I'm testing a betting system's effectiveness using Chi squared. How big does my sample need to be relative to the average odds, for the result to be meaningful?
Thanks
Jeff
Chi squared
-
CaerMyrddin
- Posts: 1271
- Joined: Mon Sep 07, 2009 10:47 am
Every result will be meaningful 
When you run such a test it will come out with some kind of metrics about it's reliability, like a significance level or a standard deviation (I must admit I can't remember now tbh). It's up to you to define what you want to work with...
When you run such a test it will come out with some kind of metrics about it's reliability, like a significance level or a standard deviation (I must admit I can't remember now tbh). It's up to you to define what you want to work with...
Thanks Antonio.
To take an extreme example, if I'm betting on 50/1 horses and one wins in the first 10 races, it doesn't prove much. But if I'm betting on 50/1 horses and 100 win in the first 1,000 races, I imagine it would be statistically significant!
The average odds in one of my samples is 11.5. How big a sample do I need before I can say 'The chances of this Chi Squared result not being replicated over a much larger sample are very, very small'?
Jeff
To take an extreme example, if I'm betting on 50/1 horses and one wins in the first 10 races, it doesn't prove much. But if I'm betting on 50/1 horses and 100 win in the first 1,000 races, I imagine it would be statistically significant!
The average odds in one of my samples is 11.5. How big a sample do I need before I can say 'The chances of this Chi Squared result not being replicated over a much larger sample are very, very small'?
Jeff
-
CaerMyrddin
- Posts: 1271
- Joined: Mon Sep 07, 2009 10:47 am
You are right.
tbh, I can't remember how to do those figures, although I have a vague memory that there were some tables with those figures plotted out? but I would suggest you would do differently. Run the test on your sample and look at the incertainty measures it comes with. Run the test again with the sample 'doubled' for instance and analyse again. If you had this new sample, was it enough for you? Do it again if needed.. This should give you a pretty accurate number about the sample size you need...
tbh, I can't remember how to do those figures, although I have a vague memory that there were some tables with those figures plotted out? but I would suggest you would do differently. Run the test on your sample and look at the incertainty measures it comes with. Run the test again with the sample 'doubled' for instance and analyse again. If you had this new sample, was it enough for you? Do it again if needed.. This should give you a pretty accurate number about the sample size you need...
That would be helpful, if anyone knows where I can find this table.CaerMyrddin wrote:tbh, I can't remember how to do those figures, although I have a vague memory that there were some tables with those figures plotted out?
I've tried searching on Google for an answer to this question, but everything I've come across is written with people doing maths degrees in mind, and is unintelligible to me!
Thanks Antonio.CaerMyrddin wrote:Run the test again with the sample 'doubled' for instance and analyse again. If you had this new sample, was it enough for you? Do it again if needed.. This should give you a pretty accurate number about the sample size you need...
The challenge when analysing betting results IMHO is that you will get long losing runs and long winning runs - that's statistically inevitable. So what I need to try to establish is how likely it is that my sample is typical.
Jeff