G. ADAM wrote:It should be roughly correct if the hit ratio is 25% - 50%, which is what we're both expecting.
If the hit ratio is 50%-99% then it will skew the ODDS higher, but if that's the case the paranormal bias is so high the amount of trials is fairly small and exponential growth kicks in on the ODDS.
There is no 50-99% in this protocol. It's either 0/4, 1/4, 2/4 or 4/4
A/ there is no contract, so the data can be sought before the stats
B/ there is no set ODDS to demonstrate, 100:1? 1000:1? 10000:1?
C/ we don't know the hit ratio or what the minimum should be
D/ the exact stats are unwieldy
E/ I don't even know if the protocol gets the channels in the right order yet
F/ Using an 'ordering' stats, I don't know whether to make 4/4 or 2/4 the pass mark
G/ If you use 2/4 as the pass mark, you will need 40 batches not 10
H/ It's a simple case of adding a few batches to overcompensate for the odds
It's like getting out your protractor to divy up the apple pie!
Just add up the hits![/quote]
You can't just add up the hits. Because this isn't a ganzfeld type protocol you've set up where there is one target and three controls and the odds at chance are 1/4. Here, you're given all the words, and asked to sort them. This is apples and oranges. You set out the odds:
G. ADAM wrote:The stats of guessing the order of 4 words.
There are 4! or 4X3X2X1 = 24 ways to order the words
1234 - 4 right
1243 - 2 right
1324 - 2 right
1342 - 1 right
1423 - 1 right
1432 - 2 right
2134 - 2 right
2143 - 0 right
2314 - 1 right
2341 - 0 right
2413 - 0 right
2431 - 1 right
3124 - 1 right
3142 - 0 right
3214 - 2 right
3241 - 1 right
3412 - 0 right
3421 - 0 right
4123 - 0 right
4132 - 1 right
4213 - 1 right
4231 - 2 right
4312 - 0 right
4321 - 0 right
0 right = 9
1 right = 8
2 right = 6
3 right = 0
4 right = 1
This above is fine, you lose yourself below:
Average is (8 + 12 + 4) / 24 = 1
2+ right = 7/24 or 3.4:1 ODDS (less than 1 chance in 3)
4 right = 24:1 ODDS (1 chance in 24)
Let's make the PASS CLAIM
AVERAGE SCORE OVER MANY TESTS >= 2/4.
So I will try to get ON AVERAGE, 2 right every test repeatable.
Now I have just have to work out how many trials to break 1000:1.
It will be roughly 3_/1000 or 10 trials.
Sound about right? 10 trials and I average 2 right every test to pass 1000:1.
I'll get the exact number of trials later.[/quote]
This doesn't seem to be the correct way to go about it at all. Here are how the odds break down:
At chance, the expectation is as follows:
0 right = 9 - 9/24 = 0.375
1 right = 8 - 8/24 = 0.333
2 right = 6 - 6/24 = 0.25
3 right = 0 - N/A
4 right = 1 - 1/24 = 0.0417
total = 0.9997 (doesn't = 1 due to rounding)
So you guessed one trial 1/4 (33.3% chance) and another at 2/4 (25% chance).
If you combine the stats and do 3/8 that would be 37.5% which obviously can't be right. What you want to test is are you getting these stats at a rate that is significantly better than expectation. I don't know exactly how to calculate that, but I do know that you just can't add them together.
For obvious reasons, you don't want to just consider getting a 2/4 to be significant since this is going to happen very often. What you want to figure out is whether you are getting 2/4 significantly more often than chance. If you accept the above, I'm willing to go and post on a probability forum to see if anyone has any suggestions on how to figure out what significant stats would be here and what sample size would be needed.
It is complicated, which is why you might want to consider going back to a protocol where you've got to guess the real word from the fakes. Then we can add up the hits and it will probably be much easier to figure out what a significant result would be.
You want to figure out if you've got a special ability right? You've been trying to establish this for 10 years. Why not do it right?