You are not logged in.
Post a reply
Topic review (newest first)
- bobbym
- 2013-01-29 21:46:36
To do better than what was done before requires more samples.
- ybot
- 2013-01-29 21:36:38
Soon, with the error %.
- bobbym
- 2013-01-29 21:32:34
How much sooner?
- ybot
- 2013-01-29 21:12:55
Ok. I want to find out the actual mean of numbers that we have small samples.
Supose the ratio is 1/34 for the first 1000 trials, 1/35 for the 2nd 1000 trials, 1/33 at the 3rd 1000 and 1/33 at the 4th 1000 trials.
I guess we would be able to know the actual mean when we have 20k.
But the quest is to infer it sooner.
What are the math tool that you use?
- bobbym
- 2013-01-29 21:02:37
That is the point! You have tons and I have nothing but one example. I do not think I can get much more out of one example.
What is it you want to do?
- ybot
- 2013-01-29 20:53:29
I have tons of data
- bobbym
- 2013-01-29 20:26:59
Hi;
Do you have the results of more runs?
- ybot
- 2013-01-29 19:24:05
As I lack of strong math knoledge I´m trying to undestand the event I withness
- bobbym
- 2013-01-29 13:25:14
Hi;
I am not following you here. What can I do a regression to? I have one piece of data, 48 / 1000.
Using that I guess that the mean is quite close to that and there is some mathematics to support that. To really nail the mean down you will need more experimental evidence, like another couple of runs, the more the better.
- ybot
- 2013-01-29 13:14:21
How could inferencial statistics help to know I want?
Can you forward a roulette example where you use the correlation, the regression, the least square method and the student distribution?
- bobbym
- 2013-01-02 01:01:00
Hi;
We finished 3 pages with no conclusions.
That is not correct. We covered the expected value of the experiment ( 48 / 1000 ), we used the central limit theorem to state that the wheel probably has an average close to ( 48 / 1000 ) and I explained the standard deviation.
I told you your expected profit with the data you gave me, 72 cents on the dollar.
We might start again.
What does 48/1000 tell?
Like any empirical experiment it can only provide evidence. Statistics you can say is designed to make mathematical sense out of data like yours.
What would you like to do next?
- ybot
- 2013-01-01 23:42:55
ybot wrote:
Hi
About french european roulette
the chance to hit es 1/37
supose we are looking for 3 standard deviation events
43 hits in 1000 trials is 3 st dev(we played 1 number)
1)what does reaching 3 st dev mean?
2)what is the difference in strentgh of hitting 76/2000, 170/5000 or 319/10000(they are all +3sd)
3)what´s the difference in PLAYING the 1000 2000 or whatever or watch some data where we you find 1 number with 3 st dev?
4)it is the same to reach 3 st dev for 1 number or 2 numbers(neighbors)?
5)having collectede data, you pick 4 numbers(isolated, not neighbors)) that their sum reaches 3 st dev. What is the difference with item 3) or if we actually play every spin?
I hope you undestood my questions
I believe they are hard to answer
Best regards
We might start again.
What does 48/1000 tell?
What are the predictions for the play of this number?
What are the chances to repeat 48/1000? (from 44 to 50/1000)
We finished 3 pages with no conclusions.
- bobbym
- 2013-01-01 02:03:27
That is what I told you. But 1000 trials is a lot larger than 30 trials. We can say that the sample ( the thousand trials ) mean is pretty close to the true mean. From what you have given it is the best estimate I can do.
- ybot
- 2013-01-01 01:53:30
But, it happened (in this example) in the first 1000 trials. The next 1000 trials it might hit 40 30 or 50, we don´t know it yet.
The real edge isn´t 72.8%, the sample is very short to have a conclusion
- bobbym
- 2013-01-01 01:39:07
Okay, for your 48 out of a 1000 model, provided that is a true estimate of the mean.
The expectation is
E = (952)(-1) + (48)(35) = 728
which means you earn 72 cents on every dollar wagered. A very lucrative deal since a Vegas wheel only earns 5.26 cents on every dollar.