Academic journal article The Journal of Parapsychology

Notes on Random Target Selection: The PRL Autoganzfeld Target and Target Set Revisited

Academic journal article The Journal of Parapsychology

Notes on Random Target Selection: The PRL Autoganzfeld Target and Target Set Revisited

Article excerpt

Skeptics and parapsychologists alike stress that target selection in parapsychological experiments must be random. Weak randomization procedures are generally counted as a major negative quality point in meta-analyses that include the relationship between study quality and study outcome (Honorton, 1985; Hyman, 1985). In the worst case, they may be grounds for dismissing the study.

One important reason for avoiding improper randomization is that biases in target selection may give the experimenter and/or subjects a clue about the potential target probabilities and therefore invalidate statistical analyses based upon the theoretical target probability.

However, even when the target selection procedure is properly random, the resulting sequence of targets may be quite structured. The 10-bit binary sequence "1111111111" may result from a random binary generator. The probability for such a sequence is equal to all other 10-bit binary sequences such as "1101001101," although the latter may appear more random than the former.

CORRECTIONS FOR RESPONSE BIASES

Response biases are tendencies for subjects to prefer a specific response over others. For instance, if we have a very rigid subject with an absolute preference to select the response "1," then this subject in a 10-trial binary choice experiment will "produce" as a response sequence "1111111111." This, of course, would yield 10 hits in case of the former target sequence. The probability that 10 hits occur by chance is smaller than I in 100 and therefore such a result is said to be statistically significant and considered to be an indication that psi may have occurred. Is that the correct conclusion? In actual experiments responses biases can be quite common, with origins as diverse as differences in the attractiveness of target images in free-response work, to the inability to remember certain targets in forced-choice tests.

A conservative approach is to adjust hit probability from the theoretical .5 to one that is corrected for the subject's response bias and the actual target sequence. If [F.sub.i] is the post-hoc relative frequency of target i and [P.sub.i] is the relative preference for target i, then the corrected target hit probability in a forced-choice, 2-alternative situation becomes:

[P.sub.corr] = ([F.sub.1] x [P.sub.1] + [F.sub.2] x [P.sub.2]) (1)

In the former case this would result in a post-hoc corrected hit probability of 1; all the extra chance hits disappear through this correction. The conclusion here would thus be that the subject was not totally clairvoyant and that nothing happened that needs an explanation.

It is obvious that this post-hoc correction is quite conservative. All responses are considered as "bias" and the possibility that psi influenced some of the decisions is not admitted. This type of correction is useful in defending an experimental result against a critical challenge that the results may have been due simply to subject response biases coinciding with chance imbalances in target selection. In certain circumstances, however, this may be too conservative, since responses which may have been the result of psi information are penalized as evidence of bias. For example, if such a correction were applied to a comparison of different types of targets, a target type that was truly better and produced more correct responses would be more severely "corrected" for supposed bias by the subject.

As long as the relative target frequencies are equal, no correction is needed, since the relative preferences, [P.sub.i], add to 1. (This can be seen in Equation 1, which yields (.5 x .3) + (.5 x .7) = .5 for a case where the relative preferences are .3 and .7 and the targets each have a relative frequency of .5) This is called the closed-deck situation. It has been argued that closed-deck designs may be preferred (e.g., see Hyman, 1994). In a closed deck, the frequency of the target alternatives is equal and randomization is replaced by random shuffle. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.