Percentage of Online Winners, Revisited Yet Again

[dfan]

I know this has been discussed many times before, but never seemed to reach any resolution. The threads tend to end with just a bunch of guesses. At least that is my impression after reading a number of recent threads on the subject. If I'm wrong I'm sure I will be corrected. Anyway, here goes my foray down this so far dead-end lane.

In the Software Forum there was a discussion of the fact that many people were finding that about 40% of the players in their PT db's were winners and what this meant. My final take on it was that the 40% was an accurate and reliable estimate, but unfortunately not of anything very interesting. It represented "the proportion of players at a typical online table that are winners for the short span of hands that appear in a typical PT database." Whoopdedoo, right?

But the question usually posed is very different: "What percentage of online players are winning players"?

And answering that is problematic for a few reasons, some conceptual, some logistical.

First problem is clearly defining the question. What is the definition of "winning player"? I would guess it would be someone who has played at least X number of hands and is + $ for lifetime online play. What should X be? 50K? 100K? Lets pick 50K.

For players with 50K hands we are set to go assuming we could get the data. But what about players who played fewer than 50k hands but quit because they were losing so much money. Shouldn't they be counted in the "losing player" category? And if so, to be fair, shouldn't players with less than 50k hands who quit a winner as "winning players"? But if that approach is followed "winner" no longer means long-term winner.

One solution is to redefine our question to: "What percentage of online players who stick with the game long enough to amass 50k hands are lifetime + $."

(For ease of discussion I'll refer now to Party players.)

If that is our definition we still need a random sample of a 1000 Party players to get an estimate that is +/- 3% with 95% confidence. Unfortunately we can't get that by observing tables because winning players are overrepresented there. If we could get Party's player list we could pick a random sample from it but we don't have that.

So how to proceed? I can only think of a few possibilities.

1. Somehow creating a "correction factor" that would adjust for the overrepresentation of of winning players at a typical tables. Anyone have any ideas on how to do that?

2. If Party ever accidentally publishes a semi-random list of players that could be used. How would that happen? Well it could result from some type of promotion. For example, you could argue that winning the bad beat jackpot is a pretty random event, so the list of winners would constitute a random list. Of course the problem is that the list would be very short and also biased in that it would include only the type of players who play in those games. Is there anything else that Party does that picks players on a close-to-random basis and reveals their names?

3. If a random list could be created then those players would have to be datamined for ideally 50k hands. Then count up the winners and divide by the total and you have your answer. But since 50k hands on a 1000 players is quite an undertaking, it might be possible to infer the answer from trends in the data. The change in % of players that are winners as hand count increases is likely a pretty well defined function. So you could plot a graph of hand count vs % win for the data you could get on these players and see where that graph leads to at 50k.

Of course Party could answer the question easily but I doubt they would be thrilled to do that. I can see them advertising "Come to Party and join the other X% that lose here."

So until the "non-random sample problem" can be solved, the question is not answerable and no amount of PT datamining will help. Anyone have any ideas or comments?