Sample Size

    • TheJap
      TheJap
      Bronze
      Joined: 10.08.2010 Posts: 6
      Hi,

      I'm wondering what people consider a good sample size of hands when making a read off data, I play MTTs mostly so don't have 100s of hands on players, only a few on the table I'm currently at, so what do you think would be a large enough sample ? Thanks.
  • 2 replies
    • purplefizz
      purplefizz
      Bronze
      Joined: 12.03.2008 Posts: 4,508
      hi TheJap,

      i will move your question to the MTT strategy discussion section so that it can receive more attention from other MTT players.

      smiles,
      wendy
    • pzhon
      pzhon
      Bronze
      Joined: 17.06.2010 Posts: 1,151
      Some stats converge rapidly, while some take much longer to converge. VPIP/PFR converge relatively rapidly. 3-bet, 4-bet, fold to 3-bet, check-raise etc. stats converge much more slowly and you often will not have enough hands for these stats to be useful in a tournament.

      An upper bound for the standard deviation of a binary statistic (something that happens 0 or 1 times per hand, like PFR) is 50%/Sqrt(# hands). It's a little lower for statistics which are closer to 0% or to 100%. So, for example, if you see someone has a VPIP of 30% after 100 hands, that could easily be +- 1 standard deviation, or about 30% + 50%/10 = 35% or 30% - 50%/10 or 25%. It's unusual for this to be off by more than 2 standard deviations.

      You don't get a chance to fold to a 3-bet every hand, particularly if you rarely raise. For a rough estimate, divide by the square root of the number of opportunities you estimate the player had in the sample instead of the number of hands.

      Be careful that some statistics group together dissimilar situations. Some players may steal from the button 40% of the time in general, but 80% of the time when you are in the big blind if their HUDs says you don't defend the big blind.