lew@ihuxr.UUCP (Lew Mammel, Jr.) (02/03/84)
This is about the article "The Persistent Paradox of Psychic Phenomena: An Engineering Perspective" by Robert G. Jahn, which appeared in Proceedings of the IEE, Vol. 70, no. 2, February 1982. The article is a general survey, but it concludes an exstensive description of a PK experiment conducted by a Princeton undergraduate, which the author supervised. A graph depicting the results of this experiment was shown on the NOVA ESP show. I'll confine my remarks to a brief critique of this experiment. The graph shown on NOVA showed "Cumulative Deviation" vs. "Number of Trials" for three cases, PK+, PK-, and BL (base line). The PK+ and PK- graphs showed trials in which the "operator" tried to psychically influence the REG (Random Event Generator) in a positive or negative direction. Each trial consisted of 200 random binary events. The trials were divided into 5 series. The graph represented data shown tabularly as: type trials mean std t-score Pt n+/n- BL 23000 100.045 6.980 .978 .164 10891/10782 PK+ 13050 100.223 6.979 3.644 e-4 6310/6004 PK- 12100 99.709 6.968 -4.596 2e-6 5462/5956 The t-score is given by (mean - 100)/(SD*sqrt(trials)), and shows the deviation of the cumlative score compared to the expected deviation. Pt is the probability of achieving that deviation (or greater) by chance, assuming a true mean of 100. n+/n- presumably shows the number of trials which came out above and below 100, but I couldn't find anything about it in the text. A few facts about the experiment: ALL trials were conducted by a single person, the experimenter, as "operator". The binary events were generated at rates of either 100 or 1000 per second by a hardware noise generator. They were tallied in an "alternating" mode to compensate for any systematic bias. It was stated that the event rate didn't affect the PK performance. That is, analyzed separately, the slow and fast rates both showed the same PK bias as the combined data. The trials were conducted in "runs" of 50 each. These could be in automatic mode, with each trial initiating automatically, or in manual mode. Read this quote from the article: This operator attempted, on instruction or volition, to distort the trial counts either toward higher or lower values. The several options of sampling number, sampling frequency, +/- polarity, and manual/automatic sequencing were variously determined by random instruction, operator preference, or experimental practicality, and recorded before the beginning of each trial. Now ... did she record her PK polarity before each trial? Read it again! Also, the device has a reset button. In fact, the whole issue of data collection isn't dealt with at all. Here's a real kicker: She tried a whole new series of 2000 event trials in which the PK+ and PK- means came out to be 1000.380 and 999.569. Jahn calls this result "curiously ambivalent". The PK seems to be limited to a fraction of an event per trial. This leads me into a whole series of ruminations on what constitutes an event. The hardware event generator sums over at least trillions of quantum events. A PK bias of 1ppb would produce perfect scores at the output. Conversely, if you counted each trial as an event with the PK bias as reported and >100 being a hit, the bias amounts to several percent. Evidently an event is whatever is required to keep the PK lurking at the fringe of the statistics. BAH! Lew Mammel, Jr. ihnp4!ihuxr!lew