The nuclear-UFO correlation is the single most cited empirical pattern in UFO/UAP research. Robert Hastings spent 40+ years interviewing more than 150 military veterans about UAP encounters at nuclear weapons sites. Robert Salas testified under oath about missiles going offline at Malmstrom AFB during a 1967 UFO event. French officials have documented incidents at their nuclear facilities. FOIA-released Air Force, FBI, and CIA documents establish a pattern of UFO activity at American nuclear weapons sites extending back to December 1948.
And yet: neither side of the debate has ever conducted a proper controlled comparison study.
Both sides are arguing from anecdotes, case studies, and crude correlations. Neither has produced a study that controls for the obvious confounders: observation density, population, flight operations, temporal trends, and the Cold War's simultaneous peak of nuclear activity AND UFO cultural interest. This is the single most useful study that could be done in the field.
Even setting aside any extraterrestrial hypothesis, this study would have practical national security value:
The ideal design borrows from environmental epidemiology — specifically the methodology used to study disease clusters near industrial facilities, where the same confounders (population density, observation capacity, reporting behavior) apply.
Unit of analysis: All US military installations, categorized into treatment (nuclear) and control (non-nuclear) groups.
| Category | Examples | Count (approx.) | Data Source |
|---|---|---|---|
| ICBM Bases | Malmstrom, Minot, F.E. Warren | 3 active | DOD public records |
| Nuclear Submarine Ports | Kings Bay, Bangor | 2 active | DOD public records |
| Nuclear Weapons Storage | Various (classified locations) | ~20 suspected | FAS Nuclear Notebook |
| DOE Nuclear Labs | Los Alamos, Sandia, Oak Ridge, Hanford | ~17 | DOE public data |
| Commercial Nuclear Plants | 94 operating reactors at ~55 sites | 55 sites | NRC Facility Locator |
| Decommissioned Nuclear | Historical weapons and power sites | ~40+ | NRC + DOE archives |
| Non-Nuclear Military (controls) | Army posts, non-nuclear AF bases, training ranges | ~400+ | DOD Installation Directory |
The critical weakness of every existing study is inadequate confounder control. A rigorous study must match or adjust for:
| Confounder | Why It Matters | Data Available? | Method |
|---|---|---|---|
| Observation Density | Nuclear sites have more security personnel, radar, cameras, restricted airspace monitoring | Partial | Proxy: base personnel count, security clearance density |
| Population Density | More people = more potential reporters (or: RAND found INVERSE relationship) | Yes | Census data, exact match or PSM |
| Flight Operations Tempo | More aircraft = more potential misidentifications | Partial | FAA flight data, MOA designation, sortie counts |
| Sky View / Light Pollution | Kirkpatrick/Utah found dark skies strongly predict UAP reports | Yes | VIIRS satellite light maps, tree canopy data |
| Reporting Culture | Military culture may encourage or discourage anomaly reporting differently at nuclear vs. non-nuclear installations | No | Would require survey data; major limitation |
| Temporal Confounds | Cold War = peak nuclear + peak UFO cultural interest simultaneously | Yes | Time fixed effects, difference-in-differences |
| Geographic Region | Southwest US has both nuclear facilities and clear skies | Yes | Region fixed effects or geographic matching |
| Classified Programs | Secret aircraft testing near nuclear sites creates genuine UAP reports of real objects | No | Cannot control; acknowledge as limitation |
Match each nuclear facility to 1-3 non-nuclear military installations with similar propensity scores calculated from: personnel count, geographic region, population within 30km radius, light pollution level, proximity to MOAs, and base age.
Advantage: Creates balanced comparison groups without parametric assumptions about confounders.
Limitation: Can only match on observables. Unmeasured confounders (reporting culture, classified programs) remain uncontrolled.
Precedent: Rosenbaum & Rubin (1983) foundational method; extensively used in environmental health studies of disease clusters near industrial facilities.
Exploit temporal variation: compare UAP reporting rates at military facilities before and after they gained or lost nuclear status. Example: bases that received nuclear weapons in the 1950s-60s vs. the same bases before nuclear deployment.
Advantage: Controls for all time-invariant unobserved confounders (geography, terrain, climate). Most powerful design if data exists.
Limitation: Requires knowing exact dates of nuclear status changes. Parallel trends assumption must hold.
Key test: Also examine bases that lost nuclear weapons after Cold War drawdowns — if UAP rates dropped, that's strong evidence.
Model UAP report counts as drawn from a negative binomial distribution (overdispersed count data, as used by Bruehl & Villarroel), with facility-level random effects and informative priors based on existing estimates.
Advantage: Naturally handles uncertainty, incorporates prior knowledge, produces posterior probability of nuclear effect. Can include Bayesian sensitivity analysis for unmeasured confounders.
What it would produce: A posterior distribution over the "nuclear effect size" — the probability that being nuclear increases UAP reporting by X%, accounting for all measured confounders and explicit assumptions about unmeasured ones.
RAND used this approach: calculate UAP hotspot statistics per census place, then test whether nuclear facility proximity is a significant predictor after controlling for other spatial variables.
Advantage: Identifies spatial clusters and tests whether nuclear sites explain clustering beyond what population, sky conditions, and military activity predict.
Already done (partially): RAND (2023) found MOA proximity significant; Kirkpatrick/Utah (2023) found dark skies significant. Neither specifically tested nuclear status as a variable.
Critical design choice: Must distinguish between three tiers of reports:
| Tier | Definition | Expected Outcome |
|---|---|---|
| Tier 1 | Any anomaly report (raw NUFORC/MUFON submissions) | Likely inflated by observation density — skeptics expect this correlates with military presence generally |
| Tier 2 | Investigated reports that remain unexplained (Blue Book "unknowns," AARO "truly anomalous") | Smaller N, but removes misidentifications — the meaningful test |
| Tier 3 | Multi-sensor confirmed anomalous events (radar + visual + IR) | Smallest N but highest evidence quality — mostly classified |
The study should run parallel analyses at all three tiers. If the nuclear effect disappears at Tier 2/3, that's evidence for reporting bias. If it persists or strengthens, that's evidence for a genuine phenomenon.
A rigorous study lives or dies on data quality. Here is an honest assessment of every available dataset.
Used by: RAND (2023), Kirkpatrick/Utah (2023), multiple academic studies. Available: nuforc.org, Kaggle, HuggingFace.
Available: archives.gov/research/catalog/catalog-bulk-downloads/uap-bulk-download, Fold3.com
Available (limited): aaro.mil/UAP-Records, EFOIA Reading Room. Bulk data: classified.
Access: mufon.com (paid membership). CMS search at mufon.com/search_database.
Available: CUFOS (cufos.org). Used by Johnson for the CUFON nuclear study.
Published: Nature Scientific Reports (Oct 2025), PASP (Oct 2025). Data: VASCO project archives.
This is the strongest data component. Nuclear facility locations, types, and operational dates are well-documented public information. The NRC provides machine-readable data including reactor status reports updated daily.
Available: nrc.gov/info-finder, nrc.gov/data, DOE national lab sites, fas.org/issues/nuclear-weapons
Available: cia.gov/readingroom/collection/ufos-fact-or-fiction, documentcloud.org, ufohastings.com
GEIPAN data was used in the 2015 French economists' study that found p=0.00013 nuclear correlation. A bilateral US-France comparison study would be powerful, as France has different military culture, geography, and population distribution but shares the nuclear variable.
Available: cnes.fr/en/projects/geipan. Used by: French economists (2015 study).
The fundamental problem: The best UAP data (multi-sensor, investigated, classified) is held by AARO and cannot be accessed by independent researchers. The best accessible data (NUFORC) is self-reported and unverified. Every study design must navigate this gap.
| What We Have | What We Need But Don't Have |
|---|---|
| Precise nuclear facility locations and dates | Precise military installation sensor inventories |
| 190,000+ civilian UAP reports with coordinates | Military UAP reports with coordinates (mostly classified) |
| Population and geographic data per installation | Security personnel counts per installation |
| MOA boundaries and airport locations | Actual flight operations tempo per base |
| Light pollution and canopy cover maps | Base-level reporting culture data (no survey exists) |
| Historical nuclear test dates and yields | Classified program test schedules near nuclear sites |
The most rigorous statistical examination to date of the nuclear-UAP link, published in Nature Scientific Reports in October 2025. Led by Beatriz Villarroel (Stockholm University/Nordita) and Benjamin Bruehl.
The study examined optical transients — short-lived flashes of light captured on astronomical photographic plates from the Palomar Observatory Sky Survey (POSS-I), taken between November 1949 and April 1957. These are NOT traditional UFO sighting reports. They are anomalous objects appearing on astronomical plates that were not stars, known satellites (none existed before Sputnik in October 1957), or catalogued celestial objects.
The research team constructed a daily dataset combining three data streams:
| Test | Purpose | Result |
|---|---|---|
| Chi-square | Whether transients are more likely on nuclear test dates | p = 0.008 (significant) |
| Relative Risk Ratio | How much more likely transients are near nuclear tests | RR = 1.45 (95% CI: 1.10–1.90) |
| Mann-Whitney U | Whether UAP reports differ inside/outside nuclear windows | U = 447,057, p = 0.008 |
| Negative Binomial Regression (GLM) | Effect of UAP count on transient count per date | +8.5% transients per additional UAP (p = 0.015) |
| Earth Shadow Analysis | Whether transients occur less within Earth's shadow (suggesting orbital reflective objects) | 22σ deficit inside shadow (highly significant) |
Defined as ±1 day of a known above-ground nuclear test. On dates within this window, transients appeared on 15.6% of days, compared to 10.8% outside the window.
Published simultaneously in Publications of the Astronomical Society of the Pacific, this paper examined the internal structure of transients on individual plates. It found "aligned multiple-transient events" — several point-like flashes appearing in a line within a single exposure — consistent with specular reflections (sun glints) from objects in near-Earth orbit. One plate showed nine aligned transients. This supports the interpretation that some transients were flat, reflective objects orbiting Earth before the satellite era.
The VASCO finding is the most statistically rigorous nuclear-UAP correlation published to date, but it measures something different from what the nuclear-UFO hypothesis claims. It shows optical transients on 1950s astronomical plates correlate with nuclear tests. Whether those transients are UAP, atmospheric debris, plate artifacts, or something else entirely remains unresolved. The 45% figure is real but its interpretation is contested. Notably, this is a temporal correlation (nuclear tests and plate anomalies on the same days), not a spatial correlation (sightings near nuclear facilities).
The Yingling, Yingling & Bell studies (2023 and 2024 follow-up) across 14 disciplines at 144 major US research universities revealed a striking paradox: curiosity exceeded skepticism in every field, yet fewer than 1% had conducted any UAP research. Faculty reported anxiety about:
The researchers identified competitive research grants as the single most important factor that would unlock faculty participation.
The Catch-22: The most useful data for a controlled comparison (military UAP reports with base-level metadata, sensor data, investigation outcomes) is held by AARO and classified. The only data available to independent researchers (NUFORC civilian reports) is the least suited for the study. Anyone who could do the study can't access the data. Anyone who has the data won't do the study.
A nuclear-UAP study faces a unique credibility problem regardless of outcome:
Skeptics will argue the researchers had an agenda. Any advocate-affiliated institution (SCU, Sol Foundation) lacks perceived independence. Any government institution (AARO) faces accusations of cherry-picking.
Advocates will argue the unclassified data was insufficient, the best evidence is hidden, or confounders were over-controlled. Government studies will be seen as cover-ups.
This is why pre-registration, open methods, and institutional independence are non-negotiable. The study must be designed so that both sides would accept the result before they know what it is.
| Development | Year | Why It Helps |
|---|---|---|
| NASA UAP Panel | 2023 | Explicitly called for stigma reduction; NASA appointed Director of UAP Research |
| Galileo Project Observatories | 2021–2026 | Three instrumented observatories, 500K objects catalogued, methodology proven |
| Sol Foundation (Stanford) | 2023 | Academic credibility via Garry Nolan; annual symposia with peer-reviewed white papers |
| Schumer-Rounds UAPDA | 2023–2026 | Bipartisan Congressional pressure for records disclosure; reintroduced 2026 |
| VASCO papers in Nature/PASP | 2025 | Proves nuclear-UAP correlation can pass peer review at mainstream journals |
| NJ State UAP Research Bill | 2025–2026 | First potential state-funded UAP research institution |
What survives: If the correlation persists after controlling for observation density, population, flight ops, sky conditions, and temporal trends, the following explanations remain viable:
| Explanation | Testable? | What Would Confirm It |
|---|---|---|
| Foreign surveillance — Adversary nations target nuclear sites with advanced reconnaissance platforms | Yes | Classified analysis would show UAP correlating with known foreign intelligence activity. National security implications would likely prevent publication. |
| Domestic classified programs — US tests advanced technology near its own nuclear sites | Partially | Would show up when AARO resolves cases. Kirkpatrick's 2024 report attributes some cases to this. 2025 Pentagon report revealed Malmstrom incident was EMP test. |
| Non-human intelligence — Something is monitoring human nuclear capability | Not directly | Would require ruling out all above explanations AND demonstrating technology beyond known capabilities. A correlation study alone cannot establish this. |
| Environmental / physical — Nuclear facilities produce electromagnetic or thermal signatures that create optical phenomena | Yes | Would predict correlation with power plants (which produce more continuous radiation/heat) at least as strong as weapons sites. |
This would be the most interesting result. It eliminates the "nuclear radiation/energy creates optical phenomena" explanation (power plants produce far more continuous energy output than dormant weapons). It points toward something about security sensitivity, classified activity, or deliberate monitoring rather than physics.
This would weaken the "classified surveillance" explanation (why would adversaries heavily surveil civilian power plants?) and strengthen either physical/environmental explanations or the hypothesis that something is monitoring nuclear technology broadly.
What this tells us: The entire nuclear-UFO narrative — one of the most enduring patterns in ufology — is a product of reporting bias, observation density, and the Cold War's simultaneous inflation of both nuclear activity and UFO cultural interest.
This outcome would be valuable precisely because it clears away decades of pattern-matching noise and refocuses the field.
Most likely outcome. Real data is messy. Several partial-result scenarios are plausible:
Also quite likely. The study might simply demonstrate that available unclassified data cannot answer the question. This itself would be an important finding:
| Institution | Data Access | Scientific Credibility | Independence | Funding | Overall Feasibility |
|---|---|---|---|---|---|
| AARO | High | Medium | Low | High | Medium |
| Galileo Project | Low | High | High | Medium | Medium |
| RAND Corporation | Medium | High | High | High | High |
| SCU | Low | Medium | Medium | Low | Low |
| Sol Foundation | Medium | High | Medium | Medium | Medium |
| University Partnership | Low | High | High | Low | Low |
| NJ State UAP Center (proposed) | Low | Medium | High | Medium | Medium |
Verdict: AARO is the only entity that COULD answer the question with classified data, but unlikely to produce a result that either side trusts. Best used as a data partner, not the lead research institution.
Verdict: Best positioned for a prospective study (deploy instruments near nuclear vs. non-nuclear sites and compare). Less suited for the historical retrospective study. Could be transformative if combined with AARO data access.
Verdict: Strongest candidate for the publicly accessible version of this study. Has methodology, credibility, and DoD relationship. Key question: would they include nuclear status as a variable in a follow-up study? Their 2023 omission is notable.
Verdict: Best positioned to do the "version 0.5" study with public data (NUFORC + NRC facilities) that establishes the methodology and identifies what classified data would be needed. Would benefit from university partnership for credibility.
Lead: University biostatistics or epidemiology department (has matched-comparison methodology expertise, no UAP advocacy perception)
Data partner: AARO (provides base-level reporting data under academic data use agreement, retains classification authority over specific cases)
Methodology consultant: RAND (already built the spatial regression framework; can advise on confounder specification)
Independent audit: Pre-registered on OSF; analysis code open-sourced; replication dataset for unclassified portion
Funding: NSF or NASA (neither has funded UAP research before, but NASA's 2023 panel recommended it; Congressional pressure building)
Publication target: PNAS, Nature Human Behaviour, or JAMA-style epidemiology journal (not a UAP-specific venue)
Estimated cost: $200K–$500K (small by research grant standards; this is a data analysis study, not a new data collection effort)
Several studies have touched on pieces of this question. None has done the full controlled comparison.
| Study | Year | Finding | N | Controls | Limitations |
|---|---|---|---|---|---|
| Johnson (CUFON) | ~2004 | RR = 1.44 nuclear vs. non-nuclear counties | 328 counties | Population, region | No control for military presence, observation density, sky conditions, or temporal trends |
| French Economists | 2015 | p = 0.00013 nuclear-UFO correlation in France | GEIPAN data | Unknown | Full methodology not widely published in English; confounder control unclear |
| RAND | 2023 | 1.2x rate within 30km of MOAs | 101,151 | Population, airports, weather stations, light pollution | Did not test nuclear status specifically; civilian data only |
| Kirkpatrick / Utah | 2023 | Dark skies + wide spaces predict UAP reports | 98,000+ | Light pollution, canopy, cloud cover, airports, military | Did not test nuclear status specifically; aggregate military variable only |
| VASCO (Bruehl & Villarroel) | 2025 | RR = 1.45 transients near nuclear test dates | 2,718 days | Pre-satellite era (no orbital debris confound) | Measures plate anomalies not UAP sightings; temporal not spatial; 1950s only; arXiv rejected |
The gap is clear: No study has combined (1) a nuclear-specific variable, (2) matched military controls, (3) multi-variate confounder adjustment, (4) multiple UAP report tiers, and (5) pre-registered methods. That study would resolve a 75-year-old question.