NUFORC UFOs
79,638 reports. 7 layers of confound strippingSystematically removing alternative explanations (social contagion, media effects, population density, instrument changes) until only the genuine signal remains.. Here's what survived.
Decision Chain: NUFORC
Ingest all NUFORC reports with minute-level timestamps. No filtering, no cleaning — everything goes in.
79,638 reports ingested. 79% of timestamps round to nearest 5 minutes.
Fit a self-exciting point process to the report timestamps. Separates "background" (independent) events from "triggered" (contagion) events.
Branching ratio 0.959. 96% of reports are triggered by prior reports. Half-life 0.6 days.
Control for when people could actually see the sky: darkness hours, cloud cover, population awake.
Observer availability explains 65.7% of the temporal pattern. 9pm peak SURVIVES.
Per-capita adjustment removes population density as a confound. Ensures rural areas aren't underweighted.
County-level normalization applied. Western US concentration SURVIVES.
Google Trends proxy for media-driven reporting waves. Regresses out spikes caused by news coverage.
Summer peak STRIPPED. Weekend effect STRIPPED. 2014 spike STRIPPED.
Detect structural breaks from changes in the reporting instrument itself.
3 changepoints detected: 1995 website launch, 2004 form redesign, 2012 Starlink contamination. Exponential growth STRIPPED.
Extract 116 features from report text. Compare distributions between independent (genuine) and triggered (contagion) groups with FDR correction.
66 of 116 features significantly different (FDR < 0.05). Independent reports enriched for disks, cigars, silver. Triggered enriched for lights, fireballs, orange.
Two populations, not one
The 4% of NUFORC reports that survive contagion stripping describe fundamentally different things than the 96% that don't. Independent observers report structured objects (disks, cigars) in metallic colors with measured language. Triggered reporters describe lights and fireballs in excited language that matches cultural templates 28% more closely. The independent group's shape distribution is stable across decades. Whatever these people saw, they didn't get the idea from each other.