The Battelle Memorial Institute began its statistical analysis of unidentified aerial object reports in 1953, under contract to the Air Technical Intelligence Center at Wright-Patterson Air Force Base. The study, designated Project STORK, processed 3,201 reports from the period 1947 to 1952 onto IBM punched cards and subjected them to the standard statistical methods of the era. It was published on 5 May 1955 as Project Blue Book Special Report No. 14, a 312-page document classified For Official Use Only. One hundred copies were printed. The public received a three-page summary that, as an Australian defence analyst later observed, “did not summarise, nor scarcely allude to the 1947-52 data.”
The study’s methodology was straightforward. Each report was evaluated and categorised: balloon, aircraft, astronomical, other, insufficient information, or unknown. The analysts then applied the chi-square test, a standard statistical procedure that measures the probability that two distributions come from the same population. The question was simple: were the unknowns just poorly observed knowns, or were they something different?
The results were unambiguous. The analysts tested the distributions of knowns versus unknowns across six observable characteristics: colour, number of objects, shape, duration of observation, speed, and light brightness. In five of the six tests, the unknowns were statistically distinct from the knowns at better than the one percent significance level. For colour, the chi-square value was 26.15 against a critical threshold of 20.1. For number of objects, 40.73 against 13.3. For shape, 29.05 against 16.8. For duration, 49.49 against 18.5. For speed, 37.93 against 15.1. Only light brightness failed to reach significance, and the analysts noted that the classifications for that characteristic were “too nebulous to be of real value.”
The speed data contained a specific finding the report could not explain away. The major contribution to the chi-square value came from “a large excess of unknowns in the over 400-mph class.” The report acknowledged that “some radar sightings, which are practically impossible to identify, show objects with speeds of 1,000 to 2,000 mph and over.”
The duration data contained another. The report stated, in its own words, that “the greater proportion of unknowns in the 31-to-60-second and 61-second-to-5-minute groups cannot be explained.”
The analysts attempted a revision, removing astronomical identifications from the known category on the grounds that astronomical objects were easy to identify and unlikely to appear as unknowns. Even after this adjustment, the number-of-objects characteristic still showed significance at the one percent level, driven by a large excess of unknowns involving eleven or more objects.
The report’s conclusion, printed in the summary that received public distribution, read: “It is considered to be highly improbable that reports of unidentified aerial objects examined in this study represent observations of technological developments outside of the range of present-day scientific knowledge.”
The data in the body of the report said otherwise. The statistical tests demonstrated, by the study’s own methodology, that the unknowns were not drawn from the same population as the knowns. They behaved differently in colour, number, shape, duration, and speed. The probability that the two populations were the same was, in five of six tests, less than one in a hundred.
Fifteen years later, an Australian Department of Defence analyst, writing an intelligence assessment of the US UFO programme, read Special Report No. 14 with the eyes of someone outside the American institutional framework. The Australian paper identified a finding the American report had buried.
The Battelle analysts had rated each sighting for reliability, based on the observer’s qualifications and the completeness of the report. The reliability categories ranged from “poor” to “excellent.” The Australian analyst extracted the data and presented it as a table:
For reports rated poor in reliability, 16.6 percent were classified as unknown. For reports rated doubtful, 13.0 percent. For reports rated good, 24.8 percent. For reports rated excellent, 33.3 percent.
The percentage of unknowns increased as the quality of the observer improved. Astronomers, radar operators, pilots, and scientists with instrumented data and accurately edited accounts produced the highest proportion of cases that could not be identified. The official position, repeated across decades, was that better data would reduce the number of unknowns. The government’s own statistical study showed the opposite.
The Australian analyst calculated the combined chi-square odds. “The probability was less than one in ten to the twenty-eighth power,” the paper stated, “that is, using the American system, the odds were ten thousand trillion trillion to one against the unknowns being the same as the knowns.” Even after the astronomical revision, “the odds were reduced to two trillion trillion to one.” The Battelle analysts, the Australian noted, had “irrationally considered the results to be inconclusive.”
The numbers were not ambiguous. The conclusion placed on top of them was.
