I have been playing around with the database at
http://b2b.sportsdatabase.com/. I think I have something that may be a
workable system but I need to know if there is a way to statistically
measure whether this is a dead end or not. The criteria is situational
in nature (home vs away, etc, as opposed to batting average formulas
and other complex data). There is 3 criteria that must be met. That's
it. Clean and simple. Includes the entire season, all teams, including
playoffs. No handicapping or date sorting. No odds filtering. When the
three criteria are met, bet to win $100 if the team is an underdog or bet to
win $100 if the team is a favorite. The results:
2004 season: 59.9% winners total profit $3355
2005 season: 57.5% winners total profit $1830
2006 season (so far): 62.2% winners total profit $950
Total games played: 339 winners 236 losers 59.0% overall win rate
My concern is there is no data available prior to 2004. Is this enough
to say this pattern is not a fluke? If you take my question, I'd like
to see the math behind your reasoning please. Thanks for reading!

The statistical formula for the error is sqrt(p*(1-p)/n), where
p=339/(339+236) and n=339+236 gives an error or approximately 2%. That
looks good, but there is a warning. Because you arrived at your method
from data, you overfit your data. That means this data produced a
profit with this system, but you need to do out of sample testing. Fix
your rules and look at new data not used to develop the rules. Very
likely your 59% win rate will drop. How much no one knows.