Why You Can’t Rely on Poker Copilot

It’s time for one of my periodical “why you can’t rely on Poker Copilot” articles.

When I have the chance to steal the blinds, I often find myself using Poker Copilot’s “Folded big blind to steal attempt” statistic for the player on the big blind. And each time I do this, the little mathematics pedant who resides in my brain nags me, “do you have enough data?” So I do I? Probably not.

Based on a load of data I have, on 9-player ring game tables, a player has an opportunity to defend the big blind against a blind steal attempt on roughly 1.26% of all hands played. On tournaments with 9-player tables, it is slightly higher, at 1.66%. That is, after you’ve played any given player 500 times, you are most likely to have seen 6 to 8 times how this player acts when he has the opportunity to defend the big blind against a blind steal attempt.

Eight times after 500 hands. Let’s assume that our opponent, Phil, in the long run (100,000 hands), defends the big blind exactly 50% of the time. What’s the likelihood that our data would show this after 500 hands against Phil? Roughly 71% of the time it would show between 35% and 65%. That is, 29% of all players that you’ve played about 500 hands against would show either a deceptively high or deceptively low value for “Folded big blind to steal attempt“.

So what’s the advice here? Always check the denominator before making an important decision based on Poker Copilot’s statistics. If the denominator is low, then the statistic is not reliable for making important decisions.