quantumleap said:How do you feel about regression testing? That's my term for going back through a sample size to discover patterns that occur at a particular percentage (not to be confused with software regression testing). Let's say I buy a database of all NBA games played so far this season. Then I test for certain conditions such as spreads greater than +10 for home teams and I find that these situations hit greater than 55% of the time.

Is this valid modeling so that I can expect to blindly bet situations such as this? Or better yet, bet selectively by either applying additional conditions or just using my own gut feel.

I do believe there are situations in sports that fit this criteria. If one could find these situations I believe you would be successful.

I have done a LOT of this. The simple formula to see if a win % is significant is multiplying the win % by 1 minus the win %, dividing by the number of samples, and taking the square root of that.

(sqrt (0.55 * 0.45) / 2000 )

Your win % is 99.7% within 3 times this number up or down from your mean. Your win % is 95% within two times this number up or down.

So the thing to keep in mind in regression analysis is that you are able to test a lot of different theories, and 1 out of every 20 will actually be a set that looks to be 95% significant, but is really just stumbled upon because you've tried 20 things.

A rule to use is that if you start to look at something that you *know* will affect win %, 2 standard errors from the mean is enough to start betting it. If you just start testin things that you *think might* affect it, and happen upon something you should have at least 2.5 standard deviations before betting on it.