College Baseball Power Rankings Theory Overview, Part One

Major League Baseball has arguably the most advanced analytics of any sport league in the world. Every year, new analytical tools are released that provide data that more closely measures the on-field product and new statistics get published that help us to better interpret this data. From Bill James first posing the most basic questions of measuring success that helped spawn a new field of baseball analysis, all the way through present day technologies such as ball- and player-tracking systems, the collective understanding of how Major League Baseball works has grown exponentially over the last several decades.

But for a variety of reasons, including league size, data availability, general interest, and, more than anything else, money, the science of college baseball analytics has fallen behind. Through a series of publications that we will present over the coming months, we will attempt to catch up, as much as possible. We will release new statistics, new methods, and new ideas that will help pull college baseball statistics out of the dark ages of batting average and ERA.

One of the first, most basic questions we can ask is this: which team is the best? It seems a simple enough question, yet it can be surprisingly difficult to answer. The most basic response would probably be to look at their record. But that has several problems. It doesn’t look at the quality of the team’s opponents: if a team plays in the Patriot League and goes 28-25, that would not be nearly as impressive as a team that goes 28-25 in the ACC or the SEC, where their competition would be much better. It also doesn’t consider margin of victory: a 15-1 win is more impressive than a 6-5 win, and teams that have a larger margin of victory should be considered better.

There are publicly available statistics that solve some of these issues. Pythagorean win percentage is a formula originally proposed by Bill James that estimates a team’s win percentage based on their runs scored and runs allowed. This formula has been proven to correlate better with future success than straight win percentage does. The problem with using Pythagorean win percentage is that it only measures runs, it does not measure who the team scored those runs on or allowed those runs to. Scoring seven runs on Oregon State last year was much more impressive than scoring seven on Alabama A&M.

On the other hand, the ranking statistic that the NCAA uses is called RPI, Rating Percentage Index. It calculates a team’s overall quality by weighting the teams win percentage, their opponent’s win percentage, and their opponent’s opponent’s win percentage. This does a pretty solid job of putting a team’s success in context based on who they have played, and who their opponents have played, but it doesn’t consider that first factor: margin of victory.

So, what is the easiest way to fix these issues with one statistic? Combine them! We will use Pythagorean-like win percentage in an RPI-like structure to take the benefits of both statistics while removing both statistics’ weaknesses. Below I will describe, in detail, how we get to our final rankings, and then present those rankings for you to peruse.

You will notice that above I said “Pythagorean-like” win percentage. Pythagorean win percentage takes runs scored squared (RS2) and divides by the sum of runs scored squared (RS2) and runs allowed squared (RA2). This means that more runs scored increases the statistic and runs allowed decreases it. This formula is very simple and convenient, but it turns out that 2 is not the most statistically significant exponent. There have been many attempts to improve upon the basic Pythagorean win percentage. Baseball Reference has said that an exponent of 1.83 is the most statistically significant, and it works slightly better than 2. This would be fine, but we can do even better. Research done by Clay Davenport of Baseball Prospectus has shown that instead of using one exponent, the most accurate solution is to use a variable exponent; that is, instead of using one exponent for every team’s record, determine the optimized exponent based on the team itself. This method can be a little mathematically complex, so I will not go into anymore detail than that, but if you are curious, read up on the Pythagenport formula. This is the name of this spin on the Pythagorean win percentage, and we will refer to it as such from here, forward.

Using the Pythagenport formula, we calculate each team’s expected win percentage. This will replace the “Team Record” in the RPI formula. We then average the Pythagenport win percentage for each of that team’s opponents (weighted by number of games played against; if a team plays Kentucky 9 times and Arizona once, Kentucky’s Pythagenport win percentage would be represented 9 times more than Arizona’s). This will act as the team’s “Opponent’s Record” in the RPI formula, and this process is done once again for the opponent’s opponents, and that number will act as the “Opponent’s Opponent’s Record” in the RPI formula.

Now, the RPI formula weighs team record as 25% of the overall rating, opponent’s record as 50%, and opponent’s opponent’s record as 25%. These numbers are most likely not optimized, but we will not worry about that for now, and for simplicities sake use the same weights. We will likely do some research into optimizing these weights in the future. But for now, that is it! Our final statistic, which we will call Team Rating (creative, I know), is a 25%-50%-25% composite of a team’s Pythagenport record, a team’s opponent’s Pythagenport record, and a team’s opponent’s opponent’s Pythagenport record.

So how does our final rankings compare to the Official NCAA RPII rankings? You can see the Top 25 below:

RankTeamTeam RatingRPI RankDifference
1North Carolina0.625654
3Oregon St.0.62021-2
5Florida St.0.617183
8Texas Tech0.60637-1
16Wake Forest0.591510-6
17Texas A&M0.593316
18Southern Miss0.589911-7
19Sam Houston St.0.58893718
22North Carolina St.0.5833275
23South Carolina0.582318
25Missouri St.0.580423-2

You can see that the differences are there, but they are not massive. All of the Team Rating Top 25 are in the RPI Top 37, and all of the RPI Top 10 are in the Team Rating Top 16. Teams are consistently regarded as generally the same caliber by both metrics. In a future article we will analyze the predictive power of Team Rating and how it compares to RPI. I hope we have piqued your interest, and I will see you next time.

Leave a Reply

%d bloggers like this: