Pomeroy question

[SlimSlowSlider]

How do you use his statistics to predict the margin of victory? Can this be done from this Pythag winning %, or is there a more complicated way of doing this?

I find his statistics very interesting, but not as user-friendly as Sagarin.

[appzter]

I'm pretty sure that since he's switched to Pythagorean Win%, you can't predict score margins.

Since you're interested, I have my own computer rating system (I use the data Pomeroy provides on his site) that can be used for that purpose, with varying degrees of success. Here are the latest rankings:

http://www.duke.edu/~mhj32/rankings.txt

Since NCAAT games are on "neutral" courts, just subtract one team's ranking from another's to find the predicted score margin.

Let me know what you think of these -- I'm looking for input.

- matt

[mgtr]

Those numbers from your system seem to reflect reality. They also say that the top teams are all pretty close, when anything can happen. Also, they suggest that teams better worry about Davidson!

[appzter]

Here are the games it says will be close (margin under 10 points) or upsets:

4 Pittsburgh, 13 Oral Roberts (6.38)
4 Vanderbilt, 13 Siena (4.74)
5 Clemson, 12 Villanova (6.49)
5 Drake, 12 Western Kentucky (4.25)
5 Michigan St., 12 Temple (7.7)
5 Notre Dame, 12 George Mason (7.52)
6 Marquette, 11 Kentucky (9.21)
6 Oklahoma, 11 Saint Joseph's (.64)
6 Purdue, 11 Baylor (.34)
6 Southern California, 11 Kansas St. (-1.37)
7 Butler, 10 South Alabama (3.43)
7 Gonzaga, 10 Davidson (.95)
7 Miami FL, 10 St. Mary's (-1.76)
7 West Virginia, 10 Arizona (3.15)
8 Brigham Young, 9 Texas A&M (-1.49)
8 Indiana, 9 Arkansas (3.58)
8 Mississippi St., 9 Oregon (.16)
8 Nevada Las Vegas, 9 Kent St. (1.19)

[MChambers]

During the regular season, Pomeroy does give a predicted score. I think you could use his system to do the same in the NCAAs, but it wouldn't be completely trivial. Read more about it here:

http://kenpom.com/blog/index.php/weblog/ratings_explanation/

[SlimSlowSlider]

Thanks Appzter, for the response and the link.

When you say "varying degrees of success," have you tracked your system. I know no system is perfect, but have you done an analysis of your system against the point spread?

[mattk11]

Hey SlimSlowSlider -

Saw you were posting about Pomeroy and Margin of Victory. I actually run a site that calculates MOV and more for the tournament. Pomeroy's data is actually an input to some of the stuff we do. You can check it out here:

http://www.bracketbrains.com

We also have detailed Power Ratings that can be found here:

http://teamrankings.com/ncb/

Certainly would be interested to hear what you think!

[gofurman]

Here are the games it says will be close (margin under 10 points) or upsets:

4 Pittsburgh, 13 Oral Roberts (6.38)
4 Vanderbilt, 13 Siena (4.74)
5 Clemson, 12 Villanova (6.49)
5 Drake, 12 Western Kentucky (4.25)
5 Michigan St., 12 Temple (7.7)
5 Notre Dame, 12 George Mason (7.52)
6 Marquette, 11 Kentucky (9.21)
6 Oklahoma, 11 Saint Joseph's (.64)
6 Purdue, 11 Baylor (.34)
6 Southern California, 11 Kansas St. (-1.37)
7 Butler, 10 South Alabama (3.43)
7 Gonzaga, 10 Davidson (.95)
7 Miami FL, 10 St. Mary's (-1.76)
7 West Virginia, 10 Arizona (3.15)
8 Brigham Young, 9 Texas A&M (-1.49)
8 Indiana, 9 Arkansas (3.58)
8 Mississippi St., 9 Oregon (.16)
8 Nevada Las Vegas, 9 Kent St. (1.19)


Interesting that all four 5s are on here and two of the 4s !

[Troublemaker]

For Pomeroy's game predictions, just click on a team's schedule. They are up now for the NCAA tourney.

My unsolicited advice is not to try to beat Vegas spreads by using computer predictions / power rankings. The spreads almost always align closely with the computers anyway.

[appzter]

Interesting that all four 5s are on here and two of the 4s !

More interesting, I think, is that it isn't predicting any 12-over-5 upsets.


Thanks Appzter, for the response and the link.

When you say "varying degrees of success," have you tracked your system. I know no system is perfect, but have you done an analysis of your system against the point spread?

This is something I've been meaning to do, but haven't gotten around to.

[JStuart]

Since NCAAT games are on "neutral" courts, just subtract one team's ranking from another's to find the predicted score margin.

Let me know what you think of these -- I'm looking for input.

- matt
And of course, UNC is playing on 'neutral ' courts throughout...

[ArtVandelay]

Hey SlimSlowSlider -

Saw you were posting about Pomeroy and Margin of Victory. I actually run a site that calculates MOV and more for the tournament. Pomeroy's data is actually an input to some of the stuff we do. You can check it out here:

http://www.bracketbrains.com

We also have detailed Power Ratings that can be found here:

http://teamrankings.com/ncb/

Certainly would be interested to hear what you think!

With respect to your "similar game" projection method, I wonder whether the sample size is large enough to draw any definitive conclusions. My stat-based knowledge is mostly focused on baseball, but it seems like in college basketball there's a lot of statistical noise that you need to control for in any sort of modeling/projection system.

Interesting that your statistical odds calculator give UCLA the lowest probability of making the FF of any of the 1 seeds.

[calltheobvious]

More interesting, I think, is that it isn't predicting any 12-over-5 upsets.

How many such upsets have been predicted by computer models in previous tournaments? My guess is you can count such instances on one hand. Those would reflect really, really bad seedings by the committee.