Dean Radin and Diane Ferrari (1990) reported a metaanalysis of 148 PK dice experiments, involving 2569 subjects and 52 experimenters, that were published between 1935 amd 1987. These experiments were intended to demonstrate the reality of psychokinesis in the throwing of dice, which was usually done mechanically. The total p-value for those trials in which subjects were attempting to influence the dice, either by making a particular face appear uppermost or by causing a dice to come to rest at a chosen location, was 10^-70, a ridiculously small number. The p-value for control trials where no PK was attempted was only .429. The authors concluded that it would take 121 unpublished nonsignificant studies int the file-drawer for every published to raise the experimental p-value to .05.
Unfortunately, quality ratings proved to correlate significantly with study outcomes, supporting the artifact interpetation of the significant effect. However, the authors had previously adopted a supplementary procedure, often used by meta-analysts, of eliminating studies that obtained either extremely high scores on the PK task or extremely low scores. (Such outliners are considered suspect by many scientists, and their inclusion can sometimes bias a statistical test.) This trimming reduced the sample size from 148 experiments to 96. The bad news is that the overall p-value for the PK effect was raised from 10^-70 to 10^-7, which, however, is still much smaller than .05. The good news is that the correlation between study outcome and quality was no longer significant (p=.372). The authors therefore concluded that real PK had been demonstrated by the dice throwing technique.
Using the blocking procedure, the authors found that the PK effect was much stronger for subjects preselected by prior testing than for unselected subjects, but it was still significant for the unselected group. They also found that the methodological quality of PK dice studies has improved significantly over the years.