The new TV program featuring John Edward, one of the currently popular "cold readers" who is "speaking with the dead," has brought the JREF a storm of inquiries in recent weeks. This vaudeville act, which made fortunes for Anna Eva Faye and the Davenport Brothers in the early 1900s, has now embraced television with great success, supported by such perceptive TV hosts as Larry King and Montel Williams, who know an attractive bamboozlement when they see it. When Edward, along with Sylvia Browne, James Van Praagh, and veteran John Anderson, emerged recently to beguile the public, we had a resurgence of interest in what the famous Fox sisters started way back in 1848 and became known as spiritualism.

The process of "cold reading" can be rather subtle and clever, though frankly I don't see in any of the current crop of readers the resilience or inventiveness that I've observed in such performers as Doris Stokes or Doris Collins, in the UK. I describe the act on this web page. We just cannot answer every inquiry individually, and I direct you to click on "The Randi Files" and then on "The Art of Cold Reading." My "Encyclopedia of Claims, Frauds and Hoaxes of the Occult and Supernatural" also has an entry on the subject.

Our good friend Eric Krieg sent us the following article by Tom Napier, who has graciously given us permission to publish it here. (I have taken the liberty of inserting a few notes in order to clarify some points). It expresses rather well the situation in regard to the standards that the JREF has established for the $1,000,000 challenge. Tom Napier has a BS in Physics, a Masters in Electronics, and lots of Common Sense. His work experience includes stints at the European Space Technology Center, and CERN (European Laboratory for Particle Physics), and he is a founder of PHACT, the Philadelphia Association for Critical Thinking.

Are the Randi Tests Fair?

by Tom Napier

It is often argued that supernormal powers are so sporadic in their operation that they cannot be tested by the type of experiment required by James Randi for the JREF $1,000,000 award. I would like to look at this contention from the point of view of a physicist with some knowledge of statistics. I should emphasize that, while I was present as an observer during the unsuccessful demonstration of the existence of the human energy field sponsored by James Randi and Bob Glickman in Philadelphia in November 1996, I have no direct connection with Randi or JREF and I certainly cannot make any representations on their behalf.

From my reading and my own observations it appears that the object of a Randi test is to distinguish with high probability between people who actually have the ability they claim and those who do not. The first decision which Randi must make is whether the claimed ability qualifies for the award, that is, does it lie outside the range of normal human physiology or beyond the laws of science as they are presently understood. No one disputes that one can bend a spoon with one's hands, no prizes for that. Bend one by looking at it as it sits on a table and you would qualify. The ability also has to be verifiable in some way. You might claim to feel cold every time a ghost walks through the room but, unless there is an independent way of determining the presence of a ghost, your claim is meaningless.

Note by Randi: before any of that takes place, we have to (a) get the claimant to state clearly what he/she can do, under what conditions, and with what accuracy. This, as Andrew Harter at the JREF will verify, is often the very hardest thing for us to ascertain. Then, as stated in the rules, we have to (b) conduct a preliminary test - with much less stringent conditions and much more attainable percentages of success. Please also note that to date, no one has successfully completed a preliminary test.

Once it has been agreed that an ability is supernormal and can be tested it is necessary to devise a suitable test protocol. This is always designed in conjunction with the person being tested. It is they after all who are making claims about what they can do. There must be some target performance which, if achieved, shows that the ability probably exists. There should be a second target which, if not achieved, shows that the ability does not exist. Between the two there will be a fuzzy area in which we cannot say for sure whether or not the ability exists. The test target must be set so that it can be easily achieved by the truly supernormal but is unlikely to be achieved by chance by someone without any abnormal ability. After all, if the probability of getting a passing score were, say, 1% then the Randi award would have been collected ages ago, even if no one had any paranormal powers whatsoever. The target must be such that probability of getting a passing score by chance is truly insignificant, perhaps less than one in a million. (Even with those odds, there is a 1% chance that one of the first 10,000 applicants would achieve a passing score by pure chance.)

"Not fair," cry the proponents, "these gifts don't work perfectly. You have to cut us some slack." On the other hand, claimants tend to be absurdly optimistic. Before being tested astrologers and dowsers have claimed 100% accuracy; it would be reasonable to hold them to the performance they claim. When Randi and Glickman tested a therapeutic touch practitioner she set her own standard. She showed she could distinguish between two people's energy fields with 100% accuracy - when she could see the subjects. When tested under exactly the same conditions, but with the subjects hidden from view, she scored 11 out of 20, a result entirely consistent with random guessing. A score of 18 or over would have been accepted as qualifying her for the full, money on the table, test.

Note by Randi: Well, not quite yet. Twenty trials were not quite enough for a preliminary test, which this was. But the practitioner, against her initial agreement with us, simply refused to go on with it.

In every test I've heard of, the experimenters have allowed a much less ambitious target to be used as the criterion of success. Far from making things hard, experimenters go out of their way to make things easy. They don't want to give the claimants the slightest excuse for failure. Still, you have to draw the line somewhere. I sometimes claim to have a gift to predict the sex of an unborn child. It is a powerful gift but it is right only 50% of the time. Why does that get a laugh from the audience? Because one can do just as well by guessing. Being right half the time proves nothing, precisely because the odds of guessing right are also 50%. Does that mean that if someone claims an ability which works only half the time, we can never prove it true? The answer is no, we just have to work out a test protocol where the probability of correct guessing is much less than 50%.

Let me give an example. We are going to test a dowser who can detect gold. If he's a typical dowser he has never carried out a scientific test of his ability yet he claims a 100% success rate. He probably means that whenever he knows gold is present he gets a dowsing reaction 100% of the time. The important thing to find out is whether he can detect the presence of gold when neither he nor anyone else present knows where the gold is. In a typical test a number of identical containers such as plastic 35 mm film cans are used. One contains a piece of gold padded with cotton wool. The others contain equal weights of lead, similarly padded. Usually the test starts with a confirmation that the conditions are suitable for dowsing. The subject is told which can contains the gold and demonstrates his ability to detect it. This test can be repeated several times; the expected result is 100% success. The containers are then shuffled, out of sight of the subject and the witnesses. Now all the dowser has to do is to repeat the former test. The only difference is that now he doesn't know which is the correct target unless his dowsing ability tells him. Of course he might hit on the gold by accident even with no dowsing ability. If there are five targets even a giftless dowser has a 20% chance of scoring a hit. That's why the test must be repeated many times. The guesser will continue to get about one hit in five attempts, the real dowser should do much better.

Regrettably, it is necessary for the test to be designed to make cheating ineffective. Thus we must switch the gold from can to can between tests, just in case there is some way of distinguishing the correct can from external marks. It is also important that no one in the test room knows which is the correct target, in case they give unconscious clues to the dowser. This doesn't mean that the dowser intends to cheat; people can use helpful information without knowing they are doing so. The test protocol must eliminate any possibility of such information being available.
If we carry out the test twice, the chance of getting hits on both tests by accident is 0.2 times 0.2 or 4%. That's still quite high. To beat the one in a million level would require nine successful tests. Still, our 100% accurate dowser should be able to do this in an hour, and walk off a million dollars richer.

So how do we test the less confident dowser, the one who claims a 50% success rate? We must increase the number of tests until the cumulative probability of guessing drops below the one in a million level. The easiest way of doing this is to increase the number of cans used in each test. If we used only two cans, both the random guesser and the 50% successful dowser would show an identical success rate. With five cans the difference between the guesser's 20% success rate and the dowser's 50% rate is not great. By using 100 cans in each test we would reduce the guesser's success rate to 1%, and the dowser should still be right half the time.

This test must still be repeated many times. We have to pick the number of tests such that there is some number of total hits which the real dowser should beat nearly all the time but which the random guesser will achieve only once in a million trials. For example, repeating this test twenty times would give the real dowser an average score of ten. In practice his actual score can vary over a wide range. However, it can be shown that he will score six or better on 47 trials out of 48 and five or better on 168 trials out of 169. If we pick a threshold score in the region of five or six hits, we are not too likely to fail a real 50% accurate dowser. The random guesser would get one hit in about 11% of such trials but has a rapidly falling probability of getting more than one hit. His probability of getting four or more hits is one in 731,101 trials, not far short of our one in a million criterion. The guesser will hit five or more times about once in 29 million trials. Thus taking five hits in twenty tests as our threshold virtually eliminates chance success and is still fair to our 50% accurate dowser.

I've gone on at some length with this example to show that it is possible, with some calculation, to devise a fair test even for abilities which don't work every time. I'm sure James Randi has access to better statisticians than I and is just as able (and willing) to design fair tests for intermittent abilities.

©2000, Tom Napier

Okay, here's the Vitruvian Man Puzzle solution . . .

The reaction to this was very heavy, much to our delight. I said at the very beginning that it was a trick question. A number of people not only solved that, but came up with a complete set of three answers - for there are three correct answers!

The original question use the word "revolution." Many persons didn't notice that, and assumed that it was congruent with "rotation." Not so. Webster's Dictionary seems just a bit loose on the differences between these words, but technically a revolution consists of a body turning around another one, as when the Earth revolves about the Sun, once every year. The word "rotate" refers to a body turning around on its own axis, illustrated by the manner in which the Earth spins about on its own axis once a day. If we accept these strict definitions, then the simple answer to our inquiry, is "one revolution." The dizzy little man goes through one revolution, while rotating at the same time. (A few readers pointed out that our Moon, though from our point of view, always presents the same face to us and thus appears not to rotate, actually does rotate once every month. Did that ever occur to you?)

But how many rotations does he make? Well, the diagrams that accompany this explanation should make it quite clear. It's a matter of relativity. To us, the spectators looking down on this strange drama, the Vitruvian Man starts out at the top of the larger wheel with his feet down and his head up - from our point of view. One-third of the way (120°) around the larger wheel, we see that his feet are now "down" and his head is now "up" again, and two-thirds of the way (240°) through his revolution, we see the same circumstance again. When he is again "standing upright," he has completed three rotations, relative to us.

However, consider it from the point of view of the long-suffering chap himself. At the top, starting his voyage, he notes that his feet are adjacent to the larger wheel. He begins to spin - to rotate - and when he again notices his feet adjacent to the larger wheel, he is halfway through his revolution. When his same position relative to the wheel occurs again, he has completed one revolution, and is firmly convinced that he has rotated only twice.
In summary, the Vitruvian Man makes only one revolution, we see him make three rotations - about his own navel! - and he himself thinks he made two rotations. As you see, it is a matter of strict definitions and of relativity - but not Relativity.
We sincerely thank all of you who took an interest in this problem. And, you may well be ready for another. This one was suggested to us by correspondence, and it's much less involved. (But NEXT week, we'll really strain your wetwear! Start thinking about wine glasses....)

Problem: consider the following algebraic expression . . .

(a-x)×(b-x)×(c-x)×(d-x) . . . . .×(z-x)=

(All the letters of the alphabet are used, all 26.)

Question: what is the most concise way in which we can express this product? You can use any mathematical notation you wish, but try to make it transferable/translatable by our differing computer systems and/or software.

BTW: Thanks to the folks who made us aware of the SNAFU in our system last week that mis-identified the previous week's archive file as, "htnl" rather than as "html." Those of you who may have been looking for former puzzle information, might have been confounded by this.

I am now assured by my clever Web-person that this has been fixed and you can go back and look at the original material now in "Randi's Opinion Archive" (or you can just click here).