RPI formula problems

[jay santos]

RPI is a really odd formula.

RPI is generally SOS heavy, meaning teams with strong SOS are overrated and teams with weak SOS are underreated.

teams with the highest SOS in the top 20:

team, RPI rank, Sagarin rank

UCLA, 1, 4
Arizona, 6, 10
Kentucky, 7, 12
Southern Illinois, 9, 20

teams with lowest SOS in the top 20:

team, RPI rank, Sagarin rank

Florida, 8, 2
T A&M, 10, 6
Memphis, 11, 9
Butler, 31, 11

So RPI is clearly SOS heavy. However, the way RPI calculates SOS errors in an SOS light way.

i.e.

though RPI ranks Arizona higher than Butler, one gets more lift in their RPI by beating #31 Butler than by beating #6 Arizona because in the RPI SOS calculation, those teams flip flop.

Very lame of college basketball to put so much stock in this formula. For as much we hate the BCS, the BCS formula is much more progressive, accurate, and fair.

[pelagius]

Originally Posted by jay santos

RPI is a really odd formula.

RPI is generally SOS heavy, meaning teams with strong SOS are overrated and teams with weak SOS are underreated.

teams with the highest SOS in the top 20:

team, RPI rank, Sagarin rank

UCLA, 1, 4
Arizona, 6, 10
Kentucky, 7, 12
Southern Illinois, 9, 20

teams with lowest SOS in the top 20:

team, RPI rank, Sagarin rank

Florida, 8, 2
T A&M, 10, 6
Memphis, 11, 9
Butler, 31, 11

So RPI is clearly SOS heavy. However, the way RPI calculates SOS errors in an SOS light way.

i.e.

though RPI ranks Arizona higher than Butler, one gets more lift in their RPI by beating #31 Butler than by beating #6 Arizona because in the RPI SOS calculation, those teams flip flop.

Very lame of college basketball to put so much stock in this formula. For as much we hate the BCS, the BCS formula is much more progressive, accurate, and fair.

Agreed, the RPI is lame; Do we know exactly how the seeding committee uses RPI? Do they use it to seed or just to determine whether a team is in or out? Are they vague about exactly how it is used?

[Indy Coug]

Have you seen my correlation analysis of the RPI with NCAA seeding? The correlation is around 0.94

[livecoug]

with outliers being MWC schools.. it's insane

why does the MVC get so much more respect at seeding time? I'll never understand

[pelagius]

Originally Posted by Indy Coug

Have you seen my correlation analysis of the RPI with NCAA seeding? The correlation is around 0.94

Nope, you got a link for me?

[Indy Coug]

The basis for this analysis is this:

1. Take the 64 or 65 teams that make the NCAA tournament and sort them by their RPI

2. The first four teams with the highest RPI get a 1 seed, the next four teams a 2 seed and so forth. This is called the "expected RPI".

3. Compare this against their actual RPI.

Here are the results from 1999 to 2005.

Diff	Overall		2005	2004	2003	2002	2001	2000	1999
-7	0.2%		0.0%	0.0%	1.6%	0.0%	0.0%	0.0%	0.0%
-5	0.7%		0.0%	0.0%	1.6%	0.0%	0.0%	1.6%	1.6%
-4	2.0%		1.5%	3.1%	0.0%	1.6%	4.7%	1.6%	1.6%
-3	3.3%		0.0%	1.5%	1.6%	7.8%	4.7%	3.1%	4.7%
-2	7.3%		12.3%	7.7%	7.8%	7.8%	4.7%	4.7%	6.3%
-1	17.8%		16.9%	18.5%	15.6%	15.6%	17.2%	21.9%	18.8%
0	38.2%		35.4%	36.9%	42.2%	37.5%	37.5%	39.1%	39.1%
1	16.2%		24.6%	23.1%	14.1%	14.1%	15.6%	9.4%	12.5%
2	6.9%		4.6%	3.1%	9.4%	9.4%	6.3%	10.9%	4.7%
3	5.3%		4.6%	4.6%	3.1%	3.1%	6.3%	7.8%	7.8%
4	1.1%		0.0%	0.0%	3.1%	0.0%	3.1%	0.0%	1.6%
5	0.7%		0.0%	1.5%	0.0%	1.6%	0.0%	0.0%	1.6%
7	0.2%		0.0%	0.0%	0.0%	1.6%	0.0%	0.0%	0.0%

Here is the correlation of expected to actual seeds by year:

Year	Correlation
1999	0.926
2000	0.940
2001	0.929
2002	0.921
2003	0.928
2004	0.948
2005	0.960

http://www.cougarboard.com/noframes/...html?id=660139

[Indy Coug]

The RPI is a perfect example of SOS hijacking the bus in the absence of MOV providing any real counterweight. This is why the BCS computer formulas are a complete farce.

[jay santos]

Originally Posted by Indy Coug

The RPI is a perfect example of SOS hijacking the bus in the absence of MOV providing any real counterweight. This is why the BCS computer formulas are a complete farce.

Computer formulas with MOV > BCS computer formulas > RPI

However the widest gap is in the jump to RPI. MOV helps, but more important is having the right balance between SOS and winning. RPI doesn't have that but at least BCS computer formulas do.

[pelagius]

Originally Posted by Indy Coug

Year	Correlation
1999	0.926
2000	0.940
2001	0.929
2002	0.921
2003	0.928
2004	0.948
2005	0.960

Thanks, Indy. It is hard to imagine that another formula or computer ranking would give a higher correlation than 0.9-0.96. An, interesting question is when the committee does deviate greatly from the RPI prediciton is there a systematic cause or does it tend to be idiosyncratic? For example, does the committee tend to deviate from the RPI prediction when there is a large difference between the RPI and say the Sagarin Computer Ranking? I am not suggesting that the committee uses the Sagarin Ranking, I just wonder if we can ex ante empirically identify deviations from the RPI prediction that way.

[jay santos]

Originally Posted by pelagius

Thanks, Indy. It is hard to imagine that another formula or computer ranking would give a higher correlation than 0.9-0.96. An, interesting question is when the committee does deviate greatly from the RPI prediciton is there a systematic cause or does it tend to be idiosyncratic? For example, does the committee tend to deviate from the RPI prediction when there is a large difference between the RPI and say the Sagarin Computer Ranking? I am not suggesting that the committee uses the Sagarin Ranking, I just wonder if we can ex ante empirically identify deviations from the RPI prediction that way.

I believe the committee is not allowed to review any other rankings besides RPI. I've heard the following as rankings criteria:

1. RPI as a primary guideline
2. Straying from RPI due to other factors
--past NCAA success such as a Gonzaga
--conference standings, i.e. you finished in your conference behind someone with a lower RPI
--last 10 games
--quality wins or record vs top 50 and top 100

The conference standings criteria is especially what did BYU in in the past, IMHO, which excites me for this year, because it should go in our favor.