My Allin Prelop Results - Statistically Relevant?

[Utah]

Given that I have played 1,000s of hours online and given that I have consistently won over that period, I am far from someone who believes that online sites have inaccurate or faulty shuffles.

However, I switched to pokerstars about 300 sngs ago and I have a return of about 30%, which is low for me. I noticed that I was losing a ton of favorable all-ins. So, I started tracking them.

Over the last 122 preflop allins:

When favorable: 78 times. Lost 38. Win Percentage: 54%
When Unfavorable: 43 times Lost 35. Win Percentage: 19%

Overall, I am 48wins and 73loses even though I was favored 64% of the time.

Is this statistically significant? How big does my sample have to be to indicate a concern?

Thanks

[NotMitch]

Without knowing how big of a favorite you were in the situations there is no way to know what this means if anything. And 122 isn't a large enough sample size anyway.

[Utah]

i can calculate that. However, I am more curious as to how large the sample has to be. Any ideas? 300? 1000? 10000?

[Tosh]

Statistically relevant in this case probably means more like tens of thousands of favourable/unfavourable hands.

[C M Burns]

how reliable your sample is is something you can calculate, I'd have to think a little more about exactly how though. First you would need the average amount you were favored and the number of hands, then u would compare your value to the expected. You can also put a value on how reliable it is. What you would be calculating is how likley is it that your result came from a distrobution where the expected result is the real average. (you could use a z-score) Anyhow you could put a number on how likley your results are chance, and calculate how many hands you would need to "prove" they were not chance. And of course the bigger the diference the less hands you would need to be conclusive.

It might be fun just to see and if a bunch of people got their hands together of allins head up pf, you could combine them. u would just need average amount fav/unfav, actualll win/loss, then number of hands. I beleive you could just average people numbers to combine them. Anyhow I may try looking at mine when i get a chance.

[Prickly Pete]

I'm assuming you're doing this by hand. Does anyone know if there's any way in Poker Tracker to track this sort of thing?

[XlgJoe]

[ QUOTE ]
When favorable: 78 times. Lost 38. Win Percentage: 54%
When Unfavorable: 43 times Lost 35. Win Percentage: 19%

[/ QUOTE ]

Just asking but wouldn't you expect the two win% to total 100%. Not that that helps answer your question, just wondering.

[PrayingMantis]

[ QUOTE ]
When favorable: 78 times. Lost 38. Win Percentage: 54%
When Unfavorable: 43 times Lost 35. Win Percentage: 19%

Overall, I am 48wins and 73loses even though I was favored 64% of the time.


[/ QUOTE ]

It was already mentioned, but I think this is important enough to repeat. If you don't take into account how much a favorite or a dog you where each all-in, these numbers mean absolutely nothing. It reminds me of a post by CrisBrown, a few months ago, where she complained about being the most unlucky on-line player, and showed some numbers from PT about all-in results (like you do here), to prove it. It looked pretty bad, and was replied by some 20+ posts, that told her she's just having a bad streak, until one poster pointed out she was getting exactly (or almost so, with that small sample) what she sould expact, if considering her advantage and disadvantage each time. She obviously didn't take it into calculation, like the rest of the posters who read her post.

[Utah]

Overall: 54% (chance of me winning)
Favorable Allins: 65%
Unfavorable Allins: 34%

To make is simple: Lets say I made a simple bet where I had a 54% percent chance of winning, what percentile of results would I fall under with placing the bet 122 times - i.e., "X percent of the time I should expect to see my results fall within this range"

My guess is that my results will not be that far out of whack.

[BigBaitsim (milo)]

You need more hands than this for statistical significance, given the normal and expected variance. My last 5 sessions, some 1K hands I've won almost every showdown situation in which I had serious doubts (i.e. two callers with QQ when the board showed an A and a 3-flush). I don't think I won a single one of these in the entire month of February.

[C M Burns]

Now that i've played with this idea some, i've found that you do not really nee that many hands to have a strong degree of statistical relevance. You should be able to anylise such a problem using the binomial distrobution. (I found a good site to do clacs http://ic.net/~jnbohr/java/CdfDemoMain.html) What you are basically doing is a series of "experiments" with a known expected outcome. The only problem is that your expected result changes with each trial, I don't know enough to figure how much of an effect this has.

But your problem is basically the same as the standard biased coin example, e.g how do u tell if it is biased. So lets say you toss it 100 times and get 75 heads is it biased. you can't really "prove" it only give a degree of likeyhood of the result and say any thing below a cirtain value is "proof". What we can calculate using the binomial distrobution is over and infinite set of 100 flips, what % would have 75 or more heads. you get a probability of far less than 1% (.99999 reversed). Generally in "science" anytime you get a result that is likley to have occured by chance less than 5% of the time it is reamed "significant" so this would be a very biased coin.

Now for the poker hands (you lost 38/78) if you were a 60% favorite on average you would expect this about 5% of the time, unlikley but considering how often these hands come up something you will see from time to time.

What if out of 100 hands where you were a .6 favorite, but only won .5, now you are out around 2% (out of all the sets of 100 of these hands)I'd say you would have to get bellow 1% to be suspicious. you could get this if you only won 50/100 hands where you were a .67 favorite, so not outside the realm of testing.

But again I am not quite sure how the fact that you are averaging over different expected probabilities factors in, and also you are selecting specific hands non randomly.

You could cirtatinly do this where you had the same expected win rate like a coin flip hand where you were the pair, with about a .52 chance to win. if out of 100 hands you are losing more than 60 you might want to be suspicious. for example you would expect to lose 65/100 1 out of 5000 sets of 100 of these hands, or at least 500000 hands of small pair head up vs. 2 overs.

well that enough thinking anyone who knows more feel free to correct me.

[Utah]

Cool. Thanks for the site. If I am using this correctly, it tells me that my results have a huge statistical relevance as the results fall well below 1 percent.

Results for favorable Allins: .009
Results for Unfavorable Allins: .15
Overall: 4.99E-4

My guess is that the actual probability of each bet would only sku the results slightly.

Also, I started tracking these results after a long series of improbable losses, so the real results of the last couple of hundred allins could even be worse.

[DaffyDuck]

Should you be concerned that you're a dog 1/3 of the time when you're all-in? I'd like to know if this is normal for a good player. I haven't tracked it so I don't know but this seems like too high of a percentage to be all-in when you're not favored.

Thanks.

[FloppedFlush]

Exactly. It not only depends whether you have an advantage or not, but also how big your advantage is.

If most of these hands in your sample were middle pocket pairs, then the numbers are pretty much exactly what you would expect. You're either a small favorite (against two overcards, winning about 54% of the time) or a big underdog (against a bigger pocket pair, winning about 19% of the time).

[Utah]

That is a good question. However, my results could be skewed by a lot of things - e.g. The number of times I am forced in with a small stack.

That being said, it would be important to add in the number of times I push in and dont get called. I will push in a lot to take the pot.