James Randi Challenges Skeptiko’s Telepathic Dog

[Topher Cooper]

Originally Posted by Paul C. Anagnostopoulos

I know what the experiments were about. I just find them strangely restrictive.

~~ Paul

Your statements didn't seem consistent with that, so my misunderstanding was, I think understandable. You didn't say, for example, "why restrict your attention to looking at a key area as in the experiments under discussion?", but rather "What is the 'key region'? ... Could you elaborate on the bolded phrase? I can't parse it."

In any case, there is nothing particularly strange about it, it is a standard method for designing scientific experiments. A more general hypothesis to be tested is operationalized to a more restricted case under specific conditions based on observations (e.g., that in many reported cases the specific behavior was going to a window or door). Very general hypotheses ("they react in some, unspecified way") are hard to test, either requiring very large data sets or complex protocols (e.g., testing whether an observing family member is able to tell whether or not the person is returning, allowing them to base this on the behavior of the animal in question).

[Paul C. Anagnostopoulos]

Originally Posted by Topher

Ah! Got it. Supervision is a forbidden interpretation but superhearing is acceptable.

Indeed. To follow your lead, we know people can hear things far away, but they can't see through buildings and hills.

So, what you are saying is that solid evidence (for the acuity of hearing in dogs) should be ignored if it leads to a contradiction to your prior beliefs?

No, it should be taken into account quite seriously. Remember, I said that I don't think super-hearing or super-smell is the explanation. However, our alternatives are mind-reading dogs, dogs that can remote view, dogs that are prescient, etc. I just can't completely toss off the super-hearing thing, even with centuries of hearing research, in favor of these other completely nonunderstood choices.

This one is, for example, designed that there are "Dogs that Know" when their owners are coming home. It doesn't say that they read their owner's minds (though there is speculation about that), that they can literally "see" their owners leaving, or that there is a specific sensory apparatus that provides that knowledge.

From the video: "Tommies telepathic link with his owner ..." I guess we have to use the alternate definition of telepathic also.

Yes, you have been using it incorrectly, but that is not how psi is defined.

Okay then, I'll start using the term psi to refer to research involving interesting sensory experiments with unknown explanations. But as soon as I hear someone make a statement in which he is assuming a "paranormal" explanation for the experiment, I'm gonna holler. I'd say that "Dogs who Know" video deserves a loud shout.

~~ Paul

[Chris Noble]

Originally Posted by David Bailey

I rather hope I never "whine", but that is perhaps beside the point.

Your 6 minutes seems to have been picked almost out of a hat, and is uncomfortably close to the entire 10 minute period which characterises the entire phenomenon!

In one of the Jaytee trials the dog began "signalling" 30 minutes before the owner began the return journey. In one of the Kane trials the dog began "signalling" 30 minutes after the owner began the return journey. In both cases the dog spent a greater proportion of time "signalling" during the return journey than during the main period but the dog's behaviour did not have any obvious relationship with the supposed telepathic signal.

Instead of whining about how unfair I am being try to come up with some criteria that would really demonstrate that the dog's behaviour is realted to the telepathic signal.

Have you ever stopped to think what "completely unambiguous" means in the context of any scientific experiment that involves statistics?

The ambiguity I was referring to was the relationship between the criteria that Sheldrake used and a "psi" explanation. As I have pointed out if the dog simply starts signalling 10 minutes after the owner leaves and remains there until the owner returns it will spend a greater proportion of time there in the return period than in the main period. If you repeat this a few times you'll get a p<0.05. If you do this enough times you'll get p<0.0005.


Compare this to somebody who claims he can predict the rolling of a die. The null hypothesis is clear. The person should by chance get 1/6 right. There is a unambiguous relationship between rejecting the null hypothesis and providing evidence for the claim.

I rather think the JREF challenge is a distraction, I would prefer to think of this as being judged as a scientific experiment than by the standards of a magic trick!

If the phenomenon is so common and so obvious that 40% or more of dog owners think their dogs can telepathically read their thoughts then it should be possible to demonstrate this in an unambiguous manner. If you have to do hundred of trials and complicated statistical analysis to find evidence then it really raises the question of how the owners can be so sure in the first place. Wishful thinking and confirmation bias seem much more likely to explain the 40% or more of dog owners who think their pets are psychic.

I just feel that if you want to suggest a 6 min cut off because of the possibility of a dog hearing the approaching car, you should provide some evidence that this is really necessary. In any case, Alex is planning some tests in which someone else drives Jane's car home - this should settle the matter.

You've got this backwards again. If you are going to argue that all normal explanations have been controlled for then you have to demonstrate this by specifically testing them. Even better - explicitly controlling for them by having the owner return in different vehicles.

Obviously the dog hearing the vehicle cannot explain the Jaytee trials where the dog began "signalling" 10-30 minutes before the owner even started the car. I believe that dogs heasring cars does contribute to the 40% or more of dog owners who think their dogs are psychic.

All of this is getting away from my main point about the criteria that Sheldrake uses. If the owner starts home after being away for 120 minutes and the return journey lasts 60 minutes should the dog start "signalling" when the owner starts the journey or is +-50 minutes near enough?

[Chris Noble]

Originally Posted by Larry Boy

How about the suggestion I made earlier?



I just spotted a problem, though. If the dog gets up from his waiting spot during the longest waiting period and then returns, you will have to cut that period in two, and this may lead to the latter part being the longer period, in which case the trial is a failure even though it may have been a success in actuality.

Dog waiting
____________ _ _ _ _ _________ _____________
|10min|10min|10min|10min|10min|10min|10min|10min|

Owner returning
__________________________________________
|10min|10min|10min|10min|10min|10min|10min|10min|


Perhaps you need to count the longer periods as one if they're longer than, say, 5 minutes and the space between them isn't more than 1 minute, or something (this will have to be decided unambiguously of course, before the experiment is done). Then you use the start of the first "block" in this combined waiting period as the measure of success.

If Tommy's behaviour is as obvious as Alex suggests then there shouldn't be a problem. Don't worry about 10 minute intervals. Just look at the largest block of time that Tommy spends signalling and compare the start of this period with the start of the return journey.

If you do a large number of trials where the owner is absent for times varying from 1-6 hours there should be a correlation between the dog's signal and the owner's return time. You can calculate the probability of the results occurring by chance.

[davidsmith73]

Originally Posted by Chris Noble

In one of the Jaytee trials the dog began "signalling" 30 minutes before the owner began the return journey. In one of the Kane trials the dog began "signalling" 30 minutes after the owner began the return journey. In both cases the dog spent a greater proportion of time "signalling" during the return journey than during the main period but the dog's behaviour did not have any obvious relationship with the supposed telepathic signal.

Instead of whining about how unfair I am being try to come up with some criteria that would really demonstrate that the dog's behaviour is realted to the telepathic signal.

Why are you assuming that the dog's behaviour represents a phenomena that can be described as a 'signal' tied to the start of the return journey?

I don't see any reason to assume that.

It's better to make less assumptions about the processes going on here. Alex et al should first try to establish a correlation between the return time and waiting behaviour. Then you can investigate whether a 'start signal' exists or whether the phenomena can be better described by some other process.

We've already established why your suggested method of analysis is not good (too much noise).

The best way to test for the initial correlation (IMO) is to look at the total time spent waiting during the return journey as a proportion of the total waiting time. If you can't explain what is statistically wrong with this method (or the method that Larry Boy suggested) then I suggest we move on, because I'm sure that Alex et al don't agree with your theoretical assumptions.

[Chris Noble]

Originally Posted by davidsmith73

Why are you assuming that the dog's behaviour represents a phenomena that can be described as a 'signal' tied to the start of the return journey?

I am not assuming anything, merely reading what Sheldrake has written. In the initial paper on Jaytee figure 2 shows the apparent correlation between the time that Jaytee started "signalling" and the start of the return journey.

I don't see any reason to assume that.

It's better to make less assumptions about the processes going on here. Alex et al should first try to establish a correlation between the return time and waiting behaviour. Then you can investigate whether a 'start signal' exists or whether the phenomena can be better described by some other process.

We've already established why your suggested method of analysis is not good (too much noise).

Alex is already suggesting that the dog starts "signalling" within a few minutes of the owner (or other person) starting the return journey. If Alex's results are to be believed then there isn't any noise to speak of. The dog spends zero time "signalling" in the main period. If the dog really does this repeatedly then what are you worried about?

The best way to test for the initial correlation (IMO) is to look at the total time spent waiting during the return journey as a proportion of the total waiting time. If you can't explain what is statistically wrong with this method (or the method that Larry Boy suggested) then I suggest we move on, because I'm sure that Alex et al don't agree with your theoretical assumptions.

Perhap's I didn't make myself clear. I think Larry's suggestion was a good one. It gives a start time for the "signalling" that can be compared to the start time of the return journey. A regression analysis can then give the probability of the results occurring by chance.

Larry's criteria do a much better job of testing the claims than Sheldrake's

What statistical test are you going to use? What would be a statistically significant result?

Your criteria would give the same value if the dog started signalling at the start of a 60 minute return journey as if it started signalling in the last 10 minutes.

The main point of the exercise is for Alex to clarify exactly what his claim is and exactly how he is going to demonstrate it.

He seems to be under the misapprehension that he can simply use Sheldrake's methods and then hand the results over to some "experts" for analysis.

[davidsmith73]

Originally Posted by Chris Noble

Alex is already suggesting that the dog starts "signalling" within a few minutes of the owner (or other person) starting the return journey. If Alex's results are to be believed then there isn't any noise to speak of. The dog spends zero time "signalling" in the main period. If the dog really does this repeatedly then what are you worried about?

I would be worried about the possibility of including too much noise in the analysis if a 'signalling' method of analysis is used, as you've described. If the dog performs as well as the taped examples then there doesn't seem to be much noise, but why adopt a design that allows for it? If you adopt the methods you've suggested then you assume the phenomena works in a particular way and run the risk of introducing noise.

What statistical test are you going to use? What would be a statistically significant result?

I would compare the proportion-of-total-waiting-time means for the main and return periods, probably using a non-parametric method.

Your criteria would give the same value if the dog started signalling at the start of a 60 minute return journey as if it started signalling in the last 10 minutes.

Yes, but either way, the result would be positive and you don't have deliberate over whether to exclude 'false starts' from the analysis due to some distraction or other. So, I'm not assuming a temporal relationship between the start of the return journey and the start of the waiting behaviour. There may well be such a relationship, but there might not.

[David Bailey]

Originally Posted by davidsmith73

I would be worried about the possibility of including too much noise in the analysis if a 'signalling' method of analysis is used, as you've described. If the dog performs as well as the taped examples then there doesn't seem to be much noise, but why adopt a design that allows for it? If you adopt the methods you've suggested then you assume the phenomena works in a particular way and run the risk of introducing noise..

I agree - and furthermore, you really want a robust experiment so that you can explore other related issues without introducing noise. Also, this is an experiment with a real dog - it may have wind one day, or an itchy paw, there really is no sense in making the experiment noisier than it need be.

One approach might be to wait for Alex to produce the full set of results, and then, assuming they are positive, for Chris to devise a simple computer model to explain the results by conventional means. The model would presumably involve some anticipatory strategy of the dog, and he could obviously adjust all the parameters to try to fit the results - that would be a fair comparison, and if a (reasonable) conventional model would indeed fit the data, I for one would be interested.

David

[alextsakiris]

Originally Posted by davidsmith73

It's better to make less assumptions about the processes going on here. Alex et al should first try to establish a correlation between the return time and waiting behaviour. Then you can investigate whether a 'start signal' exists or whether the phenomena can be better described by some other process.

We've already established why your suggested method of analysis is not good (too much noise).

The best way to test for the initial correlation (IMO) is to look at the total time spent waiting during the return journey as a proportion of the total waiting time. If you can't explain what is statistically wrong with this method (or the method that Larry Boy suggested) then I suggest we move on, because I'm sure that Alex et al don't agree with your theoretical assumptions.

Exactly!... lotta over-thinking by dogmatic Skeptics.

This whole thing is really pretty simple:
1. run a lot of trials.
2. apply reasonable controls for routine and cueing
3. see if the dog waits more when the owner is coming home.

[Larry Boy]

Originally Posted by alextsakiris

Exactly!... lotta over-thinking by dogmatic Skeptics.

This whole thing is really pretty simple

I'm a little disappointed. I thought OpenSourceScience was about finding ways to design experiments both skeptics and proponents can agree about. Your idea of a "simple" approach apparently doesn't please the skeptics, at least not here on the forum, so I think you'll have to make some compromises. Why not test the method I suggested? Chris seems to agree that it is fair, and if the start of the waiting period corresponds to the start of the owner's return as much as you claim in the videos, then there shouldn't be a problem. If you furthermore divide the trial into large enough blocks, at say 20 minute intervals over a 4 hour period, there shouldn't be any problem at all with the kinds of noise we've been discussing here.