Oldskeptic wrote:http://www.sheldrake.org/Articles&Papers/papers/telepathy/pdf/Nolan.pdf

It's a short read so go ahead.

I have some problems with this paper. First of all it is misleading when it says that 850 trials being conducted because it doesn't mean that there were that many separate tests, it is that many phone calls.

Next is that this paper was published in

the Journal of the Society for Psychical Research A somewhat partisan publication with a peer review system that might not be as skeptical as it should be.

I wondered if they had a statistician/probability expert look at the results? I don't think so.

looking at this example of the Nolan sisters a different way than Sheldrake does the percentages are not at all surprising if you look at it as an either or situation.

The phone rings it's either Anne or it's not, 50/50 probability. With Anne the hit rate was 1out of 4, 1/2 of the probability.

With Denise it was 2 out of four, the expected hit rate.

With Linda it was 2 out of 2, something not to be unexpected in only two instances.

With Maureen it was 1 out of 2, again the expected probability.

So looking at it this way there is nothing odd going on here. We have one test below expected probability, one above expected probability, and two that are the expected probability.

I don't think that's a correct way to look at it.

For any 'guess' made by the recipient, it has a 25% chance of being correct, since a randomly-selected one of

four people could be calling. It's not a 50/50 situation simply because there are two possible outcomes.

The main failing to me in this case seems to be the tiny size of the experiment.

Only having 12 calls does mean that 'statistical significance' seems a tricky concept to apply.

Also, to me, looking at the guesses, they do look rather like someone trying at least in part, and quite possibly subconsciously to work by the 'law of averages'.

The second correct guess (#4) was a name which hadn't occurred in either the guesses or the callers

The third one (#6) was a name which was yet to occur in the callers.

The fourth (#9) was one which had yet to be guessed.

The fifth (#10) was one which at that point only been one out of the previous nine callers.

The sixth (#12) was one which was, at that point, the joint-least-guessed name.

Of the incorrect guesses, many of the Maureens seem to come when there has been a distinct lack of Maureens in the recent callers.

Now, a lot of that could certainly be put down to

me looking for patterns in the data, but I wonder how many of the choices might be being made for the 'wrong' reasons.

Having a rather longer trial would seem better not merely from a statistical point of view, but also because it might make it less likely that people play by averages once well into the trial, when they would have heard all the names multiple times - something which, in the case where there

were some actual effect, could end up with people failing to listen to their inner voice when they really should - a short experiment actually seems like it could mask effects if effects actually existed, with the human temptation to choose as if there were patterns in random data.

In fact, bearing that in mind, if I was designing an experiment like this, I think I would, unknown to both the recipient and the callers, set things up so that on some calls (maybe 1 in 5 or so) the callers were *not* randomly-selected, but were selected to even out the sequence locally and make it

seem more 'human-random', those calls then being excluded when analyzing the data.

Oldskeptic wrote:There are some surprising and not so surprising things about statistics and probabilities. Not so surprising is that they are easily manipulated. The surprising thing is that they sometimes confound all logic and intuition.

One example is how many people would have to be in the same room with you to have something close to 100% probability that one of them has your same birth date, month and day? Logic and common sense says 365, but it turns out that there is a 99% probability at only 57 people. In a room of 23 people there is a 50% probability that someone will have your same birth date.

Leaving aside the 'small group shared birthday thing, as already mentioned, I'm not sure how often common sense would say '365' for the number of people needed to

definitely match

my birthday, though I think that it would be more likely to find 'common sense ' suggesting something like 183 people needed to have a 50% chance of a match, even though that is also incorrect.

Ignoring Feb 29, and assuming an even spread of births through the year, to have a 50% probability that someone in a group has a

specific birthday would take a significantly larger group - I make it 253 people, assuming I have my maths correct.

To be 99% sure someone shares my birthday, I make it 1679 people needed.

That would be explicable by saying it's like having a 365-sided die, and repeatedly rolling it, and seeing how long it takes to come up with a particular number matching my 'birthday' (for the sake of argument, the number '1')

A single throw has a 364/365 probability of not being '1'

For 'N' multiple throws, the combined probability of none of them being '1' is (364/365) to the power N

It might seem from 'common sense' that roughly 365/2 (182 or 183) 'throws' should be needed to give a ~50% chance of a hit, but a

different application of 'common sense' shows that can't be right.

Imagine someone has thrown the die 183 times and written down the results without telling me.

My birthday isn't anything 'special', so if I picked a day at random, what are the odds of it matching any of the numbers written down?

Now, clearly, it is likely that if someone did throw the die many times, they would be virtually certain to get a decent amount of duplication. For example, imagine the throws towards the end of their 183. If by some miracle duplication hadn't already happened, every throw they'd have a close to 50% chance of duplicating a number they'd already got (this is the same effect that kicks in with the 'duplicate birthday in a small group' example)

What that means is that if the die is thrown and written down 183 times, there will almost certainly be quite a few fewer than 183 different numbers thrown, and so a rather less than 50% chance that any randomly-chosen number (ie any particular birthday) will match one of those thrown.

To have 183 different numbers thrown, one needs to throw the die 183 times, and then enough extra throws to make up for any duplications that happen in the entire sequence.

I don't do sarcasm smileys, but someone as bright as you has probably figured that out already.