Significance

[Joe]

I've written a little program which some might find useful for testing systems, or just as a spin generator. You can download it for free here :

link:://:.roulettecoder.com/utilities.html

Feedback welcome. 

[Firefox]

Some great resources, well done!

[Joe]

Thanks Firefox. I'm an admirer of your common sense posts, and it's nice to see someone here who actually understands the relevant maths. 

[Steve]

Good stuff joe. The problem though is some people still think 10 spins is significant. Their logic is "a win is a win".

[Maui13]

Awesome stuff on your site! Thanks man 

[Joe]

Good stuff joe. The problem though is some people still think 10 spins is significant. Their logic is "a win is a win".

Thanks Steve. The program doesn't directly address the issue of sample size, but I'll make a note of it in the accompanying blurb, and maybe I should write a blog post about it.

BTW, I've temporarily removed the program because I found a minor bug, and also I'm adding code which will let the user save a file of results and load it again when restarting the program. This means that you can test systems over a longer period.

I'll post here again when I've done it. 

[Joe]

Awesome stuff on your site! Thanks man 

Thanks! Not ALL the software will be free, but most will be. 
I've neglected the site for a while but have more free time now and will get cracking on it.

[Maui13]

If stuff works, I'll pay for it - simple

Keep the faith and thanks for your free stuff!

[Joe]

I've now finished the mods to the software which now lets you save the results and load previously saved files so you can continue testing particular systems from one session to another.

link:://:.roulettecoder.com/utilities.html (you might have to refresh your browser to see the changes).

If you want to know a bit more about the test which I've used then see here :

link:s://en.wikipedia.org/wiki/Binomial_test

There's a calculator for it here :

link:s://:.socscistatistics.com/tests/binomial/default2.aspx

Imagine you want to find out whether you are psychic, so you toss a coin 1000 times, and each time predict whether it will come up heads or tails. You get it right 733 times, which is a lot higher than the 500 times you'd expect by chance. The question is - what's the probability that you'd get a result as extreme as 733 purely by chance?

So you could use my program to develop and test your precognitive abilities, assuming you believe that there are such things. I know there are several threads here which discuss it. 

Anyway have fun and please let me know if you find any bugs, as I haven't done a huge amount of testing.

[Firefox]

Imagine you want to find out whether you are psychic, so you toss a coin 1000 times, and each time predict whether it will come up heads or tails. You get it right 733 times, which is a lot higher than the 500 times you'd expect by chance. The question is - what's the probability that you'd get a result as extreme as 733 purely by chance?

That's quite easy to calculate without software.

Standard deviation = SQRT  (npq)=

SQRT (1000x0.5x0.5) = SQRT (250) =15.8

Binomial approximates well to normal distribution over 1000 trials, so 733 is massive.

233/15.8 = 14.74 standard deviations positive.

That's 100% . Just 3 SD is 99%.

14.74 SD is off the scale of tabulation of the normal distribution.

[Joe]

That's quite easy to calculate without software.

Ya, but since this is a roulette systems forum I wanted to keep it more intuitive, and I would guess not that many members are conversant with standard deviations and other stats terminology. The significance expressed as a probability is easier to understand IMO. The logic is quite simple and is common to all classical tests.

If H is the hypothesis (in this case, that the system doesn't work) and D is the data (the number of wins relative to losses), then the logic is :

If H is true, then it is unlikely that D or more extreme data would be observed.
D was observed.
Therefore it is unlikely that H is true.

BTW, the program doesn't need fixed probabilities because it calculates the average probability every spin and uses this in the cumulative distribution function (binomial distribution, I'm not using the Normal approximation). So you could be betting on different amounts of numbers each spin.

[Firefox]

That's good if you use the binomial as the normal distribution only fits well over a larger number of trials.

Still, I'd hope even a statistically challenged psychic with 733 right would realise they were onto something!  Will have to ask Miles what he'd get 

[precogmiles]

That's good if you use the binomial as the normal distribution only fits well over a larger number of trials.

Still, I'd hope even a statistically challenged psychic with 733 right would realise they were onto something!  Will have to ask Miles what he'd get 

I would get around 800 and above. 

[Joe]

I would get around 800 and above.

Any proof of that? 

[6th-sense]

Love the calculator joe great job