Re: Inheriting parents' chart factors

<Stevepppp@[EMAIL PROTECTED]
> wrote in message
news:5bbc8e82-bb86-4fcd-9c1b-3e6109b56999@[EMAIL PROTECTED]
> On May 17, 7:29 pm, Ray Murphy <ray...@[EMAIL PROTECTED]
> wrote:
>
>> RM: Look I can argue until the cows come home if posters are in the
>> mood for it, but I'm not doing that in these threads. I'm simply
>> trying to ****ge you into being a bit more precise - now that we've
>> finally got two researchers in the same place at the same time. You
>> haven't actually got any statistical data in your favour until you
>> present it and it's better than 0.05 and it has passed minimal
>> scrutiny at least. Until researchers get to that point, standard
>> astrology is streets ahead - even with some of the dubious systems
>> that are used in some of that work.
>
> Ok, how's this for precision:
>
> bs%   #hits  hit%(of 1 million attempts)
> 25:%  0      0%
> 24:   1      0.0001
> 23:   2      0.0002
> 22:   4      0.0004
> 21:   27     0.0027
> 20:   52     0.0052
> 19:   100    0.01
> 18:   199    0.0199
> 17:   451    0.0451
> 16:   916    0.0916
> 15:   1697   0.1697
> 14:   3179   0.3179
> 13:   5405   0.5405
> 12:   9003   0.9003
> 11:   14364  1.4364
> 10:   21871  2.1871
>
> bs=bias swing
>
> average bias swing: 3.983329%
>
> These are hard core numbers from a program I wrote that calculates
> the
> number of percent hits of a coin flip. This table is very beneficial
> and im****tant in determining statistical proof of anything.
>
> Here's what those numbers mean: the sample pool was 100 coin flips.
> The maximum bias swing achieved in the pool flipped 1 million times
> is 24% above 50% = 26 heads & 74 tails (or vice versa). So for a 100
> sample pool, it is virtually impossible to have a random bias above
> 75%. The test would have to be doped. Look at 20% bias swing. Only
> 34 times out of 1 million did the swing exceed 20%! This is still 
> very
> tiny, tiny enough to be used in a court. The odds are very low that
> a  test will achieve a bias exceeding 70%. So 70% could be 
> practically
> used as a maximum swing to "nearly" prove something statistically
> exists such as astrology.
>
> This is a very helpful chart for statistics. If you want more
> precision, I could run the program a billion runs, but I already
> know the numbers will be extremely close to those above.
>
> I've also plotted sample pool sizes of 1000, 10,000, and 100,000 of
> which I will post later. When the sample pool size increases to
> 1000, the maximum bias swing dramatically decreases, nearly by
> ten fold!
>
> Here's the program I built to get these tables (in freebasic):
>
> dim bias as single
> dim abias as single
> dim mbias as single

[Program code snipped]

RM: Yes, that sort of thing is fine for showing how existing
statistical formulae is not perfect, but it's totally irrelevant if
the replication of an experiment doesn't hold up.

For example you could argue that a p-score for an experiment which was
calculated with Chi square or the G-test should NOT be p = 0.001  but
rather p = 0.0001, but it is all a waste of time if a subsequent
experiment doesn't yield statistically significant results - either by
your method or an existing method.

The only value in your figures is in obtaining confidence to pursue
the experiment, rather than abandoning it -- and we can all do that
without any fancy numbers at all.

For example if I said "More Aries join the military than we would
expect by chance" we don't need statistics at all. Ultimately we only
need percentages and ultimately we have to demonstrate what we are
saying, and we need to do it repeatedly. If we don't - it doesn't 
work.

Astrological research (of the scientific type) doesn't go anywhere
with verbal or mathematical gymnastics - either our own or our
competitors. We have to demonstrate that what we are saying holds
up with further testing.

[....]

Ray