Boyd's World-> Breadcrumbs Back to Omaha-> There Are Other Sports?

About the author, Boyd Nation

There's probably a risk in my saying this directly below a publication date of April 1, but it turns out that the RPI's work very well.

This year, for the first time in almost a decade, I decided to enter a pool for the NCAA basketball tournament. Since I don't really follow the game on a national level, I thought I'd try to see if the ISR's would actually work on basketball. They work well for baseball, of course, and there's an amateur volleyball association using them who are pleased with the results as far as I know, but I had wondered about basketball and whether they would work there. After digging into the theoretical underpinnings, I couldn't find any reason why they wouldn't -- there's a theoretical minimum to the number of games you need for a good rating which varies from sport to sport based on how prevalent upsets are, but they're not all that common in basketball, so thirty games works out fine.

In using the ISR's to make out my bracket, I noticed that my picks were awfully boring; with a few exceptions, the seeds matched the ratings really well. Some of the exceptions were interesting, especially from a personal point of view -- I had Mississippi State at #26 and Butler at #33, so that game was disappointing but not surprising for me. The fact that the seeds matched the ratings was surprising, though; I'm not used to that. In watching past basketball selection shows, it had seemed to me that the basketball selection committee followed the RPI's a good bit more than the baseball committee does, so the next thought was that perhaps the RPI worked better for basketball than it does for baseball. It turns out that that is indeed the case, which may explain why the NCAA sticks with it; it works when it matters for the most money.

As an example, here are the top 10 from both systems for the 2002 baseball season (I'm using the core RPI formula with no bonuses here, since the bonuses differ slightly, but not really importantly, for each sport):

    RPI                         ISR

 1. Clemson                  1. Texas
 2. Florida State            2. Florida State
 3. Wake Forest              3. Stanford
 4. South Carolina           4. Clemson
 5. Georgia Tech             5. Rice
 6. Texas                    6. South Carolina
 7. Houston                  7. Houston
 8. Rice                     8. Wake Forest
 9. Stanford                 9. Georgia Tech
10. LSU                     10. Alabama

The membership's are similar, but there's not a lot of resemblance. By contrast, here's the men's basketball lists:

    RPI                         ISR

 1. Kansas                    1. Kansas
 2. Maryland                  2. Maryland
 3. Oklahoma                  3. Duke
 4. Duke                      4. Oklahoma
 5. Cincinnati                5. Cincinnati
 6. Connecticut               6. Arizona
 7. Arizona                   7. Connecticut
 8. Alabama                   8. Illinois
 9. Indiana                   9. Alabama
10. Kentucky                 10. Kentucky

Obviously, that's much more similar.

Just to finalize the proof for myself, I ran some correlations. Taking the full set of RPI and ISR rankings for six seasons worth of scores from three different sports, I get the following correlations between the ISR's and the RPI's:

       Men's Basketball          Women's Basketball           Baseball

2001         .99                        .98                     .94
2002         .99                        .98                     .93

That's striking enough, but here's an even more interesting result. I took the teams that we actually care about for the postseason and compared the results there. This correlation value eliminates all the teams who aren't in the top 50 in either list:

       Men's Basketball          Women's Basketball           Baseball

2001         .92                        .83                     .44
2002         .91                        .83                     .58

Where it matters most is where the RPI's fail worst, in the top 50. On the other hand, they work very well for men's basketball, and work fairly well for women's basketball. That makes sense given what causes most of the problems in baseball -- fewer West Coast teams and less inter-regional play. Both of those conditions exist to a smaller extent in women's basketball than in baseball, but more so than in men's basketball.

Next week, I want to take a look at the sport of college baseball itself compared to other sports in terms of predictability and parity. I have the data for the amateur and pro versions of football, men's and women's basketball, baseball, and hockey, but if you know of a source of data for other sports, let me know; I'd love to throw them in.

Rather than keep returning to the subject of pitch counts and pitcher usage in general too often for my main theme, I'm just going to run a standard feature down here where I point out potential problems; feel free to stop reading above this if the subject doesn't interest you. This will just be a quick listing of questionable starts that have caught my eye -- the general threshold for listing is 120 actual pitches or 130 estimated, although short rest will also get a pitcher listed if I catch it. Don't blame me; I'm just the messenger.

(*) Pitch count is estimated.

If you're interested in reprinting this or any other Boyd's World material for your publication or Web site, please read the reprint policy and contact me

Boyd's World-> Breadcrumbs Back to Omaha-> There Are Other Sports?

About the author, Boyd Nation