Spy vs. No Spy

One of the things I've been thinking about recently is proper use of data in shaping play against a specific opponent. On Foucault's recommendation, I'm right now reading Chen and Ankenman's Mathematics of Poker. One of the points they make is that players often overuse information they have about a certain player's tendencies. This is an artifact of two general biases, which reinforce each other. The first is that people are often overconfident about their predictions, not just about poker but about random trivia, number of jelly beans in a jar, etc.. The second is that people often "over-update", that is, weight new information too heavily relative to their prior beliefs.

Increasingly I've been wary of weighting Poker Tracker's "Aggression Factor" too heavily in making my decisions, especially with anything less than 5,000 hands on the player in question. There are a few reason for this. First, it takes AF quite a long time to converge, because you have to either bet, call, or raise for the statistic to change, whereas every hand you are dealt factors into PFR and VPIP. Second, the variety of situations after the flop is so wide that it's impossible for any one statistic to capture anything but a very rough generalization of your opponent's tendencies. Finally, the interpretation of AF is kind of ambiguous - a player who is very tight preflop with an AF of 3 is a very different animal from a loose player with an AF of 3. Also you don't know whether a low AF is a result of infrequent betting or frequent calling. Here's a spot where I almost made a bad fold because this player had a sub-0.50 AF over the 1200 hands I had on him:

Full Tilt Poker, $0.50/$1 NL Hold'em Cash Game, 8 Players
LeggoPoker.com - Hand History Converter

UTG+1: $133.35
MP1: $230.45
MP2: $18.50
CO: $151.95
Hero (BTN): $123.45
SB: $71.55
BB: $104.50
UTG: $100

Pre-Flop: dealt to Hero (BTN)
UTG folds, UTG+1 raises to $3, 3 folds, Hero calls $3, 2 folds

Flop: ($7.50) (2 Players)
UTG+1 checks, Hero bets $7, UTG+1 calls $7

Turn: ($21.50) (2 Players)
UTG+1 checks, Hero bets $15, UTG+1 calls $15

River: ($51.50) (2 Players)
UTG+1 checks, Hero bets $30, UTG+1 raises to $60, Hero calls $30

Results: $171.50 Pot ($3 Rake)
UTG+1 showed (two pair, Aces and Jacks) and LOST (-$85 NET)
Hero showed (three of a kind, Eights) and WON $168.50 (+$83.50 NET)

VPIP and PFR are more reliable, but even then, you DO see a 27/4 player showing up with AJo after an UTG+1 raise. Of course he can't be doing that every time he's dealt AJo UTG+1, but realize that you can't completely RULE OUT that hand. Even more noisy is just the summary of the last 17 hands, which is what Poker Spy gives you. Personally I think I overuse this information to such an extent (e.g., this guy hasn't played a hand in 2 rounds, his raise must really mean strength) that, along with the fact that it clogs up my screen, my computer's processor, and displays my net winnings for the session, which should be completely irrelevant to the correct play, that I'm better off not using it. I do think knowing what YOU have done over the last few hands is important for keeping track of your table image, but PAHud will display your total table stats for you, which give you a good indication of what your play at the table has been like. So I'll be trying some PokerSpy-less sessions over the next few days.

BRUECHIPS