"This is an idea that a skeptic gave:
"In ganzfeld studies, instead of 4 pictures, you start with 2 paired sets (formed randomly) of 4 pictures (8 pictures altogether) - each picture in one set is paired with a picture in the other set. One picture from one set is randomly selected as a target (both the target and set selection are random) for the sender. You go through all of the usual procedure. At the end, the receiver is randomly shown one set of 4 pictures - either the set which contained the target or the set which did not. The receiver then judges which picture was the target, as per the usual procedure. No feedback is given. It is a 'hit' if either picture in a pair was the target. Then you compare the number of hits when the same set of pictures was used for the sender and receiver vs. the number of hits when different sets of pictures were used for the sender and receiver. Everyone remains blind as to the condition throughout the entire experimental series. This design requires very little change in the procedure, yet all of a sudden it contains the most powerful tool we have in research design - a control group. In one fell swoop, it would eliminate any of the bickering and nitpicking between believers and skeptics as to what the results mean.
I am truly curious about this though. Does anyone have any inside knowledge into why control groups were never really built into the design? It seems so obvious that their absence is weird.
What do you think about the design? And why the ganzfeld studies have not controls groups?"
The suggested method does not test for the presence of telepathy any better than the existing method. The ganzfeld technique is already controlled by protocol. That means that as long as the targets are selected at random, there is no way that any response bias can systematically guess the right target. So comparing the observed hit rate vs. the chance expected hit rate over the long run is a perfectly suitable method for inferring the presence of telepathy. The adequacy of this approach has been confirmed many times by statisticians.
What the proposed idea does is test whether a "sender" is necessary to perceive information from a distance. Such tests have been done, both within the ganzfeld paradigm and in remote viewing designs. The answer is no -- which raises an epistemological question about how one would test for "pure" telepathy. So far, no one has a good answer to that question, although the closest we've come to date are studies using physiological methods, including EEG, to look for correlations in bodily responses between isolated people. Those studies suggest some form of connectivity between people, which is in alignment with the general idea of telepathy.
I've found that confusion about experimental methods often reflects a lack of familiarity with the relevant literature. This is true in any scientific discipline. In the present case, these issues have been discussed in detail for many decades by those working in the field. In addition, a statement like "In one fell swoop, it would eliminate any of the bickering and nitpicking between believers and skeptics as to what the results mean" is admirably enthusiastic, but it's also naive. The existing evidence is already persuasive to stupendously high degrees of statistical certainty. Another new experimental design is not going to change the nature of the debate, which is focused on theoretical and ideological complaints, and not on a dispassionate evaluation of the empirical data."
I would just add to this that should an easily repeatable experiment appear, without the requirement of special subjects, then that might well overwhelm prevailing skepticism. Use of special subjects is appealing and makes much sense, but the skeptic can always reply that those people are cheating, or that they cannot be easily found so independent replications cannot be performed.
That said, participants selected on the basis of some talent will almost always do better in experiments, so if you're interested in learning something rather than in convincing skeptics, it's better to go with selection, provided you have the resources to actually do that (it's a non-trivial problem).