project to collect handhistories

[dchandler321]

A major project to collect hand histories is underway. Everyone who sends in over 10,000 of their own hands will receive a free analysis of their play. More information is available at Pokernomics.com.

[smartalecc5]

Sounds awful fishy. And what would you guys be doing with my database that I worked extensively to build?

[dchandler321]

I understand your concern. The project is being conducted by an academic economist, not a professional online poker player. If you go to our website Pokernomics.com , you will see that one of the places where you can submit your hand histories is to a research center at the University of Chicago that is run by Professor Steven Levitt (the person running the project). Here is a link to the center Initiative on Chicago Price Theory. Among the faculty and Board of Directors of the Center there are 4 nobel prize winners in economics as well as other extraordinarily accomplished people.

[Derek in NYC]

Im satisfied that this is a legitimate research project. What credentials does the academic team have to give poker play feedback? Its not going to be stuff I could learn myself by poking through Pokertracker, is it, like: "You only manage to make money with middle connectors like J9s or 9Ts from the button, so stop playing them elsewhere." That would be really disappointing if that were the feedback. On the other hand, if the feedback were something more along the lines of a comparative performance (measured statistically) between how you do and how the total dB does in various situations, that sort of feedback sounds quite unique.

[dchandler321]

You've got exactly the right idea. I am familiar with Poker Tracker's software and have seen the kind of advice that it can give. In fact, we are looking to mostly request hand histories from poker tracker, as these are the kind we best know how to extract data from.

While such advice is useful for some people, it is mostly just isolating along several variables where you make the most money relative to other places. The kind of analysis we will provide is going to have the benefit of comparing your playing style against many other peoples.

With a large database of people, it would not be difficult to identify which players are most similar based upon correlation between their play in many different situations. The beauty of statistics is that we can then classify player types in any which way. Based upon this classification we can then combine hand histories from similar players and extrapolate how any individual player would have fared in situations that player has never even played. The power of aggregated data is much greater than simplistic statistics based upon a sample size of one.

Another limitations of any one individual's database is that in most cases it will not be large enough to draw its own conclusions. For example, some situations even in the course of 25-30,000 hands might have only come up 10 times. You may have played them perfectly well but had bad beats and PT would tell you that you are bad at handling those situations and have a bad win rate there even if you had played your hand perfectly well. Thus, some of the standard statistics you might receive from pokertracker have shortcomings. We can account for this.

One final note. Though Steven Levitt and others working on the project have many ideas for analysis, our project is just beginning. Thus far only 3-4 people have submitted hand histories to us and have done this independent of the creation of the site pokernomics.com. These people were responding to a posting that Steven Levitt wrote in his blog: freakonomics blog posting. The pokernomics site has more information than the posting, but the blog that he keeps (that has the same title of a recent book he's written) might also be of interest.

[Derek in NYC]

I dont know how familiar you are with the way in which traditional Pokertracker data is used, but to give you a sense, most of us focus on a few key variables to profile our opponents. Those variables are VPIP (% the player voluntarily puts money into the pot), PFR (% the player raises preflop), AF (aggression factor, by street, defined as the ratio of %bet or raise divided by % call), etc.

These variables are fairly unidimensional, and basically if you define them into non-overlapping buckets, you can create crude profiles. For example, a high VPIP and PFR with high AF postflop, might be characterized as a "maniac". Then when playing this guy, one would tend to give less credence to his raises/aggression, etc.

But the method is so crude. It fails to answer for important questions such as: "How tricky is he?" "Does he raise his draws?" "How often does he fold to a river bet?" "How often does he bluff?"

I have always felt that if you were to look at a large data sample, it would be possible to conduct multivariate regressions that could yield much richer "profiles" about player types and styles. Much like a quantitative market research study defining market segments, such profiles would tend to be less like discrete "buckets" and more like semi-overlapping multifactor profiles whose overall contour was definable (and intuitively sensible), but whose exact boundary was uncertain.

Multivariate regression is a strong tool in market research, and if it were applied to online poker, you could learn a lot about what truly "makes" a winning player.

[stigmata]

why dont you just buy a commercial database, PM krishanleong.

[dchandler321]

I had heard of the poker database that he is selling. However, to the best of my knowledge his database is only observing hands and does not participate in them. While there are many things that can be learned from such observational data, especially regarding table conditions over extended periods of time, much is left out. To really understand what determines success in poker it is crucial to see hole cards and how different hole cards are played by various player types against other various player types.

This is exactly why we are asking many different individuals to submit their own hand histories. In return, we are willing to provide people who submit a sufficiently large hand history for analysis with detailed advice about how to maximize their profits in online poker.

[sfer]

1. I think it's very likely you underestimate the data necessary to do a project like this well.
2. If you are sincere, you are going to need hundreds of books to give out.
3. You should offer a Confidentiality Agreement to contributors.

[dchandler321]

You raise a good point about the amount of data needed. For some questions we do not believe we will need that much data. Others will as you say take a great deal of hand histories from many people. Steven Levitt is prepared to give away copies of his book to people who help him with the project. Moreover, despite permanent wrist injuries from his several years of data entry as a graduate student, he will also sign them.

We understand that players have good reason to be cautious of sharing their hand histories with anyone. We have written on our webpage in several places that no data will be shared with anyone not working on the project. But we should probably make this more explicit and prominently displayed on the webpage. People who are suspicious that we are even academically affiliated can mail their hand histories directly to the Initiative on Chicago Price Theory at the University of Chicago.