4-tabling party 20+2 NL SnG

[for teh win]

Here we go again.

The thing about the 2+2 forums is no one will ever be allowed to say "I am a profitable player at limits so-and-so" unless they are a high-profile poster, because there will always be the usual crowd of "Uh-uh, your sample size is too small" posters. The sample size can indeed be too small in a particular case, but I find that in most cases the nay-sayers are wrong.

The ROI numbers in this thread are not meaningless due to sample size. They are as good as they get. And of course his ROI will drop over time if he stays at $20+2, but for reasons that may not be all that apparent.

First, let's look at the sample size in itself. 300 SNGs. That should be enough to run basically any non-parametric statistical test in the book. But let's make it simple and just look at ITM% (I know this thread is not about ITM%). A high ITM% doesn't tell you that much, but is nevertheless intuitively an indirect prerequisite for a high ROI. The null-hypothesis - a completely random string of SNGs with equal chance of ending up in any spot - would give him an ITM% of exactly 30%. Out of 300 SNGs that would mean 90 games ITM and 210 outside, which you can test against his actual ITM%. In terms of sample size, that's plenty. It really is. Many sciences have to do with less in pushing theory forward and handing out PhD's left and right.

So that is the sample size issue per se. It really is a non-issue.

I can agree it would be nice with a sample size of, say, 1000 for ROI testing, but there are also diminishing returns to getting larger sample sizes. Why? Because there is the consistency issue. What if his data isn't consistent over time? What if data was gathered from mainly games against bozos on holiday leave during Christmas? He can win against those but can he win against the future average type of $20+2 player? Well, what many people fail to realize is that this is as much an argument against a large sample size as it is for.

Nobody will be able to beat the stock market for all eternity. There will never be a marathon runner that will stay no 1 for all eternity (for obvious reasons). Who knows what $20+2 will be like in the future with stiffening competition and new books on SNG's flooding the market? And that is precisely the problem. There is no static population of poker players and games to draw reliable samples from.

A large data sample in poker can only be retrieved by investing time. Now, there are only 24 hrs in a day, so a larger data set means more months playing. Thus the longer the sample period, the more changes to the underlying population will be incorporated in the data set, making the latest data entries far more reliable in terms of predictive power. Hence only a subset of data is good.

Many games used to be softer a year ago or two, meaning old data is useless because it is not representative for the average game texture anymore. And you can only get so much data from the last few months unless you are 10-tabling 24/7.

300 is plenty. Does it give you guarantees? Of course not. Nothing will. It can only give indications. But what is going to save you when moving up in limits will never be win rate statistics but sound bankroll management. The decision to move up in limits should never come easy of course. But look at the younger tournament stars of today. If they would have followed standard 2+2 advice they would still be stuck playing $10+1, working on gathering a data set large enough to satisfy the people here.

Sorry, I just had to be the contrarian here.

[YourFoxyGrandma]

My ROI at $20+2 is 91% after 3 tournaments.

[byronkincaid]

Is that all? You need to go back down to the $10s...

[Irieguy]

Very good post. I disagree with almost all of it, but you bring up some great points and caused me to think about some things.

[ QUOTE ]
but I find that in most cases the nay-sayers are wrong.

[/ QUOTE ]

On this forum the nay-sayers tend to be quite smart, obsessively dedicated to poker theory and data analysis, and very, very, very frequently right. I've been listening for a few years, and contributing only recently, but I owe a lot to the many thoughtful posters who take the time to explain why most people are completely wrong. The concept that most people are totally wrong... and the ability to become one of the few who are not, is a profoundly important one in a zero-sum field.
[ QUOTE ]

The ROI numbers in this thread are not meaningless due to sample size. They are as good as they get.

[/ QUOTE ]

I see your point, but it's still quite wrong. It's true that it is what it is, but it just isn't worth talking about. NFL teams don't have parades for being undefeated if they win their first game of the season. It's true that they are undefeated... but it's meaningless, so nobody even uses the term. ROI discussions are similar. The number itself means nothing until you attach the confidence intervals, and for some reason nobody ever does (even though Aleo has done all the work for us in this regard.) I received a PM from somebody last night who gave me their ROI numbers for their last 500 SNGs and they included their confidence intervals and I can't tell you how much more respect for that player I gained.

[ QUOTE ]
300 SNGs. That should be enough to run basically any non-parametric statistical test in the book.

[/ QUOTE ]

The mere ability to run a test has nothing to do with the utility of the test's results.

[ QUOTE ]
Out of 300 SNGs that would mean 90 games ITM and 210 outside, which you can test against his actual ITM%. In terms of sample size, that's plenty. It really is.

[/ QUOTE ]

It depends on how certain you want to be. If you don't mind knowing your ITM +/- 8%, then yes, it's plenty.

[ QUOTE ]
So that is the sample size issue per se. It really is a non-issue.

[/ QUOTE ]

I would suggest that a failure to understand the significance of sample size implications is the single most common reason why people have psychlogical breakdowns when they are running badly. It's far from a non-issue in my opinion. In terms of the mental game of SNG poker, it may be the most important issue. That's why it's discussed so often.


[ QUOTE ]

I can agree it would be nice with a sample size of, say, 1000 for ROI testing, but there are also diminishing returns to getting larger sample sizes. Why? Because there is the consistency issue. What if his data isn't consistent over time?

[/ QUOTE ]

Fantastic point. That's one I struggle with a lot. But i'm beginning to realize that the game conditions and statistical expectations are more stable than I once believed.

[ QUOTE ]

A large data sample in poker can only be retrieved by investing time. Now, there are only 24 hrs in a day, so a larger data set means more months playing. Thus the longer the sample period, the more changes to the underlying population will be incorporated in the data set, making the latest data entries far more reliable in terms of predictive power.

[/ QUOTE ]

Again, a very good point. But I'm not sure the conclusions you start to draw at the end are as meaningfull as you think. I could be wrong about this.

[ QUOTE ]
300 is plenty. Does it give you guarantees? Of course not. Nothing will. It can only give indications. But what is going to save you when moving up in limits will never be win rate statistics but sound bankroll management. The decision to move up in limits should never come easy of course. But look at the younger tournament stars of today. If they would have followed standard 2+2 advice they would still be stuck playing $10+1, working on gathering a data set large enough to satisfy the people here.

[/ QUOTE ]

Best point of your post. Thank you for an interesting read.

Irieguy

[adanthar]

I half agree, and think that you and Irieguy are talking about two sides of more or less the same issue.

The emphasis on gigantic sample sizes that no one in their right mind would actually gather is definitely too pronounced here - but a good part of this is because most new posters have giant holes in their game.

It's not that anyone should really pony up for 1000 10+1's before going up to the rare heights that is the 95% identical 20+2 level; it's that even some of the better posters here that are beating the 10+1's soundly still post questions like 'if he showed me AK and I have QQ, shouldn't I fold because it's too early?' or 'if I have a set on an A82 board with 2 spades and my opponent called a big PF raise, should I bet big because the flush draw might hit?' and...well, they just miss the point. (No offense, folks. I went through the same thing less than a year ago; everyone does.) Until you get all of that more or less basic stuff down to a routine you shouldn't move up, and for most people that does take a few hundred SNG's to do.

It's not that anyone really cares about whether you've played a hundred SNG's or a thousand. Because I've been one tabling for most of the time I've been playing, my number of SNG's at all levels is surprisingly low (probably in the 700-800 range), but I know I'm a winning player (albeit with holes I have to fix before I can do something like quad the $200's.) But the difference is that I have a good grasp of the theory involved, and when I make a mistake I can feel it immediately afterwards. Some random guy that wins a bunch of his first 100 $10's and decides to hit the 30's isn't going to be able to do that. I know; I was that guy.