Power Rankings using only Win/Loss

[J.E.]

Not an entirely new idea, but I've developed a power ranking where only wins and losses count. I'm adjusting for strength of opponent, location and back-to-backs etc.
Instead of using point differential I'm simply scoring each game as 1 or 0 (win/loss)

I'm using linear regression, which is probably far from perfect for this type of problem, but should still lead to decent results

Here are the '14 regular season ratings

Code: Select all

╔══════════════════════════╦════════════╗
║           Team           ║ Rating*100 ║
╠══════════════════════════╬════════════╣
║ San Antonio Spurs        ║ 21         ║
║ Los Angeles Clippers     ║ 19         ║
║ Oklahoma City Thunder    ║ 17.4       ║
║ Indiana Pacers           ║ 16.2       ║
║ Portland Trail Blazers   ║ 15.4       ║
║ Houston Rockets          ║ 13.9       ║
║ Golden State Warriors    ║ 13         ║
║ Miami Heat               ║ 12         ║
║ Memphis Grizzlies        ║ 11.3       ║
║ Phoenix Suns             ║ 7.1        ║
║ Dallas Mavericks         ║ 6.9        ║
║ Chicago Bulls            ║ 5.4        ║
║ Toronto Raptors          ║ 5.2        ║
║ Washington Wizards       ║ 3          ║
║ Charlotte Bobcats        ║ 2.4        ║
║ Brooklyn Nets            ║ 0.5        ║
║ Minnesota Timberwolves   ║ -0.4       ║
║ Denver Nuggets           ║ -1.1       ║
║ Atlanta Hawks            ║ -2         ║
║ New Orleans Pelicans     ║ -4.5       ║
║ New York Knicks          ║ -5.9       ║
║ Cleveland Cavaliers      ║ -8.9       ║
║ Sacramento Kings         ║ -10.3      ║
║ Utah Jazz                ║ -12.5      ║
║ Los Angeles Lakers       ║ -13.4      ║
║ Detroit Pistons          ║ -15.3      ║
║ Boston Celtics           ║ -20.3      ║
║ Orlando Magic            ║ -20.8      ║
║ Milwaukee Bucks          ║ -27.1      ║
║ Philadelphia 76ers       ║ -27.2      ║
╚══════════════════════════╩════════════╝

This needs to be interpreted the following way: The Spurs (with a rating of 21) vs the Heat (12) would have been estimated to have a 50+21-12 = 59% chance of winning on neutral ground. Home teams win ~60% of their games. Thus, the Spurs would have been estimated to have a 60+21-12 = 69% chance of winning at home, and a 49% chance of winning in Miami.
The 76ers vs the Spurs would have a 60-27-21 = 12% chance of winning at home, and a 40-27-21 = -8% chance of winning at the Spurs... that's linear regression for ya

Here's team specific homecourt advantage using this method

Code: Select all

╔═════════════════════════════════════╦═══════════╗
║                Team                 ║ Extra HCA ║
╠═════════════════════════════════════╬═══════════╣
║ Denver Nuggets                      ║ 6.4       ║
║ Utah Jazz                           ║ 6.3       ║
║ Atlanta Hawks                       ║ 5.3       ║
║ Charlotte Bobcats                   ║ 4.7       ║
║ Indiana Pacers                      ║ 4.5       ║
║ Golden State Warriors               ║ 4.2       ║
║ New Orleans Pelicans                ║ 4         ║
║ Milwaukee Bucks                     ║ 3.9       ║
║ Washington Wizards                  ║ 3.8       ║
║ Los Angeles Clippers                ║ 3.7       ║
║ Sacramento Kings                    ║ 3.4       ║
║ Cleveland Cavaliers                 ║ 3         ║
║ New Orleans/Oklahoma City Hornets   ║ 2.6       ║
║ Portland Trail Blazers              ║ 2.2       ║
║ Memphis Grizzlies                   ║ 1.5       ║
║ New Jersey Nets                     ║ 1.3       ║
║ Chicago Bulls                       ║ 0.6       ║
║ New York Knicks                     ║ -0.2      ║
║ Orlando Magic                       ║ -0.3      ║
║ Toronto Raptors                     ║ -0.8      ║
║ Minnesota Timberwolves              ║ -1.1      ║
║ Los Angeles Lakers                  ║ -1.2      ║
║ Miami Heat                          ║ -1.4      ║
║ New Orleans Hornets                 ║ -2.1      ║
║ Detroit Pistons                     ║ -2.2      ║
║ Houston Rockets                     ║ -2.5      ║
║ Brooklyn Nets                       ║ -2.7      ║
║ Phoenix Suns                        ║ -3.1      ║
║ Seattle SuperSonics                 ║ -3.4      ║
║ Boston Celtics                      ║ -4.2      ║
║ Philadelphia 76ers                  ║ -4.5      ║
║ San Antonio Spurs                   ║ -5.1      ║
║ Dallas Mavericks                    ║ -5.2      ║
║ Oklahoma City Thunder               ║ -5.5      ║
╚═════════════════════════════════════╩═══════════╝

... effect of rest

Code: Select all

╔══════╦════╦═══════════════╦═══════════════╦═════════════════════╗
║ Rest ║ OT ║ last location ║ this location ║ effect for awayteam ║
╠══════╬════╬═══════════════╬═══════════════╬═════════════════════╣
║ 1d   ║    ║ away          ║ away          ║ 4.5                 ║
║ 1d   ║    ║ home          ║ away          ║ 3.8                 ║
║ b2b  ║    ║ away          ║ away          ║ -1.2                ║
║ b2b  ║ OT ║ home          ║ away          ║ -1.3                ║
║ b2b  ║    ║ home          ║ away          ║ -2.3                ║
║ b2b  ║ OT ║ away          ║ away          ║ -5.1                ║
╚══════╩════╩═══════════════╩═══════════════╩═════════════════════╝
╔══════╦════╦═══════════════╦═══════════════╦═════════════════════╗
║ Rest ║ OT ║ last location ║ this location ║ effect for hometeam ║
╠══════╬════╬═══════════════╬═══════════════╬═════════════════════╣
║ 1d   ║    ║ home          ║ home          ║ 5.1                 ║
║ 1d   ║    ║ away          ║ home          ║ 4.3                 ║
║ b2b  ║    ║ home          ║ home          ║ 1.6                 ║
║ b2b  ║    ║ away          ║ home          ║ -0.4                ║
║ b2b  ║ OT ║ home          ║ home          ║ -7*                 ║
║ b2b  ║ OT ║ away          ║ home          ║ -9.2                ║
╚══════╩════╩═══════════════╩═══════════════╩═════════════════════╝
╔══════╦════════╗
║ Rest ║ Effect ║
╠══════╬════════╣
║ 2d   ║ 3.9    ║
║ 3d   ║ 3      ║
╚══════╩════════╝
╔══════╦═══════════════════╗
║      ║ Additional Effect ║
╠══════╬═══════════════════╣
║ 4in5 ║ -2.4              ║
║ 3in4 ║ 1.1               ║
╚══════╩═══════════════════╝

(*) small sample size

and the 10 top/bottom teams since '02

Code: Select all

╔═════════════════════════╦══════╦════════╗
║          Team           ║ Year ║ Rating ║
╠═════════════════════════╬══════╬════════╣
║ Cleveland Cavaliers     ║ 2009 ║ 27.5   ║
║ Dallas Mavericks        ║ 2007 ║ 24.9   ║
║ Miami Heat              ║ 2013 ║ 24.1   ║
║ Los Angeles Lakers      ║ 2009 ║ 24     ║
║ Sacramento Kings        ║ 2002 ║ 23.1   ║
║ Boston Celtics          ║ 2008 ║ 22.9   ║
║ Detroit Pistons         ║ 2006 ║ 22     ║
║ Cleveland Cavaliers     ║ 2010 ║ 21.5   ║
║ Indiana Pacers          ║ 2004 ║ 21.3   ║
║ Sacramento Kings        ║ 2003 ║ 21.2   ║
║ …                       ║      ║        ║
║ New Orleans Hornets     ║ 2005 ║ -24.2  ║
║ Cleveland Cavaliers     ║ 2003 ║ -24.8  ║
║ Minnesota Timberwolves  ║ 2011 ║ -25    ║
║ Milwaukee Bucks         ║ 2014 ║ -27.1  ║
║ Philadelphia 76ers      ║ 2014 ║ -27.2  ║
║ Atlanta Hawks           ║ 2005 ║ -27.6  ║
║ Minnesota Timberwolves  ║ 2010 ║ -27.7  ║
║ Miami Heat              ║ 2008 ║ -28.5  ║
║ New Jersey Nets         ║ 2010 ║ -30.4  ║
║ Charlotte Bobcats       ║ 2012 ║ -33.1  ║
╚═════════════════════════╩══════╩════════╝

I'm strongly assuming that this type of power ranking is inferior to most power rankings that use point differential, but it would be interesting to see how a mix of this and a point differential based power ranking would perform

[mystic]

I'm strongly assuming that this type of power ranking is inferior to most power rankings that use point differential, but it would be interesting to see how a mix of this and a point differential based power ranking would perform

Funny thing: My power ranking (the result of a previous season can still be seen on the blog) was based on such a mix (just the average of a model based on the net value (ORtg-DRtg) and one based on wins) of those two, just that I did not have the sophisticated information you have about HCA and rest. I never tested that extensively, but in a 5yr retrodiction test that version had a lower RMSE than the separate models.