Dean Radin - harmless pseudo-psientist.

Ivor the Engineer said:
1) Leakage in experiment 1, which produced one of the lowest p-values, from the photo being loaded from disk at the start of the trial period.
Click to expand...


Experiment 3 had the lowest p-value. This experiment selected the photos immediately before the photos were displayed, i.e., after the SCR analysis period. If you look at table 2, Radin has calculated a p-value and size for experiments 2-4 combined. Therefore, I would say that sensory leakage is not a reasonable explanation.

2) When the experiment was run with tighter protocols (experiments 2 and 4), the difference gets much smaller, as indicated by the much larger p-values.
Click to expand...

See experiment 3. If you're arguing for a systematic reduction in results when potential disk noise is controlled for then there is no case for that. On the other hand, we could argue that the results are simply unreliable because 2 out of 4 experiments failed significance. However, the combined results are significant. Combining the results makes sense to me because otherwise we could simply cherry pick either the postive or negative experiments depending on our intent.

3) What was the purpose of using a 2:1 ratio of calm to emotional photos in experiment 3, other than to enable participants to learn calm photos were twice as likely to appear than emotional photos* during the experiment?
Click to expand...

Most probably, habituation to emotional photos during the course of the experiment.

IMO, the correct way to run the experiment is to load into memory one photo from a randomised calm list and one photo from a randomised emotional list at the start of each trial (t=-5s). Then, at t=0, a randomly generated binary value (i.e. 0 or 1) is used to select which one of these pre-loaded images is displayed. I would have thought removing the PC hardware from the room, leaving only the display, physiological monitoring equipment and a push button would have be a simple measure to avoid a source of possible leakage.
Click to expand...

Good idea. But, I think Radin's reason for moving the image selection time point was two fold. One was to get rid of the leakage potential and the other was to make sure the image was selected in the future to ensure that the experiment was "precognitive" rather than "clairvoyant", and perhaps see if the effect went away if you did that. Therefore, it wouldn't be necessary to move the equipment.

BTW, re-seeding the random number generator on every trial with the PC timer reduces the "randomness" of the output considerably, from about 4 billion cycles before repeating to a mere 86 million. Probably not important in this experiment, but it's been used to win at online poker.
Click to expand...

Perhaps note 2 addresses this? -

"To demonstrate that the QuickBasic 4.5 PRNG used in this experiment did not introduce an artifactual relationship between the physiological state and the resulting target photos, the PRNG was seeded with a number (N=1), and then used to generate one random number from 1 to 150, using the same programming code as employed in the experiment. This created a seednumber, target-number pair. This process was repeated for seed-numbers N=2 to 5000, and then a correlation was determined between the resulting pairs. If the seed-number was associated with the resulting target selection, then a positive correlation would be predicted. But no such relationship was found, r = 0.0007, p = 0.96, two-tailed."


I can think of an unlikely alternate hypothesis which is consistent with the evidence and does not tear-up well established physics: The participants manage to sub-consciously predict the output of the random number generator, not perfectly, but well enough to have better than chance results during the experiment.
Click to expand...

Its hard to argue against that interpretation if you have to insist upon it. The design of the experiment is supposed to rule it out as an explanation by using "adequate" randomisation. I assume you are arguing that the results show the randomisation was not adequate enough to rule out that explanation.

To counter that argument, the results of the experiments also suggest that as the emotionality rating of the targets increases, the pre-stimulus response difference also increases. In your interpretation, this would mean that the participants were also able to sub-consciously predict to what degree the targets were emotional, in addition to a binary emotional vs calm prediction.

Can you think of any ways to experimentally distinguish between your sub-conscious prediction hypothesis and the presentiment hypothesis?


BTW, as you pointed out, a gamblers fallacy mechanism cannot explain the results. So when you say that participants sub-consciously predicted the PRNG pattern, they must have used some other mechanism. What could it have been?
 
davidsmith73 said:
BTW, as you pointed out, a gamblers fallacy mechanism cannot explain the results. So when you say that participants sub-consciously predicted the PRNG pattern, they must have used some other mechanism. What could it have been?
Click to expand...

Go ahead and say it, David.

You know you want to.
 
Has anyone come across how the prior trial was taken into account in studies with different numbers of calm and emotional pictures? The rate of change in SCL (as it decreases to baseline) is steeper after an emotional trial and if this rate of change persists until the next trial, then it will lead to a divergence in the SCL. Any inbalance in the number of emotional to calm trials will lead this steeper rate of decrease to be sorted preferentially into the trials with the larger number of pictures - in this case, the calm trials (e.g. Experiment 3 in this paper). I realize that the data is interpreted as an increase in the pre-stimulus SCL in emotional trials, but it can equally be interpreted as a decrease in the pre-stimulus SCL in calm trials.

I have a vague recollection that I read something that looked at this issue. Has anyone come across this?

Linda
 
davidsmith73 said:
Experiment 3 had the lowest p-value. This experiment selected the photos immediately before the photos were displayed, i.e., after the SCR analysis period. If you look at table 2, Radin has calculated a p-value and size for experiments 2-4 combined. Therefore, I would say that sensory leakage is not a reasonable explanation.
Click to expand...

I agree for experiments 2-4, but in experiment 1 it is a definite possibility.

See experiment 3. If you're arguing for a systematic reduction in results when potential disk noise is controlled for then there is no case for that. On the other hand, we could argue that the results are simply unreliable because 2 out of 4 experiments failed significance. However, the combined results are significant. Combining the results makes sense to me because otherwise we could simply cherry pick either the postive or negative experiments depending on our intent.
Click to expand...

All the experiments produced results which support a real effect. Combining studies which use different equipment and protocols is questionable.

Most probably, habituation to emotional photos during the course of the experiment.
Click to expand...

Having twice as many calm as emotional pictures increases the expected run-length of calm pictures. I think this may be important.

Good idea. But, I think Radin's reason for moving the image selection time point was two fold. One was to get rid of the leakage potential and the other was to make sure the image was selected in the future to ensure that the experiment was "precognitive" rather than "clairvoyant", and perhaps see if the effect went away if you did that. Therefore, it wouldn't be necessary to move the equipment.
Click to expand...

The setup I described would still test for the presentiment effect because the actual picture which will be displayed is not decided upon until the last instant. Perhaps a clairvoyant’s skin conductance would oscillate before the picture was displayed under this protocol?:)

Another advantage I see with this approach is it removes the delay caused by accessing the disk before displaying the picture.

Perhaps note 2 addresses this? -

"To demonstrate that the QuickBasic 4.5 PRNG used in this experiment did not introduce an artifactual relationship between the physiological state and the resulting target photos, the PRNG was seeded with a number (N=1), and then used to generate one random number from 1 to 150, using the same programming code as employed in the experiment. This created a seednumber, target-number pair. This process was repeated for seed-numbers N=2 to 5000, and then a correlation was determined between the resulting pairs. If the seed-number was associated with the resulting target selection, then a positive correlation would be predicted. But no such relationship was found, r = 0.0007, p = 0.96, two-tailed."
Click to expand...

That only tests for correlation between the seed-number and the target-number. There could be higher order correlations between target-numbers.

Its hard to argue against that interpretation if you have to insist upon it. The design of the experiment is supposed to rule it out as an explanation by using "adequate" randomisation. I assume you are arguing that the results show the randomisation was not adequate enough to rule out that explanation.
Click to expand...

I’m not insisting on any one interpretation since I don’t have enough information to decide between them. My main point is there are an infinite number of hypotheses which can explain the results, many of which do not require the laws of physics to be rewritten. When choosing hypotheses, why does Radin consistently pick ones which already have a large amount of evidence against them when his experiments do not rule out less controversial hypotheses?

To counter that argument, the results of the experiments also suggest that as the emotionality rating of the targets increases, the pre-stimulus response difference also increases. In your interpretation, this would mean that the participants were also able to sub-consciously predict to what degree the targets were emotional, in addition to a binary emotional vs calm prediction.

Can you think of any ways to experimentally distinguish between your sub-conscious prediction hypothesis and the presentiment hypothesis?
Click to expand...

Yes. Check the randomisation. I don’t think prediction is the sole factor at play, BTW.

BTW, as you pointed out, a gamblers fallacy mechanism cannot explain the results. So when you say that participants sub-consciously predicted the PRNG pattern, they must have used some other mechanism. What could it have been?
Click to expand...

The gambler’s fallacy could not explain the results if it was the only factor. My guess is it is an artifact of the randomisation, the physiology of the measurement and a differential rate of change of skin conductance after a trial, depending on whether a picture was calm or emotional.
 
Last edited:
For those who have not read them yet, you will ant to refer to Ersby's posts on this subject (Ganzfield/Radin). This issue last popped up seriously some 3-4 years ago (maybe more), and Ersby gave an analysis of the experiments and Radin's analysis that was extremely fair, extremely thorough, and extremely well done.

If you want to save yourself some six months of work, you would do well to find those posts and follow his links.

N/A
 
Ivor the Engineer said:
I agree for experiments 2-4, but in experiment 1 it is a definite possibility.
Click to expand...

I agree that its a possibility in the sense that disk noise was not controlled for. But considering that experiment 3 got positive results when the disk noise was controlled for, I would doubt that it was responsible. It was more likely a real presentiment effect. In any case, the paper does not rest on the validity of that one experiment. And considering the number of published replications, I wouldn't worry about it.

All the experiments produced results which support a real effect. Combining studies which use different equipment and protocols is questionable.
Click to expand...

True, but the alternative would be to pronounce a real effect based on the two successful experiments (one if you choose to discount experiment 1) and discount the others. Personally, I wouldn't find it very conclusive having two experiments unaccounted for simply because they were unsuccessful. Combining the experiments gives us a better estimate of whether the two successful experiments were just due to chance. It's not perfect but its the best way I think.

The setup I described would still test for the presentiment effect because the actual picture which will be displayed is not decided upon until the last instant. Perhaps a clairvoyant’s skin conductance would oscillate before the picture was displayed under this protocol?:)
Click to expand...

I wasn't saying your protocol was no good. Its fine as far as I can see. But there was no need for Radin to take the hard disk into the back room because I guess Radin wanted the experiments 2-4 to be true precognition thereby accessing the disk just before the picture was presented but after the presentiment analysis period was over. You're suggestion is a good idea for a "clairvoyance" design though, which would be nice to compare with the precognition design

That only tests for correlation between the seed-number and the target-number. There could be higher order correlations between target-numbers.
Click to expand...

Is it reasonable that these higher order correlations could account for the results? A few dozen trials per participant doesn't seem much if we're talking about 86 million cycles before repetition. Or have I misunderstood?

Yes. Check the randomisation. I don’t think prediction is the sole factor at play, BTW.
Click to expand...

What aspect of the randomisation would you check?

To to clarify, both the presentiment hypothesis and your sub-conscious hypothesis say the participants are predicting the sequence of images, to some extent. You would say this is down to the participants computing some probability estimate of what the next image will be based on what has come before, yes?

The gambler’s fallacy could not explain the results if it was the only factor. My guess is it is an artifact of the randomisation, the physiology of the measurement and a differential rate of change of skin conductance after a trial, depending on whether a picture was calm or emotional.
Click to expand...


But wasn't the randomisation adequate? 86 million cycles before repetition? If the randomisation is adequate then you can't say the gambler's fallacy can explain the results. It seems that you're trying to spread the explanation around without actually establishing any of the individual parts.
 
fls said:
Has anyone come across how the prior trial was taken into account in studies with different numbers of calm and emotional pictures? The rate of change in SCL (as it decreases to baseline) is steeper after an emotional trial and if this rate of change persists until the next trial, then it will lead to a divergence in the SCL. Any inbalance in the number of emotional to calm trials will lead this steeper rate of decrease to be sorted preferentially into the trials with the larger number of pictures - in this case, the calm trials (e.g. Experiment 3 in this paper). I realize that the data is interpreted as an increase in the pre-stimulus SCL in emotional trials, but it can equally be interpreted as a decrease in the pre-stimulus SCL in calm trials.

I have a vague recollection that I read something that looked at this issue. Has anyone come across this?

Linda
Click to expand...


Is this what you're thinking of?

http://m0134.fmg.uva.nl/publications/2002/expectationbias_PA2002.pdf
 
davidsmith73 said:
Is this what you're thinking of?

http://m0134.fmg.uva.nl/publications/2002/expectationbias_PA2002.pdf
Click to expand...

No, I'm not talking about modeling an expectation effect. I'm referring to the normalization of the SCL and the clamping to zero of the normalized SCL at the start of the pre-stimulus period which has the effect of obscuring whether or not differences after the start were due to a continuation of differences that existed prior to the start (and are unrelated to the stimulus). It would be related to the detrending that was performed in the acoustic stimuli tests referenced earlier.

Linda
 
Last edited:
I've just been having another look at the audio stimulus version of the study I linked to earlier in post #209 #206.

Take a look at figure 4 on page 12. Doesn't this show the vast majority of pre-stimulus SCRs are inverted with respect to the detection criteria used for the proportions of SCR analysis?

Our definition of an SCR required that a minimum be followed by a maximum in the prestimulus region.

...

The sign of D was determined by the position of the maximum SC relative to the minimum SC. If the maximum occurred before the minimum, D was defined as negative. Figure 4 shows the distribution of D for all prestimulus regions in the pilot study.
Click to expand...
 
Last edited:
To answer my own question: No, it doesn't. D is calculated from the raw data.

Looking at the numbers, the presentiment effect is not large or consistent. On average each participant got less than one SCR in all the pre-stimulus periods during a particular trial. What is being claimed as remarkable is the proportion of SCR's before a control was lower than before a stimulus.

What effect would throwing away those results with less than six stimuli of either type have?

Since the protocol was not counterbalanced as to the number of audio versus control stimuli, we decided in advance to reject sessions with less than six stimuli of either type, to reduce the variance of the within session effect size.
Click to expand...

So there was always between 6 and 14 stimuli in the data analysed, thus a long quiet period would have to be followed by a series of stimuli, or the data would not pass through the filter. Those participants who were generating more SCR's after a long quiet period would tend to have a better chance of one occurring in a 3 second pre-stimulus period.
 
Last edited:
Radin does not solve the HD noise problem by shifting the process to "just before" the image is shown.
The subject, by association with file size or seek time, could still learn about the next photo.
The HD of experiment one would be slow and noisy by current standards, so the 'just before' will vary considerably according to file size, seek and access times (and programme design)
The sound is not only important because of leakage, but the effect upon the subject. The perception of the image (albeit late in the process) may be altered as the delay between the aural and visual stimulus varies.
According to Radin, the presentiment effect is dependent upon the novelty of the images, and in a later experiment uses sound to enhance this effect, so he should expect disk noise to be a problem from that standpoint too.
He also shifts the RNG, but still does not seem to consider that the subject may influence it, which in the light of his own work, is an error. :)
 
Last edited:
fls said:
No, I'm not talking about modeling an expectation effect. I'm referring to the normalization of the SCL and the clamping to zero of the normalized SCL at the start of the pre-stimulus period which has the effect of obscuring whether or not differences after the start were due to a continuation of differences that existed prior to the start (and are unrelated to the stimulus). It would be related to the detrending that was performed in the acoustic stimuli tests referenced earlier.

Linda
Click to expand...

I see now, sorry I didn't read your post properly. Let me just check if I've got this right.

The SCR will be huge after a subject views an emotional picture. This means that at the end of an emotional trial (I'm talking about Radin's design here) the rate of change of SCL may be greater than at the end of a calm trial and may continue into the start of the next trial. Assuming this happens, then when the first SCL of the next trial is clamped to zero, there will be a greater change in the SCL during the pre-stimulus period for trials succeeding an emotional trial compared to a calm trial. Is that right so far?

So, even if there are twice as many calm photos then I don't see how this would produce an artifact. The probability of a calm trial after viewing an emotional picture is the same as the probability of a calm trial after viewing a calm picture so the effect should cancel out shouldn't it?
 
humber said:
Radin does not solve the HD noise problem by shifting the process to "just before" the image is shown.
The subject, by association with file size or seek time, could still learn about the next photo.
Click to expand...

This is an argument about inadequate randomisation, surely. But the subjects could more easily learn about the next photo just from remembering what the previous images were. No need to listen to the disk noise!
 
davidsmith73 said:
This is an argument about inadequate randomisation, surely. But the subjects could more easily learn about the next photo just from remembering what the previous images were. No need to listen to the disk noise!
Click to expand...

Yes, I agree. But the error may be more subtle than that. If there is such an error, then the HD may transmit that information. Perhaps I am reading too much into this, but the randomisation is only tested at the RNG level, and not through to the actual picture displayed? The entiresystem as it were?

Perhaps the HD issue is unimportant, but it does go to the author's thoroughness, and one of many other examples where leakage has been given lip-service, but not eliminated.
Why not make the HD inaudible, and remove all doubt ?
 
davidsmith73 said:
This is an argument about inadequate randomisation, surely. But the subjects could more easily learn about the next photo just from remembering what the previous images were. No need to listen to the disk noise!
Click to expand...

But it certainly is a leak, right?
 
You must log in or register to reply here.

Back
Top Bottom