Even More RPI and Seeding Talk - cougarguard.com — unofficial BYU Cougars / LDS sports, football, basketball forum and message board

pelagius · 03-13-2007, 06:47 PM

I took Indy's RPI results and augmented them by doing the same thing with the Massey rating (see his post for methodology: http://www.cougarboard.com/noframes/...ml?id=2545050:)

Code:

                             rpi   NCAA   rpi mass  mass  mass 
team                   rpi  seed   seed  diff rank  seed  diff
Kentucky                12     3     8    -5    22     6    -2
Villanova               15     4     9    -5    28     7    -2
Creighton               20     5    10    -5    47    10     0
UNLV                    10     3     7    -4    16     4    -3
Arizona                 14     4     8    -4    12     3    -5
Illinois                30     8    12    -4    36     9    -3
Brigham Young           19     5     8    -3    39     9     1
Michigan St             24     6     9    -3    27     7    -2
Arkansas                35     9    12    -3    43    10    -2
Old Dominion            37     9    12    -3    71    12     0
S Illinois               7     2     4    -2    14     4     0
Duke                    16     4     6    -2    17     5    -1
Marquette               22     6     8    -2    25     6    -2
Davidson                49    11    13    -2    52    11    -2
UCLA                     2     1     2    -1     3     1    -1
Wisconsin                4     1     2    -1     2     1    -1
Pittsburgh               5     2     3    -1     9     3     0
Tennessee               13     4     5    -1    18     5     0
Nevada                  23     6     7    -1    20     5    -2
Xavier                  32     8     9    -1    61    12     3
VA Commonwealth         42    10    11    -1    62    12     1
Holy Cross              61    12    13    -1   102    14     1
New Mexico St           69    12    13    -1    69    12    -1
Wright St               73    13    14    -1    97    13    -1
TAM C. Christi          82    14    15    -1    93    13    -2
E Kentucky             128    15    16    -1   181    16     0
Ohio St                  1     1     1     0     1     1     0
North Carolina           3     1     1     0     6     2     1
Memphis                  8     2     2     0     8     2     0
Indiana                 28     7     7     0    31     8     1
Penn                    81    14    14     0   109    14     0
Oral Roberts            89    14    14     0   106    14     0
Belmont                112    15    15     0   155    15     0
North Texas            126    15    15     0   156    15     0
Central Conn           150    16    16     0   184    16     0
Jackson St             166    16    16     0   228    16     0
Florida A&M            164    16    16     0   223    16     0
Niagara                134    16    16     0   161    16     0
Florida                  6     2     1     1     4     1     0
Georgetown               9     3     2     1     7     2     0
Maryland                17     5     4     1    15     4     0
Boston College          33     8     7     1    37     9     2
Purdue                  44    10     9     1    38     9     0
Georgia Tech            52    11    10     1    30     8    -2
Texas Tech              53    11    10     1    53    11     1
Stanford                67    12    11     1    33     8    -3
Long Beach St           80    13    12     1    89    13     1
Albany NY               83    14    13     1   132    15     2
Miami OH                90    15    14     1   114    14     0
Weber St               143    16    15     1   158    15     0
Kansas                  11     3     1     2     5     2     1
Texas A&M               18     5     3     2    13     4     1
Butler                  27     7     5     2    49    11     6
Notre Dame              31     8     6     2    24     6     0
Gonzaga                 60    12    10     2    58    11     1
Winthrop                70    13    11     2    29     7    -4
G Washington            72    13    11     2    72    13     2
Oregon                  21     6     3     3    10     3     0
Texas                   25     7     4     3    21     5     1
Louisville              38     9     6     3    26     7     1
Washington St           26     7     3     4    11     3     0
Virginia Tech           34     9     5     4    40    10     5
Vanderbilt              47    10     6     4    41    10     4
USC                     40    10     5     5    23     6     1
Virginia                55    11     4     7    35     8     4

I also computed the mean absolute deviation of each ranking system:

Code:

Mean absolute deviation for Massey = 1.262
Mean absolute deviation for RPI = 1.815

At least this year on average we observe smaller deviation on average for the Massey ranking system than RPI. Furthermore, for the extreme negative rpi differences (-3 or worse) massey has a smaller difference (in magnitude) than the RPI for ever case except Arizona. Does this mean that the committee used Massey or some other computer ranking? I doubt it. I think it is more likely that there is a variable that they gave significant weight to (maybe top 50 or top 100 record) that is correlated with Massey.

MikeWaters · 03-13-2007, 06:54 PM

I want to see a comparison of the regression between RPI and seeding for mid-major vs power-conference, with a statistical test at alpha = .05.

pelagius · 03-13-2007, 07:05 PM

Quote:

Originally Posted by MikeWaters

I want to see a comparison of the regression between RPI and seeding for mid-major vs power-conference, with a statistical test at alpha = .05.

I really don't think you would get anything if you used the full field because 13-16 gets seeded very close to rpi and most of your non-power conference observations come from those seeds.

You might get something if you limited to seed 12 or better.

MikeWaters · 03-13-2007, 07:07 PM

Quote:

Originally Posted by pelagius

I really don't think you would get anything if you used the full field because 13-16 gets seeded very close to rpi and most of your non-power conference observations come from those seeds.

You might get something if you limited to seed 12 or better.

just do it for all at-larges.

SeattleUte · 03-13-2007, 07:11 PM

Quote:

Originally Posted by MikeWaters

I want to see a comparison of the regression between RPI and seeding for mid-major vs power-conference, with a statistical test at alpha = .05.

How do you square this obsession with your philosphy that those who are obsessed with ref calls are losers in the big game of life?

MikeWaters · 03-13-2007, 07:13 PM

I square this with my obsession of fighting the wealthy corrupt.

pelagius · 03-13-2007, 07:17 PM

Quote:

Originally Posted by MikeWaters

just do it for all at-larges.

For seed <= 12 (it works for the whole sample as well):

Code:

. reg seed rpiseed power;

      Source |       SS       df       MS              Number of obs =      48
-------------+------------------------------           F(  2,    45) =   26.26
       Model |  308.031875     2  154.015938           Prob > F      =  0.0000
    Residual |  263.968125    45  5.86595833           R-squared     =  0.5385
-------------+------------------------------           Adj R-squared =  0.5180
       Total |         572    47  12.1702128           Root MSE      =   2.422

------------------------------------------------------------------------------
        seed |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     rpiseed |   .6217125     .09951     6.25   0.000      .421289     .822136
       power |  -1.830879   .7853461    -2.33   0.024    -3.412648   -.2491109
       _cons |   3.703932   1.004132     3.69   0.001     1.681506    5.726359
------------------------------------------------------------------------------

So being in a power conference definitely helped. A power conference team with the same rpi is on average seeded almost two seeds lower.

Note: my power variable might have some errors in it. I added it by hand very quickly.

MikeWaters · 03-13-2007, 07:18 PM

exactly as I thought. Of course, you will never get the bigwigs to talk about this.

You should forward this to one of the espn writers like Andy Katz.

pelagius · 03-13-2007, 07:24 PM

Quote:

Originally Posted by MikeWaters

exactly as I thought. Of course, you will never get the bigwigs to talk about this.

You should forward this to one of the espn writers like Andy Katz.

Actually, it is possible that the RPI is biased toward non-power conferences and the seeding committee straightened things out (For example I think the RPI overcompensates for road wins and if nonpower play more road games ...):

Look at what happens if we do the same thing with the massey rankings:

Code:

. reg seed mseed power;

      Source |       SS       df       MS              Number of obs =      48
-------------+------------------------------           F(  2,    45) =   43.96
       Model |  378.358437     2  189.179219           Prob > F      =  0.0000
    Residual |  193.641563    45  4.30314583           R-squared     =  0.6615
-------------+------------------------------           Adj R-squared =  0.6464
       Total |         572    47  12.1702128           Root MSE      =  2.0744

------------------------------------------------------------------------------
        seed |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       mseed |    .777461    .093222     8.34   0.000     .5897022    .9652198
       power |  -.2265484   .7286263    -0.31   0.757    -1.694077     1.24098
       _cons |   1.558384   1.000078     1.56   0.126    -.4558772    3.572645
------------------------------------------------------------------------------

MikeWaters · 03-13-2007, 07:30 PM

however your chances of winning a road game are much less. the power conferences don't play any road games against mid-majors.

03-13-2007, 06:54 PM	#2
MikeWaters Demiurge Join Date: Aug 2005 Posts: 36,365	I want to see a comparison of the regression between RPI and seeding for mid-major vs power-conference, with a statistical test at alpha = .05.

03-13-2007, 07:13 PM	#6
MikeWaters Demiurge Join Date: Aug 2005 Posts: 36,365	I square this with my obsession of fighting the wealthy corrupt.

03-13-2007, 07:18 PM	#8
MikeWaters Demiurge Join Date: Aug 2005 Posts: 36,365	exactly as I thought. Of course, you will never get the bigwigs to talk about this. You should forward this to one of the espn writers like Andy Katz.

03-13-2007, 07:30 PM	#10
MikeWaters Demiurge Join Date: Aug 2005 Posts: 36,365	however your chances of winning a road game are much less. the power conferences don't play any road games against mid-majors.