What the Arithmetic Rests On

The committee concluded that death-coincident apparitions occurred vastly more often than chance allowed, and the objections to that conclusion arrived immediately and have not been improved on much since. The sample was not random: volunteers asked people they knew, and a person with a story to tell is more likely to answer than one without. The data are retrospective self-report, sometimes about events years earlier, and memory reshapes a vague impression into a specific one once the outcome is known, dating it more precisely than it deserves. The twelve-hour window is an analytic choice, and a different window gives a different rate. And the comparison assumes the two events are independent, which fails in exactly the cases most likely to be reported, since someone whose relative is gravely ill is both more likely to think about them and more likely to have them die.

None of this makes the census worthless. It remains the only attempt anyone has made to establish a population rate for the experience, and its raw finding, that roughly one person in ten reports a waking hallucination at some point, has held up reasonably well against modern surveys. It is the inference to coincidence, not the survey, that the criticism reaches.