Nylon Calculus: The (updated) math of transition defense

Transition possessions are some of the most efficient in basketball. What causes one team to be better than another at preventing them?

A lot of work has been done historically on the relationship between offensive rebounding and transition defense. Our own Seth Partnow had an article which established, among other things, that merely positioning a player near the ball for a chance at an offensive rebound increases the efficiency of the other team in transition on the subsequent possession. Kevin Ferrigan, when he did the regression for Daily RAPM Estimate, found that if he split out offensive rebounds, they didn’t come out as statistically significant, and one of the proposed reasons was the decline in transition defense. Even as a general trend across the NBA, offensive rebounding rates have cratered and the 46 best defensive rebounding teams ever have all happened within the last ten years.

But that study was two years ago, and the available data has changed. Seth had access to older SportVU positioning data, which is not made publicly available anymore. But he also was forced to define transition by proxy — using the time at which the shot occurred to define a transition possession. With Synergy data publicly available on stats.nba.com, there’s access now to a more precise definition of a transition possession, and so it makes sense to readdress the topic.

Since I don’t have access to that SportVU data, though, I’m going to have to use a slightly more imprecise method for quantifying rebounding aggression.

To start, what if we regress offensive rebounding percent onto the opponent’s points per possession in transition? The sample here is the last three years of teams, minus two that just randomly don’t appear in the database, for a total of 88 team.

The result is statistically insignificant at any meaningful threshold.

But what if the teams that are getting more offensive rebounds aren’t actively pursuing them? After all, it would make sense for more offensive rebounds to come from at least two sources: Actively pursuing them and rebounding skill. So we need to create a control for rebounding skill, and total rebounding percent makes the most sense. It avoids rewarding teams that explicitly crash the glass on one end (Looking at you, Steve Clifford) and allows us something of an unbiased estimator towards the total ability without any scheme based interference.

Well, by regressing offensive rebounding percent and total rebounding percent onto the opponent’s points per possession in transition, the results are a little more clear.

If a team were to hold its total rebounding percent constant, while increasing offensive rebounding percent, you would expect an increase increase in the efficiency of their opponents in transition. That constant assumption doesn’t necessarily hold in reality, where usually if you’re getting more of the offensive rebounds, you’re also getting more of the total rebounds. But since the coefficient on total rebounding percent is easily more than twice that of offensive rebounding percent, and there’s typically about double the amount of shots when you account for both ends of the court I’m comfortable with stating that even if total rebounding increased at a corresponding rate to the offensive rebounds (Not at a straight one-to-one ratio, but at the near fifty percent rate that virtually any NBA team will hold since their opponent will get to take shots), there would still be a clear detrimental effect.

But now, what about adding turnovers into the mix? Using the same transition proxy in a different article, Seth determined that following a steal, teams had a significantly higher effective field goal percentage early in the shot clock.

Well, we’ll start in the same way we did with offensive rebounds and just regress a team’s turnover percent onto their efficiency in transition.

Like before, we get that it isn’t statistically meaningful. But in Seth’s article, he was specifically looking at steals, so I’ll have to separate out live ball turnovers and dead ball turnovers. In order to do this, I’m taking opponent steals, dividing by total turnovers, and multiplying that by the total turnover percent. That gives a live ball turnover percent, and subtracting that from the total turnover percent gives the dead ball number.  So then turn around and regress those to efficiency.

Unlike the first time around though, both of them come out meaningless, in contrast to previous studies.

But we’ve only been looking at how efficient teams are in transition. It makes sense that when a player crashes the offensive boards, that the subsequent transition attack into fewer players is more efficient. What if we were to look at how those things have an effect on the rate at which teams get out in transition? After all, that’s one of the advantages of the new play type data — having that already split out.

Well, keeping our total rebounding percent control on offensive rebounding percent, the result is not statistically significant at 95 percent.  

It appears that teams don’t necessarily proactively look for the transition opportunity when their opponent crashes the boards, they’re just more successful when they do decide to go because they’re attacking into fewer people. Given the p-value’s close proximity to the threshold, and even hitting a moderately common threshold at 90 percent, this effect is not as clear in the data as some of the other conclusions in this article, but since, in my experience, sports data very rarely needs a 90 percent threshold, I’m still confident asserting that as the case, but I’m certainly not going to war over it.

But for turnovers?  Regressing turnover percent onto transition frequency is clearly statistically significant.

Even splitting it out into the previous live ball and dead ball numbers like we did shows up with both of them as significant.  

Which I actually don’t have a great explanation for the second part of, since it doesn’t make sense that a team that commits a higher rate of dead ball turnovers would also give up more transition opportunities without some third factor at play.  That third factor may be a general carelessness with the ball, since our live ball turnover percent is correlated to dead ball turnover percent, but regardless.

All that to say, the newer data makes it quite clear that the prior studies establishing both a trade-off between offensive rebounding to transition defense and the compounding negative effect of steals on transition defense were correct.  Further, with that new data, we can more directly see the exact mechanism by which that happens. With offensive rebounding, teams don’t conclusively present more active opportunities to leak out in transition by crashing the boards, they just increase the quality of existing ones. In contrast, with turnovers, they don’t create higher quality transition opportunities for the opponent, they just create more of them, and since transition opportunities are already some of the most efficient looks in basketball, the overall result is bad. In either case, both hurt the overall defense in a clear and statistically demonstrable way, and that’s why the NBA is prioritizing reducing both of those things more and more.