Hack 59. Rank with the Best of Them


There are many ways to use data to make judgments about who is best in any sport. All the intuitive ways to compare performance in individual sports have validity concerns, however.

My friends and I are a competitive lot. Our arena of combat, most recently, has been poker. On a regular basis, my friends and I gather at my home and take part in a Texas Hold 'Em poker tournament. It's an informal affair, but we all take it very seriously. The way our poker tournaments work, everyone starts with the same amount of chips, and when they are gone they are gone. There is a first one out, a last one out, and everything in between. So, for example, if seven people play, someone comes in first, second, third, fourth, fifth, sixth, and seventh.

We all think of ourselves as pretty good and, being competitive, we have longed for an objective method of comparing performance across tournaments. As one of the statisticians in the group, I took it upon myself to devise various ways of producing some sort of objective index that would allow all participants to compare their performance with each other to decide once and for all who is the best player and who is only lucky now and again. This is the story of my quest and the statistical solutions I chose. Not to give the ending away, but I learned that there is no single best solution.

How to Rank Fairly

This business of how to identify the best is a common problem for competitive organizations such as sports leagues and associations. The problem is how to summarize performance across a variety of categories, venues, and occasions.

There are three methods commonly used in the world of sports to make determinations about who is the "best." All of the approaches make some intuitive sense, though each method has its own specific advantages and disadvantages.

First, let's take a look at the nature of the data I had to analyze. Your data will likely be similar, whether you run your weekly home Monopoly game or you run the Professional Golf Association. Though poker is not a sport, any organized competitive endeavor provides data for rankings. Table 5-16 shows the results from eight tournaments in my own summer poker league.

Table Summer poker league data
  Paul Lisa Billy BJ Mark Bruce Cathy Tim David
5/14 6 5 4 3 2 1   
5/21 3 6 4 5 7 2 1  
5/28   5 4 1 3 2  
6/4   4 6 3 7 2 5 1
6/11   4 5 6 1 2 3 
6/18   5 4 2 3 1  
6/25   1 4 3 5 2  
7/2   1 5 4 3 2  


You can see that nine players took part in at least one tournament, but no event had participation from all players. If a person received no points on a given night, it was because she didn't play. This is commonly the case in sports such as golf and tennis as well.

On two occasions, seven people played, but on other occasions, as few as five sat down together. Four people have played in all eight tournaments. (These are the hard-core players who have to admit that they have a bit of a problem recognizing what is important in life.) One player, David, played in only one tournament.

The points under each player's name indicate the order in which they went out. If there are six players and you go out first, you get one point for taking last place. If you are the winner among six players, you get six points for taking first.

Notice a couple of things about this point system. First, you get at least a point just for showing up. Second, you get more points for winning a tournament with more players.


How, then, to rank players in the poker league? Here are three common solutions, all of which work to some extent.

Total points

The first thought that came to mind in my situation was to simply add up the points across tournaments and rank players based on their total points. This is the approach taken when celebrities are ranked by income or bank robbers are ranked by their number of crimes. Just participating a lot moves you up in these rankings. To be golfer of the year, you have to have played in many events, in addition to performing OK in them.

Mean performance

A second method is to average the points by dividing the total points by the number of tournaments in which a player participated. The beauty of producing an average is that you get a number that represents a typical level of performance. This is ideal for measuring something elusive, such as talent. Your average performance at poker (or anything else) should be the best single indicator of ability.

Total wins

A third method, the simplest and most commonly used in team sports, is to count victories. The player who wins most often is the best player. This method works well for tournament-style poker (the kind we play) and any events in which there is one competitor who is the clear winner.

Comparing the Three Methods

Though each ranking approach has some clear advantages and does the job adequately, Table 5-17 shows the values for each player under all three ranking systems.

Table Summarizing poker performance
  Paul Lisa Billy BJ Mark Bruce Cathy Tim David
Points911283628251281
Mean4.55.53.54.53.53.131.714.01.0
Wins112122000


All three scoring systems make sense. But the question about who is the best has a different answer under each of the three systems! This is certainly a frustrating finding for a poker scientist like me. Because one could defend any of the three methods as the "best" way to rank, it is a bit of a paradox that each method produces a different "best" poker player. Table 5-18 shows how the rankings differ under each scoring method.

Table Poker rankings
  Paul Lisa Billy BJ Mark Bruce Cathy Tim David
Points762.512.54589
Mean2.515.52.55.57849
Wins442422666


Notice how the "best player" is different under each system. BJ is the best under the Points system. Lisa is the best under the Mean system. Three people tie for first under the Wins system, but BJ and Lisa are not among them. The only real agreement across the three methods is that David is ranked as the worst player. (Sorry, David, but numbers don't lie. And sorry about the public ridicule. Maybe I can make it up to you with a free copy of this book?)

I broke ties when assigning rankings by averaging the ranking among those who were tied. In other words, Billy, Mark, and myself were all tied for the number one ranking under the Wins system, so the ranks of 1, 2, and 3 average to 2, and that was our ranking.


If three different scoring systems result in three different rankings, it is clear they cannot all be equally valid. They cannot all produce scores that truly reflect the variable of interest, which is poker-playing ability defined in the same way. The solution does not involve picking the single best approach. It was not my goal to identify the best system and go with it; my goal was to provide valid information and let others interpret the data how they want.

My solution was to provide all three rankings based on the three scoring methods. That way, players could choose to focus on the ranking results from the method that makes the most sense to them.

The End of the Story

The system that made the most sense to the players in my poker league turned out to be the one that ranked them the highest. Imagine that.

I sleep at night secure in the knowledge that any of the methods is probably acceptable and "accurate." After all, none of the three methods makes the mistake of identifying me as the one best player. That's got to be some sort of validity evidence in and of itself!

Real-life professional sports organizations have dealt with the advantages and disadvantages of each system by creating composite point systems. Some of the tinkering to improve ranking systems in tennis and golf (and tournament poker, too) includes:

  • Combining performance data over a long period of time

  • Awarding more points for winning more difficult tournaments

  • Using both the mean performance and total points together to reward excellence and frequent participation

It is a bit ironic that these systems that are likely fairer and more accurate are often perceived by the press and fans as overly complex and crazy. Attempts to make the ranking systems more valid have resulted, often, in a rejection of the systems by the public as invalid.




Statistics Hacks
Statistics Hacks: Tips & Tools for Measuring the World and Beating the Odds
ISBN: 0596101643
EAN: 2147483647
Year: 2004
Pages: 114
Authors: Bruce Frey

Similar book on Amazon

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net