Bench Use and the Great Stats Debate

First off, I want to apologize for last night, because TJ Leaf’s injury was my fault. Last night was the first game I’d attended since March 8, 2008 (aka Josh Shipp’s circus shot to beat Cal), so I was clearly tempting fate by showing up to Pauley Pavilion and the basketball gods made sure to punish me for my hubris. Apologies to everyone involved.

That said, I think we need to have a talk about bench usage and when you should or shouldn’t go to your walk-ons. Suffice to say I read the entire conversation in Joe’s postgame and I am here to tell you every single one of you is wrong and I am the one and only arbiter of truth.

...

Ok, not really, but I will try and break down some thoughts on this.

Let’s Talk About James, Baby

My background is baseball, so let’s just say I am very keyed in to the whole sabermetrics debate, because it’s one the baseball community has been having for decades now.

Am I surprised that Bill James, the forefather of baseball sabermetrics, dipped his feet into basketball? Not remotely. Basketball has always been the next frontier for sabermetrics, and arguments that basketball people are currently having regarding efficiency, possible outcomes, and the like are arguments that baseball people still have.

Bill’s reasoning is sound mathmetically, and as the article points out, there was only one game in the 10,000 sampled that didn’t work with his created formula. That said, here’s the dirty secret that sabermetrics buffs like myself will constantly tell you: models are never perfect. Simply put, we create these models to better explain what should happen under perfect conditions, but we have to readily admit that human error can play a factor in the outcome. And occasionally a team comes around that flies in the face of existing models (to use baseball, the 2015 Royals were the anti-sabermetrics team a guy like Harold Reynolds dreamed of).

Now, in baseball, our models have remained fairly solid, in part because baseball itself has remained pretty static for decades. If you brought Babe Ruth forward in time to today’s game, he’d be able to figure out the minor changes rather quickly. This doesn’t work as well in basketball, because the game itself has undergone major changes to the rules in the past, such as the 3-point line and the shot clock.

Ironically enough, basketball sabermetrics have created a shift in how the game is played that creates some issues with Bill James’s model. In recent years, ideas from the sabermetrics community have taken root in the style of play in the NBA, with the rise of pace and space putting greater emphasis on the 3-point shot (including the corner 3, one of the most efficient shot in basketball).

How does this fit with the college game? Well, unlike football, where ideas tend to filter up from high school to college to pros, basketball tactics tend to move top-down from the pro game. Perhaps this is a byproduct of the one-and-done era, but there’s a reason college basketball pundits have gone crazy over UCLA’s offense and compared it to pro teams like the Warriors.

Now, I don’t think we’re necessarily at the point where the Bill James model needs to be revised, but consider the graph in this article from earlier this year. James didn’t have to account for the large rise in efficient offensive play over the past few years, and while it wouldn’t have affected this game in the same way (Washington, with their AdjO efficiency of 108 and missing Fultz wouldn’t have fit inside that chart), it would be interesting to see the article revised to account for this shift.

The injury effect, or developing a bench

Part of the reason the hysteria over whether or not Steve Alford left the starters in too long exists at all happens to do with TJ Leaf. Frankly, if TJ Leaf doesn’t get injured less than 5 minutes into the game in a freak accident, I’d think the furor over Alford’s substitution pattern in this game would be as pronounced.

That doesn’t mean we shouldn’t talk about it, though.

It shouldn’t be much of a reach to say that the more minutes you put in a game, the greater the risk of injury. Seems simple, right? This article goes into detail from a physiological perspective, but most of the base points should be pretty obvious. And, from an NBA perspective, this article from a few years ago made a correlation between the number of minutes played by the Bulls in garbage time and their injury woes.

Now, I don’t think there’s as big of a straight line from that Bulls article to the college game as one might think. Taking out the obvious differences like the longer games, you have to consider the relatively stable game schedules in the college game compared to the pros. For UCLA, the standard has been (for the most part) a game on Thursday and a game on Saturday. Obviously, this can shift in different directions, such as how the Washington game was on a Wednesday, but for the most part, you can reasonably expect at least one day of rest between games, and the relative stability of the schedule in the college game allows for better gauging of rest for players that can’t exist in the pro level, where schedule density can cause issues.

There’s a more interesting argument to be had regarding bench development, and for this I’ll reference this article by Will Schreefer. In it, Schreefer discovered having a deeper bench is not necessarily a recipe for success in the postseason; if anything, shortening your bench is usually a better indicator that you’ll outperform expectations.

Schreefer notes that having a deeper bench is more important in the regular season, if only to combat fatigue over the longer stretch of games, but there’s an underlying principle to Schreefer’s analysis: the starters are going to be better than the bench players by a considerable margin. This is not to say that there aren’t exceptions; after all Aaron Holiday exists on this UCLA team, but it is a general indicator of quality.

So considering that, how much in-game time should you realistically allocate to help develop bench players? Honestly, it depends. For some younger players, you may want to start them with shorter minutes, if only to get them acclimated to the game. Other players may benefit more from a sink-or-swim model. With the one-and-done model, there seems to be less emphasis in general from top teams towards developing their bench, instead focusing more in-game minutes on getting starting units and key bench players time together to gel (I’m basing this off the fact that UCLA, Baylor, Villanova, Kansas, and Duke all sit inside the bottom 50 of KenPom’s bench minutes metric).

To that end, I’m not surprised Steve Alford chose to go so late in the game before going to the walk-ons fully. Losing Leaf as early as they did provided the Bruins an opportunity to get a solid 30 minutes of in-game time to learn how to play and run the offense without Leaf. It’s a similar situation to earlier this season, where the Bruins were without the services of Thomas Welsh for a 3-game set, except figuring out how to play without such an integral player like Leaf required as much in-game time as possible. Even with the blowout, in the moment you don’t know the long-term prognosis for Leaf and it’s a risk I’d be willing to take.

There’s an alternate theory that UCLA needed the extra time to continue working on the defense, and while I agree, that wasn’t the pressing concern in this game, especially once Washington’s biggest offensive threat was ruled out.

Doing What You’re Taught

Let’s talk about this statement from Steve Alford after the game:

I alluded to this earlier, but there’s always been a bit of a power struggle between statisticians and coaches. Part of this has to do with human nature - we do what we’re taught. For a lot of old-school coaches, there’s a clear hesitance towards changing how you approach things, especially if that method has worked out for you in the past.

For Alford himself, his substitution patterns in this game were consistent to how they have been in the past, and his logic has been consistent each time, citing competitive spirit. But it does raise the question on how that compares to other coaches in the NCAA. Bruinette brought up Dana Altmann and his bench usage in a blowout win against Colorado here, so I’ll go look at some other spots.

How about Mark Few of Gonzaga? The Bulldogs had a similar late-season blowout victory against an outmatched conference opponent last week, when they beat San Diego by 58 points. In that game, Few took the last starter out at the 7:19 mark of the second half, though the team didn’t take out all regular contributors until the 3:23 mark. So Few definitely falls into the “take them out early” camp.

How about North Carolina coach Roy Williams? Looking at their recent blowout of Pitt, the Tar Heels had achieved an insurmountable lead at the 4 minute mark according to Bill James. Yet, Williams didn’t sub completely remove the starters until the 1:43 mark of the game, and didn’t go to the walk-ons until 0:57. Coincidentally, Williams’s move to the bench coincided with Pitt’s similar move, suggesting Roy Williams and Steve Alford probably agree on when to go to the walk-ons in a conference game (and looking through minute logs, this seems consistent for North Carolina regardless of the margin of victory).

Finally, one more: Kentucky’s John Calipari. Looking at their late-season blowout of Tennessee, Calipari waited until 1:30 left in the game to take his starters out. At that point, the Wildcats were up by 25 points, and even then, only one walk-on managed to make the floor during this sub: the coach’s son Brad. Every other player was a rotation player averaging over 8 minutes a game.

(And if you’re wondering why I didn’t compare the current top 2 teams in the AP, it’s because Kansas never blows teams out, and I couldn’t find substitution info for Villanova’s recent blow-out of Seton Hall, though going off the box score would lead me to say that Jay Wright left starters in until the sub-minute mark.)

If anything, this should show that there isn’t much of a consensus as to when you should pull your starters among even the best college coaches. That’s hardly surprising, and the clear separation between when many of these coaches take their foot off the gas pedal and when sabermetrics says you should isn’t surprising in the grand battle between the two sides.

Does it all really matter?

No. Nothing I have written has ever really mattered.

But to go meta on you all, it’s honestly hard to say. Like I said at the beginning, there isn’t a clear-cut correct answer to the question of when you should call the dogs off. And on that same regard, it’s why debate on the topic can get so heated. Both sides have legitimate points that I tried to cover, and if I didn’t, it’s because I don’t care about your argument I legitimately forgot about it.

So again, I am correct, and everyone should listen to me.

Go Bruins!