Here is a quote from Dean Radin’s book, The Conscious Universe: The Scientific Truth of Psychic Phenomena, published by HarperEdge, 1997, from pages 114 – 115. (The change in font size and color is by the P-I-A editors)
"Honorton and Ferrari surveyed the English-language scientific literature to retrieve all experiments reporting forced-choice precognition tests. They found 309 studies, reported in 113 articles published from 1935 to 1987, and contributed by sixty-two different investigators. The database consisted of nearly two million individual trials by more than fifty thousand subjects. The methods used in these studies ranged from the use of ESP cards to fully automated, computer-generated, randomly presented symbols. The most frequently used participants were college students (in about 40 percent of the studies), and the least frequent were the experimenters themselves. (in about 5 percent of the studies). People had been tested both individually and in groups.
"The future targets were selected in many ways. Some studies used quasi-random methods relying on naturalistic events, like the average daily low temperatures recorded in a large group of cities located throughout the world. Other studies used informal methods such as dice tossing and card shuffling, or more formal techniques such as the use of tables of preprinted random numbers and electronic random-number generators. The time interval between the guesses and the generation of the future target ranged from milliseconds to a year.
"The combined results of the 309 studies produced odds against chance of 1025 to one—that is, ten million billion billion to one. This eliminated chance as a viable explanation. The possibility of a selective-reporting bias—the file-drawer problem—was also eliminated by determining that the number of unpublished, unsuccessful studies required to eliminate these astronomical odds was 14,268. Further analysis showed that twenty three of the sixty-two investigators (37 percent) had reported successful studies, so the overall results were not due to one or two wildly successful experiments. In other words, the precognition effect had been successfully replicated across many different experimenters.
"Successfully replicating an effect does not mean that the results observed in different experiments will be identical, because there will always be some variations in study designs and participants. Instead, we would expect the results to be about the same, known in statistical terms as "homogeneous." In meta-analyses it is not expected that the effects observed in different studies will be homogeneous until the "outliers" are trimmed away. These are studies that for one reason or another produced wildly large or wildly small effects, possibly because of design problems, or the use of dramatically different procedures or personnel, or just by chance. In any case, to be sure that the same results have been replicated, a homogeneous set of effects is commonly created by trimming from the full set 10 percent of the studies producing the largest effects and 10 percent of the studies producing the smallest effects.
"After performing this trimming, Honorton and Ferrari were left with 248 studies, and the total number of investigators was reduced from sixty-two to fifty-seven. But the combined effect of the remaining 80 percent of the data still produced odds against chance of a billion to one. This means that these fifty-seven investigators observed precognition effects that were effectively the same, and these effects could not be attributed to chance or to selective reporting."