I have a program that can do this pretty automatically, but you can do it on your own!
There are five steps to finding the final score of any game:
1. Find each team's (adjusted) Offensive and Defensive Efficiency (their Points Per 100 Possessions)  from kenpom.com
For Example:
Duke's Offense  115.4
Duke's Defense  80.7
North Carolina's Offense  119.8
North Carolina's Defense  87.6
2. Find out where the game is being played. If it's a neutral game (not at either team's court), you don't have to adjust the data  you can find this at any sports website, but at kenpom.com, if you click on a team, then click schedule, you can tell if it's (H)ome, (A)way, or (N)eutral
For the home team, add 1.4% to their offense, and subtract 1.4% from their defense.
For the away team, subtract 1.4% from their offense, and add 1.4% to their defense.
For Example:
Duke's Home Offense = 115.4+(.014x115.4)=117.02
Duke's Home Defense = 80.7(.014x80.7)=79.57
North Carolina's Away Offense = 119.8(.014x119.8)=118.12
North Carolina's Away Defense = 87.6+(.014x87.6)=88.83
3. Find each team's adjusted pace (Possessions per game), and the League Average Pace from Kenpom.com's stats page
Duke's Pace = 71.5
North Carolina's Pace = 76.3
League Average = 67.6
Multiply these together, and divide by the league average.
Duke vs. UNC Pace = 71.5 x 76.3/67.6 = 80.7 possessions
4. For each team, multiply their Offense with their Opponent's Defense, then Divide by the League Average
League Average (just so happens to be)=100 points (per 100 possessions)
Duke's PPP=Duke O vs. UNC D=117.02x88.83/100=103.95 points per possession this game
North Carolina's PPP=UNC's O vs. Duke's D=118.12x79.57/100=93.99 points per possession this game
5. Multiply the predicted PPP (from #4) by the predicted game pace, and divide by 100 to get the final score
Duke's Points=103.95x80.7/100= 84 points
North Carolina's Points=93.99x80.7/100= 76 points
I still think UNC will win, though. :)
Ok, so....why does this work?
It actually has to do with fractions canceling out. Let's find out why:
The adjusted numbers are actually based on percentages. For example,
UNC's adjusted offensive efficiency is found by doing the following:
UNC's Avg Offense/Opponents' Avg. Defense x The League Avg Offense/Opponent's Avg Defense x Your Opponents' Average Defense
What this gives us is 3 numbers:
1. UNC's % of offense vs. opponents' defense
2. the league average offense% compared to UNC's opponents' defense
3. and the opponents' defense.
The first number tells us how well UNC has played against their opponents. The second tells us how an average team would play against UNC's opponents. Thirdly, again, we have the opponents' defense. If a team plays worse than the average team, they should be ranked lower accordingly. That's how the adjustments work.
When we cancel out the opponents' defense from the equation, we are left with:
UNC's average offense x the league's average efficiency / UNC's opponents' average defense. Look familiar? It's the opposite of our efficiency prediction calculation. (This calculation, by the way, works the same with defense, and pace.)
Now: the predicted efficiencies. Let's explain the opposite.
When UNC plays Duke, we compare UNC's offense to the average team, then Duke's Defense to the average team. The formula here is:
UNC's offense/League Average Defense x Duke's Defense/League Average Offense x Average Efficiency of all teams. This accurately portrays how UNC plays against the average, combined with how Duke plays against the average, in percents (since they are percents of the average, we then multiply by the league average).
This cancels out in much the same way, giving us UNC's Offense x Duke's Defense / League Average Efficiency.
There you have it!
Subscribe to:
Post Comments (Atom)
Blog Archive

▼
2009
(61)

▼
January
(17)
 ELRS
 Selection Sunday Predictions....
 WFU v. Duke, Part II: Weaknesses
 Duke v. WFU, UPDATED AGAIN
 16841
 One problem with kenpom.com
 nathan's sweet 16
 Clemson @ UNC
 Right now....
 Effective Efficiency
 Hmmmmm.....
 The Underr/Overrateds, and UNCDuke, continued
 Why Pitt would beat UNC in the NCAA tournament, bu...
 Wake Forest v. UNC pt II
 How To Predict The Final Score
 Wake Forest  North Carolina Game Preview
 Nathan's Normal Distribution Formula

▼
January
(17)
How often does this actually work?
ReplyDeleteExactly correct? Maybe 1% of the time...it's not magic!
ReplyDeleteIn terms of getting the exact point spread, Ken Pomeroy is probably 50%ish ATS.
Good stuff. Nathan, where do you get is the 1.4% adjustment to efficiency derived from? Thanks.
ReplyDelete