"Future Telling": A Meta-Analysis of Forced-Choice Precognition Experiments, 1935-1987

📌 Appears in:

⚡ Bem Feeling The Future ⚡ Meta Debate 📖 Newcomer Survey

Plain English Summary

This is the big one for precognition research — a massive review pulling together 309 experiments conducted over more than fifty years, involving nearly 2 million trials and 50,000+ participants. In these "forced-choice" tests (where people try to guess a future random outcome), the overall hit rate was small but wildly statistically significant, with odds against chance of roughly a trillion trillion to one. About 30% of individual studies hit significance on their own, and you'd need over 14,000 unpublished negative studies hiding in file drawers to erase the effect — a powerful argument against cherry-picking. Here's what's really interesting: study quality didn't weaken the results; higher-quality work actually showed slightly stronger effects. The researchers also pinpointed the sweet spot for success: use people who seem naturally gifted, test them one-on-one, give them immediate feedback after each guess, and keep the time gap short. Studies that nailed all four conditions produced an astonishing 87.5% significant hit rate. The authors made a memorable comparison: this precognition effect is comparable in size to well-accepted medical findings like aspirin preventing heart attacks, an argument that's been repeated in psi research ever since.

Abstract

We report a meta-analysis of forced-choice precognition experiments published in the English-language parapsychological literature between 1935 and 1987. These studies involve attempts by subjects to predict the identity of target stimuli selected randomly over intervals ranging from several hundred milliseconds to one year following the subjects' responses. We retrieved 309 studies reported by 62 investigators. Nearly two million individual trials were contributed by more than 50,000 subjects. Study outcomes are assessed by overall level of statistical significance and effect size. There is a small, but reliable overall effect (z = 11.41, p = 6.3 x 10~*). Thirty percent of the studies (by 40 investigators) are significant at the 5% significance level. Assessment of vulnerability to selective reporting indicates that a ratio of 46 unreported studies averaging null results would be required for each reported study in order to reduce the overall result to nonsignificance. No systematic relationship was found between study outcomes and eight indices of research quality. Effect size has remained essentially constant over the survey period, whereas research quality has improved substantially. Four moderating variables appear to covary significantly with study outcome: Studies using subjects selected on the basis of prior testing performance show significantly larger effects than studies using unselected subjects. Subjects tested individually by an experimenter show significantly larger effects than those tested in groups. Studies in which subjects are given trial-by-trial or run-score feedback have significantly larger effects than those with delayed or no subject feedback. Studies with brief intervals between subjects' responses and target generation show significantly stronger effects than studies involving longer intervals. The combined impact of these moderating variables appears to be very strong. Independently significant outcomes are observed in seven of the eight studies using selected subjects, who were tested individually and received trial-by-trial feedback.