Difference btwn. 30+3 and 50+5

[bball904]

[ QUOTE ]
I'd say I'm 99% confident the numbers are accurate...you got any rationale behind your skepticism or just 'feel'?


[/ QUOTE ]

OK, pshreck was making his claim on feel. He does have experience that suggests his feel is right. Is his feel right, or should we be more inclined to trust the numbers in this case? I do have some background to discuss this beyond the feel level and I certainly have an opinion here, so I will share.

[ QUOTE ]
However, one persons results can be just about anything if you have enough people out there doing something.


[/ QUOTE ]

Exactly! Statistics are a funny animal. Many posters have lived periods that fall well outside the 1st or 99th percentile of any confidence intervals we can calculate. I certainly fall into that category. For example, if I were a 20% ROI player at 55's, I would have a 99% confidence (quoting your Monte Carlo simulation) that my results over 50 sng's would be no higher than 76.4%. Well, I am certainly not a 20% player (17.0% over 930 55's), but I do have a stretch of 50 sng's with a 83.6% ROI. On the flip side, still assuming I'm a 20% player (sorry, I don't have a chart for my 17% skill level), I could be 99% confident that over 200 sng's, the worst I could expect would be <font color="red"> (6.4%) </font> ROI. Unfortunately, I have managed to have a string of 200 sng's with an ROI of <font color="red"> (13.6%) </font> .

How can this be? One player, over the course of only 930 tournaments, has broken the model on both the positive and negative side.

The reason is quite simply really. All the discussion of "meaningful" statistical evaluations that are regularly conducted on this forum need to be taken with a grain of salt.

Are your Monte Carlo simulations scientifically credible? No! For that to be the case, we'd need to be looking at data that comes from independent random trials. One table tournaments at Party Poker certainly do not meet that criteria. In addition, your null model, which assumes an even distribution of 1st, 2nd and 3rd place finishes, further distorts the analysis. I would bet the original poster had a much higher distribution of 1st place finishes for him to have a 36% ROI. I'm quite sure that there are very few players that have a playing style such that their ITM distrubions stay normalized like that over time.

The bottom line is that running a Monte Carlo simulation on sit-and-go poker tournaments is not a valid statistical analysis. I did find it quite interesting, as I do with most of the statistical discussions on this board, but please do not quote the results as scientifically credible in the future.

What does all this mean? Well, for one thing, I agree very much with pshreck's feel regarding anyone making any sort of conclusion based on 200 sng's.

[Irieguy]

Nice post.

I find Monte Carlo simulations quite helpful for developing some perspective on what to expect... but they are certainly not the end-all, be-all. I also have lived below the 1st and above the 99th percentile with my results. What is often overlooked when discussing the likelihood of certain outcomes is the massive number of trials we are all generating. If I had to pull the trigger of a 100-chamber gun knowing there was only one bullet in it... I could point it at my head and feel confident that I would survive. But if I had to pull the trigger once a minute for one year, I wouldn't much like my chances.

Pokerscott is not necessarily wrong (the evidence can't be wrong, it's just evidence)... it's just that pshreck's comments are more accurate from a practical standpoint. You don't know a thing after 200 SNGs. Not a single, worthwhile thing.

Irieguy

[Pokerscott]

[ QUOTE ]
What is often overlooked when discussing the likelihood of certain outcomes is the massive number of trials we are all generating. If I had to pull the trigger of a 100-chamber gun knowing there was only one bullet in it... I could point it at my head and feel confident that I would survive. But if I had to pull the trigger once a minute for one year, I wouldn't much like my chances.


[/ QUOTE ]

That is a great analogy. If you play 2500 SnGs you have played 25 seperate 100 SnG sessions. That is like pulling the trigger 25 times. Sooner or later you are going to get something unusual. In fact, if you cherry pick the best and worst 100 SnG stretch in a 2500 stretch I bet the extremes you mention are not at all uncommon.

However if you really are a 10% ROI player, then the next 100 SnGs will likely fall in the range I gave (with 90% confidence [img]/images/graemlins/wink.gif[/img] )

Pokerscott