An Interview with Rupert Sheldrake from the Seattle Post Intelligencer 1st April 2003
The Psychic Staring Effect: An Artifact of Pseudo Randomization , from Skeptical Inquirer magazine : September/October 2000, which makes a pretty good case that Sheldrake's "results" are a consequence of the poorly randomized sequences he chose to use for his test cases.
Short form: When choosing "when to stare and when not to stare" in his tests to see if subjects could "sense" when the "starer" was staring at them or not, Sheldrake used poorly randomized sequences which "spread out" the staring periods more than would be the case in a truly random test. The problem with this is that it is well known that when humans (i.e. the "starees" in his tests) guess things "at random", their guesses are more "spread out" and include more "alternations" than is the case for truly random sequences.
Thus, while human-style guessing will score no better than chance against *uniformly* random events, they will score *better* than chance against any sequence (like Sheldrake's trials) which are "random" in ways biased towards "human style" guessing (e.g. "spread out" trials).
For example, if people are asked to guess a random arrangement of 3 white balls and 7 black balls, their guesses will more often look like "BBWBBWBBWB" (with the 3 white balls "spread out" across the 10 slots) than they will the equally likely arrangement of, say, "BWBWWBBBBB", where the white balls happen to have come out more "clustered".
So when asked to guess "test" sequences of balls which happen to be not truly random, but instead are arranged to be more "spread out" than would be the case by chance, their guesses are considerably more likely to get "hits", BY PURE CHANCE, than you'd expect by simple guessing, and it would give the false *impression* that they had somehow "seen" the sequences they were trying to guess (by ESP?).
When Marks and Colwell repeated Sheldrake's test protocol but with *purely* random test sequences, the positive "results" evaporated, and there was no indication that test subjects could successfully "sense" when they were being stared at.