Information Overload

"Lack of information is not the typical problem in our decision processes (although lack of the right information sometimes is). The world is constantly drenching us with information through eyes and ears -- millions of bits per second. According to the best evidence, we can handle only about fifty. The limit is not information but our capacity to attend to it."
-- Herbert Simon

Poker used to be so simple. It only involved one table, a maximum of nine opponents, and not much more than 45 hands an hour. Online poker changed everything. Today, players must deal with multiple tables, hundreds of hands an hour, detailed up-to-date statistics, heads-up displays (HUDs), site lobbies, waiting lists, and the increasing necessity of monitoring a range of sites for game selection. It seems like the complexity is forever growing. As fish become less plentiful, professionals have to process a greater and greater amount of information to maximize EV. New software tools give professionals new ways of extracting edges, but they also often add to a player's information processing requirements. And where once there was an abundance of fish operating in plain sight, at mid-stakes and up you now have to search every under every nook and cranny to find them.

All this is surely enough to test the limits of the human brain. While much of this information is too valuable to ignore -- you don't want to inefficiently use your time because you aren't playing on all the sites available to you -- the information is also too much to handle in its raw form. You have to efficiently process the signal from the noise in order stand any chance of dealing with it all. The aim of this article is to explore some ways of coping with this overload by concentrating primarily on the most important information. It's all about finding the best way to budget your limited endowment of attention.

HUDs are one of the unique defining innovations of online poker, making this as good a place as any to start. Some players have a frightening array of numbers on their HUD, providing information on a number of highly specific situations. While more information is in an ideal world always better than less, this can only exacerbate the problem of information overload -- particularly if the stats you use take an unreliable number of hands to converge. This is true of many of the more specific postflop stats, where the effective sample sizes can be very small even if you have hundreds or thousands of hands on a player. This can very easily lead to incorrect decisions based on faulty data.

Some players, including a number of well-known high stakes players who generally face a small pool of potential opponents, choose to not use a HUD at all. I got comfortable playing without a HUD for a while, initially because I was playing on a site without HUD-compatibility, but it soon became a bad habit without much justification. I think that HUDs are absolutely mandatory for all players in today's games, if only to pinpoint the information needed in game selection. All but the highest stakes players will generally be facing player pools that are quite large, either because they're playing over multiple sites, or because their game has a lot of liquidity.

Try and imagine a theoretically perfect stat for game selection; what would it look like? An ideal stat would one that gives a perfect indication of whether a given table is worth sitting at. It would clearly indicate which tables are worth joining, which tables are worth staying at, and which tables are worth leaving. It wouldn't generate any false positives; it would never mistakenly direct you towards unprofitable games. Additionally, it shouldn't miss good games, either because it doesn't have sufficient data or because it fails to capture an aspect of bad play. This perfect game selection stat would allow you to easily find all the profitable games available to you, subject to the number of games you can play simultaneously, allowing you to focus almost all your attention on playing. It would greatly limit the amount of information that needs attending to, freeing up scarce cognitive resources.

Sample size is a really important issue when it comes to game selection stats, as you don't want to rely on stats that require a huge amount of data to give reliable signals. That's why I've never been a huge fan of using aggregate win/loss amounts on sites like TableRatings. Unless someone's a very consistent winner or loser over a large sample, then looking at their graph won't tell you anything useful. An account that has lost a significant sum in a small number of high stakes hands could well be a fish. It could also be a high-stakes regular from another site who has created a new account and hit an instant downswing in a high variance game.

Table level statistics such as the average number of players to the flop and the average pot size are much too broad and subject to random errors. Instead, you have to look at the qualities of the individual players. Now in my games, I'm always on the lookout to find at least one bad player. One bad player in a six-max game is both about the minimum required to make a game profitable after rake, and is also as good a game as you'll usually get -- as the rest of the seats generally get filled by regulars once a known fish sits.

Focusing in on that one bad player: what differentiates them from the regulars? They lose, of course, but remember the issues of sample size and false positives when looking at win/loss amounts. A statistic to efficiently differentiate losers from winners must therefore capture some aspect of playing style, and it must do so in small samples with few false positives. The sample size issue eliminates a majority of postflop stats, as these can take a long time to converge. A lopsided distribution of postflop aggression is an easy marker of fishy play; some fish tend to overload one street with aggression while playing the others much more passively. For the sake of balance regulars tend to distribute their aggression throughout the hand much more evenly. But these stats will take at least a few hundred hands to converge, which is far too slow to make this a primary fish marker.

Source