cooper@pbsvax.dec.com (07/04/86)
Bill Jefferys ({allegra,ihnp4}!{ut-sally,knew}!utastro!bill, bill@astro.UTEXAS.EDU) has claimed as an explanation for the decline effect in parapsychology (the tendency for the scoring rate to decrease over the course of an experiment) the following: >Not if you have selected for further study precisely those individuals >who scored high initially, and if you include those earlier trials >in the overall test results. I admitted that such the error of inappropriately mixing screening with testing trials might have occurred on rare occasions in the parapsychological literature (as a chemist might admit that dirty test tubes may have been the actual cause of some published results). I thought it was *very* unlikely that any such flaw was the source of any claimed evidence for the decline effect which had appeared in a refereed parapsychology journal or other legitimate piece of parapsychological literature, however. The error would have been too obvious in this case to have been missed by the referees. I therefore challenged him to produce even a single example of this, by "direct or indirect" citation. He responded with: >The University of Texas doesn't subscribe to parapsychological journals, >so I can't satisfy you completely. However, > >A quick trip to the library turned up the book, "Parapsychology: Science >or Magic?", by J. E. Alcock (Pergamon, 1981). His discussion and explanation >of the "Decline Effect" is basically the same as the one I suggested. This is a legitimate and reasonable answer to my request. It is the type of thing which I had in mind when I was speaking of an "indirect" citation. Unfortunately, in this case, it is not enough. I own or have read a fair amount of the critical literature, but I do not own nor have I read this book. I checked for it in the MIT library system but no copy is owned by them either. This is, of course, no criticism of the book. It just means that I cannot effectively respond without further information. Since you have access to this book please post the citations to the literature which the author uses to support his/your thesis. If he does not provide such citations, then this book does not even partially answer my challenge. I do not question that others have made this accusation before. I have seen or heard it made a number of times. I have just not seen any concrete evidence to back it up. >He also cites Spencer Brown ("Probability and Scientific Inference", 1957) >who showed that data from 100,000 published random numbers, analyzed using >the Quartile method in use at Duke University, showed a very significant >Quartile Decline effect. Sorry, I fail to see how this supports your thesis at all. G. Spencer-Brown is best known for his book "The Laws of Form." (I'm doing this off the top of my head so I can't give publication info) It is the thesis of this work that there is something fundamentally wrong with all existing formal logic, and therefore, with the foundations of all modern mathematics. A friend of mine once said of it, "It would clearly be a work of genius if only it made any sense at all." Many people who read it, including those with a fair amount of mathematical sophistication, are left with the feeling that Spencer-Brown seems to have said *something* of importance but it's completely unclear what. Spencer-Brown considered himself a critic of parapsychology. He felt that parapsychologists felt that they were investigating a physical phenomena while in reality, all they were doing was demonstrating that probability and statistical theory were fundamentally flawed (of course). Specifically, he believed that published random number tables were not "random". He never really specified the nature of their non-randomness except that it would conveniently result in the success of many parapsychological tests whatever use they made of the table (the dominant methodology for randomization during the 50's in parapsychology was to use published random number tables, such as the RAND 1,000,000 random digits, to determine the targets for the experiment in one or another ways). I should say that I would consider the demonstration of such an error in elementary statistics and probability to be a *very* unexpected outcome. In my opinion, it would, however, justify the most basic claim of parapsychologists, i.e., that psi is something real and important. I have heard similar statements by other parapsychologists. Of all the resolutions I can think of for the mystery of psi, this one would have one of the largest impacts on the practice of science and engineering. Anyway -- To prove his point Spencer-Brown performed the cited experiment. He entered a random number table (probably the RAND table, but I don't remember for sure) at two arbitrary points. He declared one entry point as being for his targets and the other as being for his calls. He then matched the next N digits (I'll take Bill's word that it was 100,000) one by one and came up with a significant number of matches and a scoring decline. (I may be misremembering the details here, its been years since I read it. I have the feeling that his actual procedure was a bit more complex than this, but I'm pretty sure that this was the essence). I won't bother to discuss generally the various interpretations of this experiment here, as its irrelevant to the point. If Spencer-Brown did what he claimed (entered the table with the first pair of entry points he tried) then this is either an irrelevant fluke, or *disproves* your point, since the decline effect appears without screening trials to be inappropriately included with the data. If Spencer-Brown actually tried multiple sets of entry points until he found some which gave many early hits, then it still gives no support to your claims. No one is denying that a decline effect would be produced if someone committed this obvious and egregious error (although, the apparent *continuous* decline which appears to be the characteristic of the actual effect would not be produced this way). Spencer-Brown was a *critic* of parapsychology, however unconventional, rather than a parapsychologist. His behavior therefore cannot be taken to say anything about the way that parapsychologists conduct experiments. >I also would guess that some of the articles cited by Marks (1986: _Nature_ >vol 320, pp. 119-124) would treat this issue, particularly Refs. 42-45. >Unfortunately, I don't have access to any of them so I don't know for >sure. I am only familiar with three of those four references. None of these touch upon the issue. I doubt if the fourth supports your thesis. First, because if John Beloff, probably Great Britain's leading parapsychologist, had made such a claim, I suspect I would have heard about it. Secondly, the decline effect does not seem to me to be particularly germane to the topic of the paper, i.e., the inappropriateness of strict repeatability as a criterion for the acceptance of evidence of psi. Third, I have heard Beloff discuss this topic elsewhere, and he made no mention of the decline effect. As for the other references, none that I have read provide any solid support for your thesis. I am willing to accept an "indirect" reference through the critical literature if I have access to the work in question. But a reference to a set of references covering a moderate amount of the technical critical literature of the last 20 years, with a comment of "something here probably supports my claim" is hardly a reasonable response. >The odd thing is that the "Decline Effect" is cited by parapsychologists >as "Evidence" of the reality of Psi. In any other field, such an "effect" >would be cited as evidence that the original observations were flawed >in some way, or were the result of a statistical fluke. Unless it were found repeatedly under a variety of experimental conditions and scientists involved were at all competent. Then they would give it a name (such as "the decline effect" or "conditioning extinction" or "pulsar slowdown" :-) and attempt to study it. Topher Cooper USENET: ...{allegra,decvax,ihnp4,ucbvax}!decwrl!pbsvax.dec.com!cooper INTERNET: cooper%pbsvax.DEC@decwrl.dec.com Disclaimer: This contains my own opinions, and I am solely responsible for them.
bill@utastro.UUCP (William H. Jefferys) (07/10/86)
Topher Cooper says: >Bill Jefferys ({allegra,ihnp4}!{ut-sally,knew}!utastro!bill, >bill@astro.UTEXAS.EDU) has claimed as an explanation for the decline effect >in parapsychology (the tendency for the scoring rate to decrease over the >course of an experiment) the following: >>Not if you have selected for further study precisely those individuals >>who scored high initially, and if you include those earlier trials >>in the overall test results. >I admitted that such the error of inappropriately mixing screening with >testing trials might have occurred on rare occasions in the >parapsychological literature (as a chemist might admit that dirty test >tubes may have been the actual cause of some published results). I thought >it was *very* unlikely that any such flaw was the source of any claimed >evidence for the decline effect which had appeared in a refereed >parapsychology journal or other legitimate piece of parapsychological >literature, however. The error would have been too obvious in this case >to have been missed by the referees. Aha! I see the problem now. We have been talking at cross-purposes. It is my fault; I was trying to keep my response to Dave short, and I worded it very badly. I didn't mean to accuse experimenters of such an obvious mistake. But I agree that that is how my comment reads. I have to apologize to you. I had in mind a more subtle kind of unconscious "selection" that could take place after the trials began in earnest. There could be a "selection effect" of the following sort: Alcock cites a number of studies showing that initial success in this kind of experiment is much more likely to lead to belief by the subjects (and presumably by the experimenter) that nonchance effects are involved, than is initial performance at the chance level followed by increasing success. (The outcomes of the "random" results in some of these studies were, unknown to the subjects, manipulated by the experimenter). In any population of subjects that has been selected for further study, a certain fraction will (purely by chance) continue to score well for a while, while others will revert quickly to the chance level. (This assumes there are in fact no other unknown biases affecting the study). There are a number of factors working that will tend to keep the subjects who continue to score well in the study for more and more trials, while those who initially fall by the wayside will tend to drop out. One is the tendency of the experimenter to want to go with a winner. The more spectacular the initial success, the more time will be spent with such a subject. After all, the name of the game is to publish papers, and no experimenter likes publishing papers about his failures. Another is the tendency of the subject to drop out if he or she is not showing "paranormal abilities". The sooner such a subject's performance drops to the chance level, the less the reinforcement, and the more likely it is that he or she will drop out. This would happen even if the experimenter tried to keep all subjects in to the bitter end of a prescribed number of trials. The net result would be a set of short trials with subjects who scored at the chance level; a smaller set of longer trials with subjects who initially scored above chance but reverted to chance rather quickly; and a small set of subjects who initially scored spectacularly, who got a lot of attention and many trials, and who also reverted to chance levels. In other words, a "Decline Effect". The reason why one doesn't see an "Increase Effect" is that few if any studies would do extensive testing on subjects that start out poorly. The effect could be compounded if there are biases (e.g., unconscious cueing) which initially inflate scores, but which come under better control as attention is focussed on a high-scoring subject. Is this an unreasonable scenario? If so, why? (Topher, if you respond to this, please E-mail me a copy, as I will be at a meeting for a few days & our machine expires news in 3 days). -- Glend. I can call spirits from the vasty deep. Hot. Why, so can I, or so can any man; But will they come when you do call for them? -- Henry IV Pt. I, III, i, 53 Bill Jefferys 8-% Astronomy Dept, University of Texas, Austin TX 78712 (USnail) {allegra,ihnp4}!{ut-sally,noao}!utastro!bill (UUCP) bill@astro.UTEXAS.EDU. (Internet)
cooper@pbsvax.dec.com (Topher Cooper DTN-225-5819) (07/16/86)
Bill Jefferys ({allegra,ihnp4}!{ut-sally,knew}!utastro!bill, bill@astro.UTEXAS.EDU) has proposed an explanation for the decline effect in parapsychology. Essentially he proposes that parapsychologists tend to "stick" with subjects who happen to score well initially (the reader is referred to <981@utastro.UUCP> for details). This would produce a decline effect by selection. He asks if this is "an unreasonable scenario." It is not unreasonable but it is irrelevant to what we have been discussing. There is not one "decline effect" but many. Among them is the run decline effect, the score-sheet decline effect, the run series decline effect, the experiment decline effect, and the subject decline effect. These differ in the unit over which the decline is measured. A run is a series of trials which are done essentially in "one go". Classically, a run corresponded to a single deck of 25 Zener cards. The run decline effect is the tendency for the average score for the first trials of an experiment to be higher than the second, which in turn is higher than the third etc. The score-sheet decline has to do with the old manual scoring sheets. It was found that there was a frequently a declining trend from the average of the first run on each sheet to the last run on each sheet. The run series decline is a tendency for scores to decline over runs in a run series. A run series a part of an experiment over which conditions for any given subject are held constant. The experiment decline is a tendency for the average score per run to decrease over the course of an experiment. The subject decline effect is the tendency for the performance of a subject used in a series of experiments to decline over time. It is frequently noted by researchers that good subjects seem to "burn out". These different decline effects can be put together into a single phenomena by assuming that there is a continuous decline for any given subject. Novel conditions seem to cause some recovery over the basic trend. This is, however, speculation -- the different decline effects may come from distinct causes. The best established of these effects are the run decline and the experiment decline. We have been very specifically speaking of the experiment decline. For example, in the article in which he makes this proposal Bill Jefferys quotes me as saying: >> . . . the decline effect >>in parapsychology (the tendency for the scoring rate to decrease over the >>course of an experiment) . . . Nowhere in this discussion has anyone disputed this definition. A properly designed experiment of the general form that a parapsychology experiment takes must include a clearly specified termination criteria. The termination criteria specifies when the experiment is to stop. The termination criteria must be shown to be independent of the hypothesis under study. Almost universally, the simple termination criteria of stopping when the/each subject has performed a specified number of trials is used. One therefore does *not* have some initially good subjects with many (declining) trials, and (many more) initially poor subjects with a few trials as required by Bill's hypothesis. In a given experiment one has one subject or several subjects with the same number of trials for each. The number of trials is determined before the experiment is conducted. The decline effect is then either demonstrated across all the subjects or for those subjects whose *total* score is the highest. Bill says that > This would happen even if the experimenter >tried to keep all subjects in to the bitter end of a prescribed number >of trials. but does not explain how -- presumably, by the experimenter failing. In such a case, the experimental protocol would have been violated. According to standard practice, either the experiment would have to be abandoned or the experimental report would have to mention this fact. Selection biases from weak subjects dropping out are well understood in parapsychology. Any analysis of such an experiment, for the decline effect or for any other purpose, would have to take this into account. The scenario you propose *might* explain the *subject* decline effect, but it requires that high psi scores be a result of chance. This is the one hypothesis which can be categorically rejected. The odds against the results of parapsychological experiments being due to chance rather than rather than one or more systematic biases (conventional or unconventional) are astronomical, to say the least. The odds against subject differences being by chance are only slightly smaller. A variant on this hypothesis might, however, reasonably explain the subject decline effect. If subjects' average score (as somehow magically measured independently of any experiment decline effect) tends to both increase and decrease over periods of years (e.g., oscillate) then subjects who are used repeatedly in experiments would tend to be selected when they are scoring high. Inevitably, they would show a decline over many experiments. As far as I know the subject decline effect has simply been an observation by experimenters. I know of no study which purports to demonstrate it as a consistent effect. It is therefore at best a very weakly demonstrated effect which is, nevertheless, plausible as an extrapolation of the experiment decline effect. This hypothesis, though interesting, does little to change this basic picture. >The effect could be compounded if there are biases (e.g., unconscious >cueing) which initially inflate scores, but which come under better >control as attention is focussed on a high-scoring subject. It would be except that care is, of course, taken to apply controls uniformly across subjects and across time. This is another form of the "incompetent experimenter" assumption. Topher Cooper USENET: ...{allegra,decvax,ihnp4,ucbvax}!decwrl!pbsvax.dec.com!cooper INTERNET: cooper%pbsvax.DEC@decwrl.dec.com Disclaimer: This contains my own opinions, and I am solely responsible for them.