Watt's analysis of judges' ratings in her 2014 dream precognition study

I grasped the nettle and read the second of the references mentioned by ersby, which is available here:
http://www.patriziotressoldi.it/cmssimpl...nali16.pdf

It seems the idea is that:
(1) The person has precognitive access to the whole range of possible future events. Not too much is said about this in the paper, and the relationship with experimenter psi isn't discussed, but there's an anecdote about an associative precognitive remote viewing in which the subject, despite having scored a "hit", refused to view the target.
(2) Precognition applies not only to cases in which the subject later perceives the information, but also to cases where the subject remains blind and only the experimenter perceives it. There's even a reference to an experiment where the results were analysed by computer, and no human being ever perceived the target. The authors acknowledge that precognitively reading the "mind" of a computer is "seemingly implausible", but argue that it can't be ruled out.

Apparently Decision Augmentation Theory is weirder than I'd realised. The SPR Facebook page has a link to a paper by Edwin May and Sonali Bhatt Marwaha, discussing an experiment in which a deceased grandmaster purportedly played a game of chess, through a medium, against Viktor Korchnoi:
https://www.academia.edu/31084244/An_Alt...Chess_Game

They say that the moves chosen by the medium can be explained by Decision Augmentation Theory, because once the choices had been made, records of them would exist, and the medium could perceive these precognitively before making his choice. So DAT can include a temporal loop, in which a choice is made using precognitive information about what choice will have been made. (But where does the information come from to start with?)

Just to note that in Hyman's (1994) re-analysis of recent Ganzfeld studies by Honorton and coworkers, he found that there was a strong correlation between the hit rate and the frequency with which targets were presented to the subjects. Targets offered more frequently tended to be selected in a larger percentage of cases than targets presented rarely.
http://deanradin.com/evidence/Hyman1994.pdf

As subjects tend to be biased towards selecting particular types of images, that might be consistent with the idea that it was the (supposedly random) choice of target that was being influenced by psi (perhaps retroactively) , rather than the subject's choice. However, there were some other odd features of this correlation, notably the fact that the hit rate was actually reduced on the first appearance of the target, and enhanced only on subsequent appearances.

Just to note that in Hyman's (1994) re-analysis of recent Ganzfeld studies by Honorton and coworkers, he found that there was a strong correlation between the hit rate and the frequency with which targets were presented to the subjects. Targets offered more frequently tended to be selected in a larger percentage of cases  than targets presented rarely.
http://deanradin.com/evidence/Hyman1994.pdf

As subjects tend to be biased towards selecting particular types of images, that might be consistent with the idea that it was the (supposedly random) choice of target that was being influenced by psi (perhaps retroactively) , rather than the subject's choice. However, there were some other odd features of this correlation, notably the fact that the hit rate was actually reduced on the first appearance of the target, and enhanced only on subsequent appearances.

I wondered if I had fallen prey to a statistical fallacy there, but I think Hyman's observation is consistent with that idea. Because if a certain target image is more likely to be chosen because of response bias, then it will tend to have a higher hit rate than less popular images (if it was so popular that it was chosen 100% of the time, then it would necessarily have a 100% hit rate). And if the process of choosing the target were influenced by the guess made by the subject, then popular images would also be presented to the subjects as targets more often.

Honorton and coworkers did perform a number of tests of how uniformly distributed their random numbers were. But I think the only test that would have shown up a tendency of more popular images to be picked more often as targets was performed on data from control sessions conducted in the absence of subjects.

I still can't think of a reason why the effect found by Hyman should be reversed for the first appearance of the target, though.

I'll just note here that the pilot studies for Mossibridge et al.'s "Future Photon Experiment" might also be consistent with the idea of a psychokinetic influence on a random number generator, rather than retrocausation.

The first pilot study found that the fringe pattern in a double-slit experiment during the first minute was correlated with the overall duration of the illumination, which was chosen randomly after 90 seconds. But a second pilot study found that when the overall illumination was chosen randomly after 22 seconds, there was no significant correlation. That seems consistent with the idea, because the data for the full minute hadn't yet been collected when the random choice was made, so in the absence of retrocausation the random choice could be influenced only by the first 22 seconds' data, and when the subsequent data were added any correlation could be expected to become much weaker.

However, this was only an unpublished pilot study, and it's indicated that problems were found with the calibration of the device.