NBA Rankings

By: Daniel Tokarz

Updated April 13th

See the bottom of the page for a basic intro to the model as well as directions for its usage.

For a more in depth look at the math behind the model, click here

For a guide on the R script we use to make the model, click here

Rank Team YUSAG Coefficient
1 Houston Rockets 8.73
2 Toronto Raptors 7.03
3 Golden State Warriors 5.89
4 Utah Jazz 5.25
5 Philadelphia 76ers 4.49
6 San Antonio Spurs 4.41
7 Oklahoma City Thunder 4.10
8 Cleveland Cavaliers 3.17
9 Boston Celtics 2.63
10 Portland Trail Blazers 2.57
11 Minnesota Timberwolves 1.98
12 New Orleans Pelicans 1.69
13 Denver Nuggets 1.29
14 Indiana Pacers 0.84
15 Charlotte Hornets 0.05
16 Miami Heat -0.13
17 Washington Wizards -0.17
18 Los Angeles Clippers -0.30
19 Milwaukee Bucks -1.06
20 Detroit Pistons -1.15
21 Los Angeles Lakers -1.73
22 Dallas Mavericks -2.79
23 Brooklyn Nets -4.32
24 New York Knicks -4.89
25 Orlando Magic -6.05
26 Atlanta Hawks -6.18
27 Sacramento Kings -6.33
28 Memphis Grizzlies -7.10
29 Chicago Bulls -8.27
30 Phoenix Suns -9.84

Basic Introduction to the NBA YUSAG Coefficient

      Based on the work YUSAG done in creating our NCAA football and basketball rankings, I set out to create a model for ranking NBA teams using the same system.  I’ll upload a more thorough explanation of my mathematical model in the next few days, but basically, we create a linear model based upon three variables, the team playing, their opponent and the location.  It uses these values to predict the point differential of a game involving a given team and the average NBA team at a neutral location.  The game data was taken off of Basketball Reference. 


How to use this Model

      For a given game between two NBA teams, subtracting the YUSAG coefficient for the road team from the home team and giving the home team a slight boost (~2.1 points) gives the predicted point differential for a given game.

​      To find the probability that the team will win this game based upon this point differential, plug the differential “d” into this function W(d) which we extracted from our other linear model (more math on this section coming soon as well!)   

W(d) = 1/(1+e^(-.1472d))