The following is a letter that I wrote recently to a well-known person in the poker world … asking for help with regards to rankings on thepokerdb. I’m copying it below so other people can add their thoughts on this issue. I took out parts that aren’t relevant to the general public, but I marked them with a [snip].

[snip],

I have a problem that you might be able to help with … or at least point me in the right direction. I run the site thepokerdb.com and we’ve neglected the task of rating players (and the derivative rankings that would result) for quite awhile, mainly because it seems so difficult. I’m going to at least make an attempt in the next week or two to write a formula to rate players from 0 to 1 with 0 being the worst performers and 1 being the best. I’d present it on the site out to three decimal places, effectively giving everyone a rating from 0 to 1000.

In terms of the factors to be included, I’ve come up with at least a preliminary list of what I find important in judging performance strictly from a results perspective:

1. Played
2. FT Rate/Field Size
3. ITM Rate/Field Size
4. Top 3 Rate/Field Size
5. Average Finish
6. ROI
7. Average Buyins Won

It’s clear that they should not be equally weighted, but I’m not sure how to best weight them. I’m also fairly certain it would be incorrect to linearly apply the respective “credit” to the rating coming from each factor (ie, the first 1000 tournaments played are worth more than the last 1000 out of 20000 — not that we have that many results for anyone anyway). Basically I think there may be a marginal drop in relative utility for most of the stats as more samples are included, but the approximate rate of decrease has eluded me. Another issue is finding a way in the formula to “cap” the ranking at a certain level with 1000 being the best player(s) with regards to past performance. Lastly, it’s unclear how to apply these statistics to different time periods … such as the relative importance of number played over the last week v the last month v the last year. Obviously we’re not dealing with statistically significant sample sizes, but the fact remains that our users would like to see at least a reasonable attempt at rating people over non-statistically significant sample sizes.

[snip]. I’m sure you’re busy and I know there aren’t any “right” answers to this question (and it might not even matter if there were right answers due to sample size issues), but I’d appreciate any help or guidance with regards to where I can find help.

Thanks,

–
Nat Arem
www.thepokerdb.com

——–

UPDATE: Add an 8th factor: Average Buyin. Thanks for pointing that one out Ken.

Related posts:

1. the thepokerdb logo challenge
2. My jailbait rankings are out of control
3. return of the aim convo post
4. Emory Law & thepokerdb update
5. Start of thepokerdb