web analytics

How Does the Committee Seed? Introducing ExpSd

March 14, 2011
By

The Bracket was revealed yesterday.  Quick thoughts and long ramblings below:

I have two rankings systems: my Bayesian predictive power ratings, which tell how good the teams are, and my DSMRPI ratings, which tell how much they have accomplished.  I would put teams in and seed them based on DSMRPI, which looks purely at win-loss data for the team, but for SoS looks at the Ken Pomeroy adjusted efficiency differentials to get a better idea of how good the opponents actually were.

Four teams were left out that I would have put in based on DSMRPI:

  1. Virginia Tech (as a 9 seed or so)
  2. Boston College (as an 11 seed or so)
  3. Minnesota (as a 12 seed or so)
  4. Cleveland St. (as a 12 seed or so)

The four teams to be displaced?

  1. VCU (I have them as 16th out!)
  2. USC (5th out)
  3. Georgia (4th out)
  4. Marquette (3rd out)

Of course, by my Bayesian ratings, Marquette is actually quite a good team and a worthy participant (they just perhaps didn’t *deserve* to get in based on record and SoS).  In fact, they’re 28th overall in the Bayesian ratings.  Perhaps the committee looks a *little bit* at how good the team actually is and mostly at their resume.

Why not run a regression and find out?  I used KenPom rank, Bayesian rank, and DSMRPI rank to predict the seed chosen by the committee.  KenPom rank was not significant and negative, indicating that recent-play-weighted efficiency is used more by the committee than a season-wide efficiency.  So after dropping KenPom out, here’s the result:

CoefficientsStandard Errort StatP-value
Intercept1.8638453320.527053.53640.000924
Bayesian0.0406520630.025651.58480.119712
DSMRPI0.1356830710.027404.95130.000010

So basically, the committee looks at 3.3 parts resume (win-loss + efficiency SoS) and 1 part adjusted efficiency differential, weighted toward more recent results.  The R^2 was over 70%, which was good given that I was regressing onto seed-line, which is in bunches of 4 (adding noise).

And I can’t complain with that method of choice.  Combine mostly resume with a little Bayesian ratings, and that’s a good way to do the seeding.  I’ll call this new rating ExpSd, since it’s the seed/rating we’d expect the committee to follow.

Of course, even by that evaluation method a few teams are a long way from where they perhaps should be.  All of the data is below.  Utah State is the big discrepancy in the seed line.  VT still looks like it should be in at the expense of VCU.

For the latest version of the Bayesian Rankings and this data, see Google Docs.

TeamConfBayesianRkDSMRPIRkConfAtLSeedExpSdExpSd
Ohio St.B1033.4125.8110111
KansasB1231.3324.2310121
DukeACC32.0223.6410131
PittsburghBE27.3420.81001193
San Diego St.MWC24.21125.6210241
North CarolinaACC22.71522.1501262
Notre DameBE25.41021.3801282
FloridaSEC21.31818.815012164
PurdueB1025.6821.6701352
Brigham YoungMWC23.51321.9601372
SyracuseBE25.5919.912013103
ConnecticutBE21.71720.711103144
WisconsinB1026.7719.813014113
TexasB1227.3519.714014123
KentuckySEC26.9618.817104154
LouisvilleBE24.21217.021014185
West VirginiaBE21.81618.618015175
ArizonaP1018.53118.816015195
Kansas St.B1219.32415.427015277
VanderbiltSEC18.43214.334015328
GeorgetownBE19.62220.89016134
St. John'sBE16.63917.419016226
CincinnatiBE18.53015.328016287
XavierA1018.82615.230016308
WashingtonP1022.91415.329107257
TempleA1017.33615.031017318
Texas A&MB1217.23714.933017339
UCLAP1014.75113.5390174111
Nevada Las VegasMWC19.52316.422018216
MichiganB1016.24215.032018349
George MasonCAA15.94413.2400184010
ButlerHorz16.34112.4431084311
IllinoisB1020.12015.725019236
VillanovaBE18.92515.526019267
Old DominionCAA15.34914.0351093710
TennesseeSEC15.44812.2460194612
Michigan St.B1017.03817.2200110246
Penn St.B1015.54616.1240110298
Florida St.ACC16.44013.6370110369
GeorgiaSEC12.96111.754011055
MissouriB1218.13412.84101113810
MarquetteBE18.72811.85101114411
GonzagaWCC18.72711.06010114913
Southern CaliforniaP1015.44711.755011151
Virginia CommonwealthCAA9.7759.469011171
Utah St.WAC20.51916.3231012205
RichmondA1015.25013.63810123910
ClemsonACC19.62112.04801124211
UABCUSA13.56012.24501124812
MemphisCUSA10.67210.86410126714
BelmontASun18.33310.86310135613
OaklandSum10.77011.85210135713
PrincetonIvy6.59711.75310136413
Morehead St.OVC5.71015.89910139814
BucknellPat4.51146.88810149114
Indiana St.MVC4.91116.79010149214
St. Peter'sMAAC2.91305.8100101410415
WoffordSC7.9894.1121101411215
Long IslandNEC1.41486.987101510215
Northern ColoradoBSky3.01294.2119101512315
UC Santa BarbaraBW2.31343.8125101512916
AkronMAC4.61133.1132101513116
NC AshevilleBSth-0.51650.1165101616216
Boston UniversityAE-0.1162-0.1166101616316
HamptonMEAC-5.8226-1.7190101619617
Arkansas Little RockSB-5.4221-2.3194101619717
Texas San AntonioSlnd-6.4231-2.0193101619817
Alabama St.SWAC-13.9298-10.9300101630317
Virginia TechACC18.13513.83600100359
MinnesotaB1013.85412.344001004512
Boston CollegeACC12.86212.642001004712
NorthwesternB1012.76311.9500010050100
Cleveland St.Horz8.78012.0470010052100
St. Mary'sWCC16.14311.1590010053100
Colorado St.MWC9.37712.0490010054100
CaliforniaP1013.65711.2570010058100
Washington St.P1014.05310.9620010059100
Oklahoma St.B1211.86711.2580010060100
New MexicoMWC15.74510.6650010061100
ColoradoB1213.55910.9610010062100
Miami FLACC14.25210.2660010063100
HarvardIvy7.69111.4560010065100
MarylandACC18.6298.3760010066100
AlabamaSEC13.8568.9720010068100
Southern MississippiCUSA9.9739.6670010069100
Missouri St.MVC9.7749.6680010070100
Texas El PasoCUSA11.9658.7740010072100
MarshallCUSA9.0799.0710010073100
Seton HallBE13.5587.9790010074100
Wichita St.MVC12.3647.5820010075100
DaytonA108.3837.9770010076100
MississippiSEC11.9667.2840010077100
ValparaisoHorz5.31048.7730010078100
College of CharlestonSC8.2867.6810010080100
DrexelCAA5.21078.4750010081100
NebraskaB1213.8556.5910010082100
RutgersBE8.3827.2850010083100
TulsaCUSA8.3847.0860010085100
Boise St.WAC8.5816.3920010086100
BaylorB1211.6685.9960010087100
DuquesneA1010.6715.51050010094100
IonaMAAC11.4695.51060010097100

Tags: , , , , ,

2 Responses to How Does the Committee Seed? Introducing ExpSd

  1. Nathan Walker on March 14, 2011 at 3:56 pm

    Interesting post. Pomeroy’s numbers are adjusted for recency, however.

    Since yours are adjusted more accurately, adding his numbers to the equation adds nothing of statistical significance.

    • DanielM on March 14, 2011 at 4:30 pm

      I didn’t think his were adjusted for recency?

Leave a Reply

Your email address will not be published. Required fields are marked *

Current day month ye@r *

DSMok1 on Twitter

To-Do List

  1. Salary and contract value discussions and charts
  2. Multi-year APM/RAPM with aging incorporated
  3. Revise ASPM based on multi-year RAPM with aging
  4. ASPM within-year stability/cross validation
  5. Historical ASPM Tableau visualizations
  6. Create Excel VBA recursive web scraping tutorial
  7. Comparison of residual exponents for rankings
  8. Comparison of various "value metrics" ability to "explain" wins
  9. Publication of spreadsheets used
  10. Work on using Bayesian priors in Adjusted +/-
  11. Work on K-Means clustering for player categorization
  12. Learn ridge regression
  13. Temporally locally-weighted rankings
  14. WOWY as validation of replacement level
  15. Revise ASPM with latest RAPM data
  16. Conversion of ASPM to" wins"
  17. Lineup Bayesian APM
  18. Lineup RAPM
  19. Learn SQL