Thursday, June 25, 2009

Some more preliminary analysis.

Although I will be more particular in the future (most likely dividing things by percentiles), I initially just split my sample into those who have large (admirable) win rates, are above even, below even, or are losing chips at a swift rate. I chose to look at box plots- which show the range of the data, the median, and the 25th an 75th percentile.

Here are some charts. I apologize for the format within the post. I'm in a hurry. But click on them and they should explain what they represent.




















Posted by at

5 comments:

  1. AnonymousJuly 28, 2009 at 9:03 AM

    Hi Kirk

    Isn't it interesting that the big winner and big losers have the same preflop raise %. Doesn't that suggest that pre-flop strategy is pretty standard and widely known?

    Also, since everyone knows the 'correct' strategy, deviation from it can be disasterous ( if they can also adjust).

    Mark ( marrek in the forums).

    ReplyDelete
  2. Kirk ThompsonJuly 28, 2009 at 2:25 PM

    A few thoughts:

    Initially I thought,

    "Yes! Preflop play is largely solved."

    Now I think this,

    "A solid preflop game does not spare you the variance of postflop play over a sasmple even as great as 10k hands."

    Why?

    *The median winrate of the 10k hands crew is 10bb/100 higher than the entire sample set.

    *The biggest loser among those with >10k hands is the player who doesn't play this way.

    *I haven't parsed out short stackers and players with glaring postflop play errors.

    I think I will be analyzing preflop strategy by position with a greater sample of players (I doubt a cutoff of 100 hands/position would introduce too many errors)

    ReplyDelete
  3. UK RookieAugust 1, 2009 at 2:36 AM

    Hey Kirk, Marrek,

    It's chp03djw from the forums. I'm finally back online after my move to the UK. Thought I'd check out how this analysis is going.

    I suspect the 'biggest losers' preflop raise distribution is two tailed, where the 'biggest winners' distribution is going to be closer to normal.

    Plotting these distributions as histograms might give a bit more information.

    I'm just surprised how little variation there is in the distributions between the groups, for many of the variables.

    Rather than analysing the preflops stats by position, perhaps you can make better use of your data by coming up with a positional awareness index, maybe BT-VPIP / VPIP.

    ReplyDelete
  4. Kirk ThompsonAugust 4, 2009 at 5:01 PM

    I agree 100% that positional play must matter. I blurbed a little on this on stox (I'm trying to keep the "BLOG POSTS" to graphs and more formal analyses.

    As it turns out, there are some fairly intuitive indications when you look at covariance. I'm not 100% on whether or not my crude normalization is acceptable for analyzing covariance, but it's obvious that a great many of the "Golden Rules" of good poker jump right off the page.

    *Isolation raises in position, steals, and raise-or-fold play in reraised pots all have positive covariance with BB/100.

    *"Calling off" at or near the commitment threshold (facing a 3- or 4-bet) preflop, VPIP vs. EP raisers, VPIP in the blinds all have negative covariance with BB/100.

    I want to stress that the average winrate for this gang is positive. Many of these folks have only limited grasp of the fundamentals and only a few spewtrons are present in this group*, but they are good enough to win. The dead money is from players who put in far fewer than 10k hands over the course of a month. These folks are by and large trying to play well and as a sample, represent the winners.

    *Just me and a couple other spewtrons, that is :).

    ReplyDelete
  5. Kirk ThompsonAugust 4, 2009 at 5:03 PM

    Here is the post on stox.

    The strongest positive covariance with BB/100 in this sample is W$@SD. Or, as I like to call it, "running good".

    ReplyDelete

Subscribe to: Post Comments (Atom)