The practice of adjusting offensive and defensive performance to opponent strength was popularized by basketball statistician Dean Oliver in his 2002 book, Basketball on Paper.
Now an assistant coach with the Washington Wizards, Oliver exclusively analyzed the NBA. But his method was actually better suited to college basketball, which struggles to characterize itself due to the broad diversity of its 362 participants.
In Oliver’s method, you estimate how one team’s offense would perform against another team’s defense by multiplying their respective points scored and allowed per 100 possessions, using the league average as a baseline.
A margin of victory is estimated. And considering that not all teams are created equal, the better teams are expected to beat the lesser teams by a certain margin.
If you overperform? You move up in the analytics. If you underperform, your adjusted efficiencies decrease and you move down, even if you win the game.
Although Kenpom has been publishing AdjEMs for almost 25 years, when I met him in 2008 he was focused on optimizing in-game probabilities; the underlying adjusted efficiencies were just the tool that made that analysis more accurate.
There still wasn’t broad interest in adjusted efficiencies until 2018, when the NCAA added it to the Selection Committee Team Sheets. For the record, I think this was a mistake. I also disagree with the practice of ranking teams by their AdjEM, in other words, their “Kenpom rank.” But I’ve made my argument, and it’s not going to stop.
But I do want college basketball fans to understand that (1) the calculations are straightforward, and (2) there isn’t one right way or one right answer. And I’ll illustrate using the data from Providence’s recent game against Wisconsin. Why that game? Because it was so damn FUN.
Let’s start with the pregame adjusted efficiency data:
| Pregame | Offense | Defense |
| Providence | 107.9 | 95.8 |
| Wisconsin | 112.4 | 93.3 |
Up to that point, Providence had scored 107.9 points per 100 possessions, Wisconsin allowed 93.3, and the NCAA average was 103.
The NCAA average efficiency is what the most average team is expected to score and allow on a neutral court. But in math terms it means the value where the cumulative offensive and defensive performances across all games among all teams balances out.
Now let’s apply Oliver’s formula to estimate the outcome of Wisconsin at Providence:
For Providence on offense: (AdjustOff*AdjustDef/ AveEff) or (107.9*93.3/103.0); we expect Providence to score 97.7 points per every 100 possessions against Wisconsin on a neutral court.
We then multiply the 97.7 PPP by the estimated number of possessions (72), add 3.2 points for Kenpom’s estimate of home court advantage, and you get Providence’s estimated points scored in the game, or 73.57.
Similarly, for Wisconsin: [(112.4*95.8)/103]*72 = 75.27
The pregame prediction was thus 76-74.
This is an oversimplification so you can see how AdjEMs are used to predict game scores. There is an additional step using a formula popularized by baseball statistician Bill James, which I wrote more in detail about here.
Another caveat is that Pomeroy doesn’t actually follow the original Oliver method of adjusting efficiencies. He found that multiplying efficiencies can overemphasize results at the margins. Great teams will appear even better and the worst teams will appear even worse.
In order to narrow this spread, Pomeroy switched to an additive system. Instead of multiplying, he adds or subtracts, again using the D1 average as a baseline.
Let’s redo using Pomeroy’s method:
Providence adjusted offense – average adjusted defense; 107.9 – 103 = 4.9
Average adjusted offense – Wisconsin adjusted defense; 103 – 93.3 = 9.7
Providence is +4.9 points per 100 better on offense and Wisconsin is +9.7 points per 100 better on defense. Simply put, Wisconsin’s defense (on paper anyway 😛) is better than Providence’s offense. So we add Providence’s spread over average to Wisconsin’s defense to get Providence’s estimated total points scored per 100:
93.3 + 4.9 = 98.1
The assumption here is that against the D1 average, Wisconsin will allow 93.3 PPP. But since Providence is 4.9 PPP better than the D1 average, they will score an additional 4.9 points per 100 possessions.
Multiply by the number of estimated possessions, 72. Then add 3.2 points for home court and we get our estimated total of 73.8 points scored.
The end result is the same, but you see how the varying methods produce slight differences in the calculation. In the original Oliver method, Providence was estimated to score 73.57. In the Kenpom method, 73.8.
The adjusted efficiency can also be calculated if expressed in percentage terms.
Providence’s offense at 1.08 is 4.8% better than the average of 1.03, Wisconsin’s defense at 93.3 is 9.4% better than average, so Providence should score (9.4-4.8) or 4.6% more against Wisconsin than the average, which is also 98 points per 100 possessions.
A final note about Torvik. For one, he still uses the original Dean Oliver method to adjust for opponent strength. Another difference is his treatment of home court advantage. Finally, recency is weighted differently.
These methodological differences are why Torvik’s version of AdjEM and resulting game predictions vary from Kenpom. This also underscores that there is no single correct or universally accepted method. Kenpom isn’t necessarily the best, he was just here first.
Post Game Bump
Providence 72 Wisconsin 59 in 67 possessions
As it turns out, the pregame prediction was wrong. No one from Wisconsin showed up except for AJ Storr, while Devin Carter and Jayden Pierre played like the splash brothers.
The unadjusted result was 1.07 PPP for Providence and .88 for Wisconsin. But Wisconsin was (is?) good, 25th in Kenpom. So how do we adjust the raw PPP in order to give Providence more credit in the analytics?
First, the formula for adjusting raw PPP data for an individual game based on opponent strength:
Adjusted offense = PPP scored*(Opponent pregame AdjDE / ave PPP)
Adjusted defense = PPP allowed*(Opponent pregame AdjOE / ave PPP)
1.07*(93.3/103) = 1.18 Providence’s adjusted points scored, an increase from 1.07
.88*(112.4/103) = .80 Providence’s adjusted points allowed, a decrease from .88
So although Providence finished with a .19 PPP advantage in real terms, they get credit in the advanced metrics for a .38 PPP victory. In other words, as far as Kenpom is concerned, Providence won the game 86-58(!).
| Offense | Defense | |
| Providence | 108.6 | 94.3 |
| Wisconsin | 110.8 | 93.9 |
Providence’s AdjEM jumped 2.4 points per 100 after the game (1.7 improvement on defense and .7 on defense), which is a lot – about 15%. And Providence’s rank consequently also improved, from 57th to 46th.
So now, you don’t need to refresh Kenpom after the game to see how Providence’s AdjEM and subsequent rank change, you can do the calculation yourself!
That’s all for Part 1. Part 2 next week I’ll go into detail on the actual game simulations, both Kenpom’s and my own.
Wisconsin at Providence
10,000 simulations using Torvik’s data from preseason through now.
Average score
Wisconsin 75 (-.5 o/u 148.5)
Providence 74Providence wins 56.6% of outcomes due to home court advantage. #pcbb #GoFriars pic.twitter.com/q4Yht99oKI
— FriarStats (@StatsFriar) November 14, 2023