"Hermes" <hermes@[EMAIL PROTECTED]
> wrote in message
news:hermes-E1BF23.14221001052008@[EMAIL PROTECTED]
> : : On May 1, 9:08 am, "Ray Murphy" <ray...@[EMAIL PROTECTED]
> wrote:
> : : > Astrology Statistics for dummies:
[....]
> : The experimental situation for star signs is actually
> : more like tossing a dodecahedron dice with one of the
> : signs on each of the 12 surfaces 2911 times.
> :
> : At birth you have a choice of 12 signs not just 2.
> : A choice of two is gender. :)
>
> I simulated this and actually the probability distributions
> are pretty similar for the two things.
>
> http://www.exactphilosophy.net/stat/
>
> Actually I plot 4 graphs (also described on my site):
>
> ++ A
> ++ * Toss a coin 500 times, count heads and tails
> ++ * Repeat this and count how many times the same
> ++ difference {head count - tail count} occurs
> ++ * Plot the counts for each difference
> ++
> ++ C
> ++ * Toss a coin 500 times, count heads
> ++ * Toss a coin 500 times, count tails
> ++ * Repeat this and count how many times the same
> ++ difference {head count - tail count} occurs
> ++ * Plot the counts for each difference
> ++ * Curve C is thinner than A - I guess by a factor
> ++ of square root of 2, since in case A the distance
> ++ between one toss and the expected value is always
> ++ doubled and in case C it is statistically doubled,
> ++ i.e. doubled by sqrt(2), which gives an overall
> ++ ratio of 2/sqrt(2) = sqrt(2)
> ++
> ++ B
> ++ * Toss a dodecahedron ("zodiac") dice 3000 times
> ++ * Count numbers that occur separately
> ++ * Pick any two numbers between 1 and 12
> ++ * Repeat this and count how many times the same
> ++ difference {count for number 1 - count for
> ++ number 2} occurs
> ++ * Plot the counts for each difference
> ++ * Practically the same width as curve A
> ++ * I leave it as an excercise to the reader to
> ++ calculate why (i.e. haven't done it :)
> ++
> ++ D
> ++ * Toss a dodecahedron ("zodiac") dice 3000 times
> ++ * Count numbers that occur separately
> ++ * Pick two numbers so that difference is maximal
> ++ * Repeat this and count how many times the same
> ++ difference {count for number 1 - count for
> ++ number 2} occurs
> ++ * Plot the counts for each difference
> ++ * Two peaks since obviously zero difference is
> ++ very improbable due to the selection process
> ++ * Number at top right is percentage of chance
> ++ that difference is >68; typically around 9%
RM: This online Java program shows us how random distributions can
occur. It's a good idea to change the settings to view it in different
ways.
http://www.fourmilab.ch/rpkp/experiments/pipeorgan/
For anyone who is unfamiliar with statistics, that moving graph will
not tell them much, but they need to know that the thing researchers
are all looking for, is a graph (from their research work) that looks
different to the "bell curve" shape of those movong graphs.
If it DOES look different their p-score will be below 0.05.
So if an astrologer says "My observations for Aries gave me a
p-score of p < 0.05" - then we know immediately that their
observation may be worth following up because the .05 tells us
that such a finding can only occur about 1 in 20 times we look.
If an astrologer finds a result that gives p = 0.03 then we're
supposed to say -- "Neat"
If the astrologer says they found p = 0.01" we're suppoded to
say "Heeey!"
If we find p = 0.001 we're supposed to say "Let's all have a
look at this!"
> )o+
Ray
|
