>If psi exists we have not the slightest
>idea how it works and so we're unjustified in throwing data out just because
>it doesn't work. Birthdays, astrological charts, and interesting patterns on
>play slips are as legitimate as anything else. There is a insidious trap in
>science that should be feared, unconsciously "correcting" the raw data until
>the desired results are obtained.
I think you have misunderstood what I'm saying here. *Of course* I didn't
`throw out' birthdays, etc. But I knew in advance from many decades of
psych tests on college kds etc, and from previous psi studies, that
populations don't free-respond literally *at random*. In the 1930s, Zenith
Radio in the USA ran a long series of ESP tests that asked people to mark a
sequence of binary choices and mail them in. They got many tens of
thousands of answers, When these were compiled and analysed (rather
witlessly, but psych stats was young then), they announced to a stunned
world that ESP had been proved beyond doubt. There was a quite startling
correspondence between the collective answers and the targets.
Then someone smarter noticed what is now obvious - that in almost every
week's responses, there had been a joint tendency for people to generate
the sequence 11211 (I think it was). If the choice was Black/White, people
en masse tended to choose White White Black White White, say. When all
this stuff was screened for such templates and biases, and the randomly
generated target arrays matched up, it turned out that by chance there had
been slightly rather more target sequences compatible with that format than
not. Worse, subtle cueing meant that people were sometimes led to start
their guesses with one target rather than the other, which then entrained
the rest of the sequence. And so on. At a stroke, the evidence for ESP in
the Zenith data vanished, to relieved guffaws from the skeptics.
Forty years later I came along and went through it again, using a rather
denuded data base (the original data had been lost, and all that's on
record is the total number of guesses for each week and the standard
deviation scores for each target, from which one can reconstruct a rude
profile of the true guesses. When one edits out the underlying and
repeated 11211 pattern, something quite interesting emerges. It is now
obvious that something *was* going on. Put it another way - people will
always collectively choose White in the first place, but the vote is two
percent higher (say) when White *was* the target than when it wasn't.
I have never published these results, because they are post hoc and too
grainy to be truly interesting. But it showed me that when I came to look
at Lotto, I shouldn't go searching for, say, the 6 largest votes out of 45
each draw, taking that as the majority vote. If I'd done that, I'd have
got pretty much the same 6 each week, because people collectively select 7
1.209 times as often as they might if there were no bias, 13 is chosen
1.215 times too often, 20 scores 1.247 higher; while 41 only gets 0.634
times as much of the vote as it ought, 43 gets 0.784, and so on. All this
would wash out automatically if I'd had a very large number of draws to
work with, but in fact in my pool of data some numbers came up as winners,
by chance, more often than others. (The numbers produced by the game
randomiser do *not* correlate with the population preferences, so that
stands as powerful evidence *against* any collective psychokinetic or
mind-over-matter effects in such games.)
The answer is obvious: don't compare 7's vote with 41's, but 7-when-it-wins
with 7-when-it-loses.
That's all I was talking about. No data or animals were harmed in the
course of this experiment.
Damien Broderick