2. Non-Fatal Reportable-Event Frequency Fatal-accident rate 2001–2017 ≈ **0.006 per 100,000 flight hours** (~1 per 16.3M hr) — unmeasurable for marginal effects. Non-fatal endpoints by tier: | Endpoint | Tier | Approx. volume | Source | |---|---|---|---| | Part 121 accidents (mostly non-fatal) | Mandatory (NTSB) | ~29/yr; rate ~0.18/100k hr | [NTSB](https://www.ntsb.gov/safety/data/Pages/Part121AccidentSurvivability.aspx) | | NTSB Part 830 reportable incidents | Mandatory | > accident count; not cleanly tabulated **[UNVERIFIED count]** | NTSB | | FAA AIDS incidents | Mandatory/objective | ~200,000 A&I records 1973–present | [FAA AIDS/ASIAS](https://www.asias.faa.gov/) | | ASRS total reports | Voluntary | 106,087 in 2023, ~52% air-carrier crew | [NASA ASRS 2024](https://ntrs.nasa.gov/api/citations/20240014226/downloads/ICASS%202024%20ASRS.pdf) | | ASAP reports | Voluntary | 262 operators/767 programs; per-program counts not public **[UNVERIFIED]** | [FAA Report to Congress](https://www.faa.gov/sites/faa.gov/files/2021-11/Report-to-Congress-on-ASAP-and-FOQA.pdf) | | FOQA exceedances | Continuous/objective | unstable approaches common; go-arounds ~1–3/1,000 approaches | [FSF Go-Around Study](https://flightsafety.org/wp-content/uploads/2017/03/Go-around-study_final.pdf) | **Regional share — decisive.** Regionals are >40% of ASIAS participants and ~50% of data analyzed ([DOT IG ASIAS](https://www.oig.dot.gov/sites/default/files/FAA%20ASIAS%20Final%20Report%20-%2003.10.2021.pdf)); they fly 3–6 short legs/day vs 1–2 long legs at majors, generating far more *cycles* (highest-risk phases) per hour. **Power verdict.** A single large regional plausibly generates thousands–tens of thousands of objective exceedance events/year vs a fatal expectation near zero — **~4–6 orders of magnitude** more signal. The study is easily powered on event volume; binding constraints are confounding (§3) and linkage (§4). **[Per-regional FOQA counts are proprietary; the order-of-magnitude inference applies published go-around/unstable-approach rates to regional departure volumes.]** --- ## 3. Proxy-Validity Confounds (rigorous and symmetric) **A. Operational exposure (experience ⟂ exposure).** Regionals fly more cycles, shorter legs, more secondary airports, more weather per hour. Per-hour normalization conflates exposure with experience. **Mitigation: normalize per departure/cycle; condition on airport/weather/fleet.** **B. Reporting-culture confound for VOLUNTARY data — sign-flipping.** ASAP/ASRS counts reflect reporting propensity; a stronger just-culture reports MORE and looks WORSE ([reporting-vs-accidents dynamics](https://www.sciencedirect.com/science/article/abs/pii/S0925753521001958)). Can reverse the experience effect's sign. **Mitigation: voluntary counts not a primary outcome.** **C. Selection confound (experience ⟂ pairing).** 14 CFR 121.438 *forces* low-time-in-type FOs to be paired with the captain flying all takeoffs/landings at hard airports/poor conditions — the target pilots are operationally shielded by rule, attenuating and confounding any experience effect. Base/fleet/route also correlate with seniority. **Mitigation: model captain experience and pairing; treat 121.438 shielding as an effect-modifier.** **D. Post-2013 low-tail compression** (§1) — natural experiment weak near the floor. **Mandatory vs voluntary partition.** Preferred (objective): NTSB Part 830, FAA AIDS, FOQA exceedances (machine-generated). Sensitivity only (propensity-driven): ASAP, ASRS. --- ## 4. Design Implication **Population/denominator — YES.** The feeder network is the correct population and denominator — the only Part 121 setting with meaningful low-experience mass and ~50% of national objective safety data. A valid pre-registered design must: 1. **Contrast = within-regional experience stratification** (R-ATP tier, total hours ordinal, time-in-type bands tied to 121.438, new-hire vs seasoned), carrying captain experience/pairing as covariates — **not** regional-vs-major. 2. **Primary outcome = mandatory/objective events** (FOQA exceedances + NTSB/AIDS incidents); voluntary reports only as pre-specified sensitivity with a reporting-rate adjustment. 3. **Exposure normalization per departure/cycle**, not per hour. 4. **Non-inferiority framing** with a pre-registered margin justified independently of the locked ledger. 5. **Confounder control**: airport/weather/fleet/base effects; captain experience; 121.438 indicator; calendar time (post-2013/post-COVID regime shifts). **The decisive blocker — event-level experience linkage.** Public datasets do **NOT** reliably link individual FOQA exceedances or AIDS/NTSB incidents to the crew's flight-hour totals and R-ATP tier. FOQA is de-identified by design; ASIAS aggregates. **[NOT PUBLICLY LINKABLE]** A valid study almost certainly requires **carrier-side data access** (a regional's own FOQA + crew records, or an ASIAS-mediated protected study). This is the single largest design risk and must be settled *before* pre-registration. --- ## Implications for the FOQA Precursor Pre-Registration 1. Frame it **within-regional**; disclaim regional-vs-major as confounded/out of scope. 2. **Objective primary endpoints** (FOQA exceedances + mandatory NTSB/AIDS); pre-specify exact exceedance definitions/thresholds. 3. **Quarantine voluntary data** to sensitivity analyses with a stated reporting-propensity adjustment; pre-register the sign-flip risk. 4. **Normalize per departure/cycle.** 5. **Resolve linkage before locking** — carrier FOQA+HR join or protected ASIAS study; if unobtainable, the precursor experience-curve test is infeasible and must be re-scoped (weaker ecological analysis) rather than over-claimed. 6. Treat **"hours" as one noisy axis**; pre-register multiple experience operationalizations. 7. **Power is not the constraint; confounding and linkage are** — pre-register those as the substantive commitments. **Scope reminder:** EXPLORATORY feasibility input to a fresh, separate pre-registration. It does not re-open FRISA's locked verdict. What the public data supports: feeders concentrate low-experience FOs and generate abundant objective precursors. What it cannot support without carrier-side access: event-level linkage of those precursors to individual crew experience — the crux on which the whole precursor design turns. --- ### Verification flags - **Well-verified:** NTSB Part 121 counts/rates; ASRS 2023 volume/air-carrier share; regional ASIAS share (DOT IG); GAO pilot-supply; R-ATP thresholds and 121.438 text; PSA 5342 and Colgan crew hours. - **[UNVERIFIED count]:** per-program ASAP volumes; clean public NTSB Part 830 incident time series. - **[NOT PUBLICLY LINKABLE]:** event-level join of FOQA/AIDS/NTSB events to individual crew flight-hour totals and R-ATP tier — the key feasibility blocker.