Special Report No. 14: Visualisation · Signals from the Periphery

← Intro 中文

View

THE SORT

Subject

—

Value

—

Verdict

—

SPECIAL REPORT NO. 14
The day the Air Force counted the saucers
3,201 cases, 1947-1952, on punched cards
21.5% unknown is one of ten honest answers
best reports 33.3% unknown, worst 16.6%
its own test says they are not one population
pick a denominator · run the chi square · read the file

…

…

…

Evidence

fact ⇄ what kind of claim · live

3,201 sightings charted from about 4,000 reports, 1947 to 1952

Figure 1 · the largest study of its kind ever run

MEASURED

Unknown is 689 = 21.5% of all sightings, and 434 = 19.7% of object sightings

Figures 2 and 3 · on the object basis, 474 = 21.5% is AIRCRAFT

MEASURED

1,501 of the 2,199 object sightings are from 1952 alone

Figure 4 · six years of data, 68.3% of it one summer

MEASURED

Excellent reports 33.3% unknown, poor 16.6%, doubtful 13.0%

Figure 8 · the top of the claim holds, the middle does not

MEASURED

Five characteristics of six reject one population at better than 1%

Tables II-VII · the report's own test, on the report's own data

MEASURED

66% of the brightness cases and 38% of the speed cases are "not stated"

the missing-data bucket carries a third of the speed signal

MEASURED

Table X's OTHER row prints 1.76 where its own numbers give 0.18

corrected, that table falls below the 5% line printed beside it

MEASURED

Exact p values, the contingency statistic, and χ² = 63 on Figure 8

computed here from the printed counts · 1955 had only a lookup table

MODELLED

A measured difference is not an identification

the report knew this, said so, and is quoted in both directions anyway

READING

View

…

Counting basis

The rule

Characteristic

The knowns

The statistic

Drop "not stated"the bucket the report never priced

The quality table

Insufficient-info. linethe bin competing for the same cases

Labels

Readout · live

Unknown share

—

The fraction

—

Spread, 10 rules

—

χ² on the bench

—

p (computed)

—

Excellent → poor

—

…

What is measured vs modelled here

Measured: every count on this bench is printed in the 1955 report. Figures 2, 3, 4 and 8, Table I, and all twelve chi square tables are carried complete, including the "not stated" buckets and the merged rows. Modelled: the exact p values, the contingency form of the test, the chi square on Figure 8, the ten-rule spread and the 17.1% that the report's own re-evaluation implies. All of it is arithmetic on the printed counts, checked by scripts/sr14-tune.mjs. Reported: the "under 3%" figure, which belongs to the October 1955 public release and to a later case load, not to this file. Reading: what the unknowns were. The bench declines, and so did the report.

The arithmetic, in full

The report's statistic scales the knowns to the unknowns' total and sums (K−n)²/K, treating the knowns as a fixed reference distribution. The textbook comparison of two samples is a 2 × m contingency chi square with expectations from the margins; both are on the switch, and they barely disagree. Degrees of freedom are one less than the number of buckets, and the revised tables merge any bucket whose adjusted knowns fell to ten or fewer, exactly as the report did. p comes from the regularised upper incomplete gamma function, which 1955 could not evaluate: the report looked up 5% and 1% in a table, and those two values are still drawn. The reliability test is a 4 × 2 contingency on Figure 8, which the report never ran.

Where this bench takes liberties

The 21.5% and "under 3%" markers are drawn on the same ruler for contrast, but they are not the same kind of number: the first is Figure 2, the second is the 1955 public statement about a later programme, and the inspector says so. The "apply the re-review" rule exists only on the object basis, because the panel worked case folders. OTHER in these figures absorbs light phenomena, birds, clouds and dust, and psychological manifestations, which are separate codes in the underlying tables. Table I's counts are 69 short of the object-sighting totals, because some cards carried no usable time. And the Table X erratum is this bench's own reading of the report's arithmetic, stated so you can check it: 51 and 54 give 0.18, and the page prints 1.76.

…1 / 23

…

…

…