Given the small sample of games in a college basketball season, even if every player returns, a team’s defense is surprisingly unpredictable. As an anecdotal illustration of that point, in last week’s column I noted that North Dakota St. brought back essentially its entire roster in 2014, and yet NDSU’s defense was much worse than in 2013. But brilliant South Carolina Gamecock writer @chickenhoops emailed me and pointed out that North Dakota St. wasn’t necessarily a worse defensive team last year, they were just unlucky. The Bison’s collapse on defense was almost entirely driven by factors outside their control. Notably, North Dakota St.’s opponents hit an incredibly high percentage of three pointers, and made an extremely high percentage of free throws on the season. I had noted that North Dakota St.’s defense fell from 59^th to 131^st. But ChickenHoops noted that if opponents had made the same percentage of threes or the same percentage of free throws as the average D1 team, North Dakota St.’s defense would have ranked 70^th in 2013 and 74^th in 2014.

Most people would agree that teams have little control over their opponent’s free throw percentage. And Ken Pomeroy has argued that an opponent’s three point percentage is something teams also have little control over. (Ken argues that teams have control over whether opponents take threes, but have substantially less control over whether those shots fall.)

And as ChickenHoops proposes, we can easily recalculate each team’s points allowed per 100 possessions, assuming their opponents hit the D1 average percentage of free throws and threes. This new measure of defense will be the points allowed per 100 possessions, assuming more typical luck. In the next table I show how each team’s defense would have looked last season using this new measure.

In the major conferences, a team like Vanderbilt stands out as fortunate last year. While the average team made roughly 70% of its free throws last year, Vanderbilt’s opponents made only 65% of their free throws. And while the average team made 35% of its threes last year, Vanderbilt opponents made only 30% of their threes.

On the flip side, a team like Notre Dame was probably very unfortunate to have such a bad defensive season last year. Opponents made 75% of their free throws and 39% of their threes against Notre Dame last year, far above the national averages.

Though not listed in the above table, Wichita St. was also extremely fortunate last year. The Shockers opponents made only 65% of their free throws and 31% of their threes last year. Wichita St.’s opponents rarely had scorching shooting nights, and when they did, (Evansville making 5 of 11 threes and 15 of 16 free throws), the game was enough of a mismatch that it didn’t matter. The Kentucky game was arguably the first time on the season that one of Wichita St.’s quality opponents had an unusually good shooting night.

This is worth revisiting in the middle of next season, to see if the margin-of-victory rankings are misleading us about the top teams. But it can make a meaningful difference. With this alternative metric, Notre Dame was not the 99th best team in the nation last year, but they were the 82nd best team.

Another way to think about this is to think about performance between seasons. A team’s adjusted defense is correlated year-to-year. But if I use this new measure of defense, the year-to-year correlation is meaningfully higher. The improvement in the unexplained variation is a bit difficult to put into words, but perhaps a comparison will help. One of the factors that matters in my model is the height of each team’s center. Using this new measure of defense improves the performance of my prediction model four times as much as adding the height of each team’s center to the model.

Since my prediction model is a bit complicated, for now let’s think about the simplest possible prediction model which accounts for returning minutes (but which does not account for individual player stats, player heights, recruiting rankings, or coaching). I’m discussing results based on the last 10 seasons of data. In this simple prediction model, if a team brings back the average number of minutes, having a defense that is 1 point better predicts that the defense will be 0.67 points better the following season. But when we use this new measure of defense, a defense that is 1 point better predicts that the defense will be 0.72 points better the following season. For a team on the unfortunate end of the scale, like Notre Dame, using the new measure of defense essentially moves them up about 12 spots in my projections for next year.

It is not fair to say that teams have no control over their opponent’s free throw percentage. FT defense depends on which players a team fouls. And Pomeroy has admitted that three point percentage includes some information. But college basketball analysts should be thinking of this a bit like Batting Average on Balls in Play in baseball. If a team is way too high or way too low in free throw defense or three point percentage defense, that probably is a bit about luck. Both within seasons and between seasons, we shouldn’t necessarily expect that component of defensive performance to persist.