How many games makes a W-L record statistically significant?

[ImtheFranchise]

Ever since I've started picking NBA spreads and totals, I am 161-127 (55.9%) over 288 bets.
I read the section in Wong's Sharp's Sports but dunno if i completely understood it - is this enough of a trial period to conclude that I beat the baseline 52.5% for picks?

[Sinister Cat]

This thread may help: http://forum.sbrforum.com/players-ta...r-success.html

[reno cool]

If you show over 5% edge on 500+ even money games that starts to get significant. Your record in itself is not yet good enough to conclude you're a winning player to a degree of certainty I would consider important.

[ImtheFranchise]

Ok, so even if i ran at 56% over 500 bets it wouldnt be a confident enough trail?
Sinister, link does not work.

[reno cool]

500 bets at 56% is 280-220

280*2.1=588
500*1.1=550

you would have bet 550units and returned 588. thats 106.9% return or edge of 6.9%. If you can do this I would guess something more than luck is involved.

However, using past record alone is kind of dangerous. It would help if you had good reasoning behind your strategy also.

[Peep]

"Past results are no guarantee of future results" is the crux of the problem.

I would just keep betting, if you are winning money, great.

I am assuming you got these results going forward. If you are backfitting this whole system you are in deep dodo.

[Sinister Cat]

Sinister, link does not work.

Okay I fixed that link. Using the method in that thread, assuming everything is -110, I got around an 88% chance that you are profitable going forward.

[Ganchrow]

Okay I fixed that link. Using the method in that thread, assuming everything is -110, I got around an 88% chance that you are profitable going forward.

Spot on results, but I'd phrase the conclusion differently.

There's a roughly 1-88%=12% probability that a handicapper no better than breakeven would see results of this magnitude or better.

Anyway, here's how the numbers would breakdown:

Obtaining approximate results in Excel using the methodology outlined above:

Z = (161/1.1 - 127) / sqrt(288*10/11) ≈ 1.1967
p(z < Z) = NORMSDIST(1.1967) ≈ 88.43%

Obtaining exact results in Excel using the binomial distribution:

p = BINOMDIST(161-1,288,52.381%,1) ≈ 87.25%

Of course, just as Peep says, if you're backfitting and/or looking at some nonrandom subset of your actual betting history then these results would be effectively meaningless.

[hajune]

Ever since I've started picking NBA spreads and totals, I am 161-127 (55.9%) over 288 bets.
I read the section in Wong's Sharp's Sports but dunno if i completely understood it - is this enough of a trial period to conclude that I beat the baseline 52.5% for picks?

The answer is probably yes. Bookmakers sometimes adjust their spreads if a series of people key in to a type of handicapping method many are using. Thereis much to consider , but you probably are a long term winner using whatever you are doing.

[jtuck]

Had been wondering the same thing myself. Thanks guys

[reno cool]

The answer is probably yes. Bookmakers sometimes adjust their spreads if a series of people key in to a type of handicapping method many are using. Thereis much to consider , but you probably are a long term winner using whatever you are doing.

I disagree. If you play regularly you are very likely to have such a run as just a coin flip bettor.(Especially if you change strategies until you go on a good run.) No doubt it's a positive sign. Put 2-3 such runs together in a row before getting excited.