Were advanced statistics the reason why Brad Stevens gave veteran Kris Humphries his first start of the season over Jared Sullinger on Tuesday night? Perhaps. Humphries has played extraordinary basketball in January, averaging almost a double double in only 26 minutes per game, which could explain why Stevens started him.
Unfortunately, Sullinger is battling through a left hand injury and has seen his performance on the court drop like a rock. He's shooting only 29.6 percent has seen his playing time dip every game this month.
Despite his struggles as of late, the move came out of nowhere considering Brandon Bass and Sullinger had started 21 straight games alongside each other.
That's why the lineup change triggered a few thoughts in my mind:
- What frontcourt pairings have been most effective for the Celtics this season?
- What analytics being fed to Stevens could've influenced the starting lineup change?
- Who should start at power forward and center?
So, I decided to conduct an analysis to attempt to answer those questions. In this article I will look specifically at how the Boston Celtics' frontcourt combinations have performed together this season before I offer lineup adjustments based on what the statistics suggest.
This analysis will make heavy use of an advanced statistic called PPP, which is short for Points Per 100 Possessions. In the simplest terms, PPP is used to detail how many points a team/player scores or allows per 100 possessions.
More specifically, PPP can be calculated to see how a team performs when a specific player or group of players was on or off the court.
100 points is roughly the average amount of points scored/allowed per 100 possessions, so any offensive PPP higher than 100 means more points than average were scored, whereas a defensive PPP higher means more were allowed.
For example, last season Boston's defense allowed 98.9 PPP when Kevin Garnett was on the floor, but 108.9 PPP when he was off. That differential can lead us to assume that the Celtics were significantly better when he was playing.
In this study, PPP will be used to describe how the Celtics perform as a team with a certain players on or off the floor at the same time this season. Graph 1 to the right details what the Boston Celtics offensive and defensive PPP are so far this season as a whole. They are scoring 102.1 points per 100 possessions, and allowing 105.9 points per 100 possessions. The offensive PPP is right around the league average, but the defensive PPP is below average.
The best way to continue this breakdown is to simply look at how the Celtics function with a specific big man on or off the court. These statistics are listed in Graph 2, shown below.
Three sets of numbers really pop off the stat sheet to me. First, it appears the Celtics really aren't any good offensively when Vitor Faverani is on the floor. They score only 93.7 PPP when he's playing compared to 104.9 PPP when he's off. That's an astonishing difference, and it might explain why he has seen his playing time drop significantly this season.
I will be eliminating Faverani from the pairing section of the study because his play is so subpar statistically. It also simplifies the amount of data to come in the next section, which will benefit the readability of the article.
It surprised me very much to see how much more efficiently Boston scores with Brandon Bass off the floor. They score a below average 98.6 PPP with him on the court, but a respectable 107.5 with him on the bench. Considering Bass' early season success, I expected his PPP splits to stay relatively constant.
This is contrary to the numbers indicated by Kris Humphries' PPP. The Celtics score 108.3 PPP with him on the court to only 99.6 with him off. That data is basically the reverse of Bass', which again, could be a reason why Humphries has seen his playing time rise in recent weeks.
Lastly, the defense is fairly consistent no matter who is on the floor. But the stats do indicate that the defense is at its worst when Sullinger is off the court, allowing 108.9 PPP, and at its best with him on, at 102.9.
But these statistics can't explain everything. There are so many variables to account for considering each Celtics big man plays alongside different teammates and therefore faces certain types of lineups.
That's why it's necessary to look more closely at the groupings of players. Brad Stevens had his statistical guru, Drew Cannon, do just that at Butler, and now he's doing the same with the Celtics. It's important to remember that statistics can't paint the big picture, but they can provide you with some of the colors necessary to understand some of the fine details.
This is where the study gets a little more complex. As I said in the introduction, I want to find out what pair of bigs (power forward and center) produce best together, so I calculated what the PPP was for all 18 possible lineup combinations. Stats listed include combinations of players (IE: Bass and Sullinger), but also how the stats differ with one of them off the court (IE: Bass and no Sullinger, or Sullinger and no Bass).
Don't be alarmed by the amount of numbers on Graph 3, it's not as overwhelming as it seems. The first two columns indicate when a player is on the court, the third column is when one player is off the court, the fourth column is the team's offensive PPP, and the final one is the team's defensive PPP.
Boston's defense really hurts when Brandon Bass is paired with Kelly Olynyk or Kris Humphries, since the team allows a dismal 116 PPP when either combination is on the floor.
However, the Celtics do only allow 102.1 PPP when Bass is paired with Jared Sullinger, which happens to be the ample sample size.
Other than that, the defensive combinations are fairly consistent according to the statistics, with the high being 103.5 (Sullinger and Olynyk) and the low being 101.4 (Humphries and Olynyk).
Interestingly, Kris Humphries is involved with the three best offensive combinations, since the team averages over 111 PPP no matter who he's paired with. Meanwhile, the team struggles, scoring 97 PPP, with the combination of Sullinger and Bass in the game.
Looking at the stats when one player from each pair is removed from the court implies that the team executes drastically better offensively without Bass than it does with him. For example, the Celtics score 104.4 PPP with Olynyk on the court and Bass off compared to only 97.8 PPP with Bass on and Olynyk off.
The Celtics' numbers are also better when Sullinger or Olynyk are paired with a player other than Bass. The team plays especially well when Olynyk is paired with Humphries, with 112 PPP.
The stats suggest Bass isn't a fit
Brandon Bass and Jared Sullinger had started 21 games straight together this season until Sullinger was benched on January 8th against the Clippers. The combo has played 501 total minutes together and defended fairly well, only allowing 102.1 PPP.
It's when looking at the offensive production things get interesting. When Sullinger and Bass are on the court together, the Celtics score a subpar 97 PPP. But when Brandon Bass is off the court and Sullinger stays on, that number skyrockets up to 115.4 PPP (a similar differential appears when Bass is paired with Kelly Olynyk or Humphries, see Graph 3 above).
I'll admit, that surprised me quite a bit. I thought the team played slightly better using combinations not including Bass, but a difference of 18 points is staggering. These stats indicate that Bass and Sullinger is not a very good combination to utilize in the starting lineup.
Of course, Brad Stevens changed things up on Tuesday night by starting Humphries over Sullinger. But why not bench Bass when the statistics heavily suggest that the team is worse with him on the floor?
Other than the fact Sullinger has struggled offensively since injuring his left hand, I'm having a hard time finding an answer. Even though the Celtics score a terrific 111.8 PPP when Humphries and Bass are paired together, the defense gives up a lackluster 116.2 PPP.
When Bass is paired with Sullinger, the team can defend, but it can't score; with Humphries they can score, but they can't defend; and with Kelly Olynyk, the Celtics struggle on both ends of the floor. To put it plainly, the stats confidently indicate that Brandon Bass doesn't fit well alongside any big on this team.
Humphries is a glue guy
After Wednesday night's loss to the Clippers, Brad Stevens said, "Hump's just playing at such a good level ... He's been very productive. The more minutes that we have him on the floor, I think it's better for our team right now."
That's for sure, and the numbers are indisputable, which can clearly be viewed in Graph 4 to the right. Humphries is averaging 14.3 points and 10.5 rebounds per 36 minutes this season, but also a career-high 2.2 assists. The offense plays very well no matter who Hump is paired with in the frontcourt, and so does the defense -- unless he's on the floor with Bass.
When the offense gets stagnant with Bass holding the ball for too long, Humphries instead provides a better flow; and when the pick-and-roll defense gets beat with Olynyk, Humphries would be there to contain the penetration. Humphries' per-36 numbers might level off as he continues to receive increased minutes, but it's obvious that he's playing at a high level right now.
Coach Stevens clearly made the right decision inserting him into the starting lineup considering his ability to adequately play alongside any other big man on the roster. For the Celtics to have success this season, they must find consistency and stability. So far, the stats suggest that Humphries is the answer for that problem.
What should the rotation be according to the stats?
Remember back in the preseason when almost everyone expected Brad Stevens to pair Kelly Olynyk with Jared Sullinger? Well, that happened for a little while when Vitor Faverani had his stint starting alongside Bass and Humphries got no playing time, but it didn't last very long. Olynyk got hurt on November 22nd and didn't return for almost three weeks. Since then, the two youngsters have played only 41 minutes together, even though they are still putting up numbers consistent with what they did early in the year.
Using the statistics found as a basis for my suggestion, I would propose to Brad Stevens that he should start Kris Humphries alongside a healthy Jared Sullinger. If the statistics accurately portray court performance, this would upgrade the Celtics on offense due to their scoring efficiency. Humphries and Sullinger would also provide the Celtics with a prominent rebounding duo, which would cut down on second possessions for opposing teams, improving the defense statistically.
The data would then suggest that Kelly Olynyk should come off the bench for Humphries. This would be used to take advantage of his effectiveness playing alongside Sullinger, as the Celtics score 110.3 PPP when they're both on the floor, and allow only 103.5 PPP.
Brandon Bass could then be substituted in for a fatigued Sullinger to play with Olynyk for a short period of time until Humphries was ready to sub in for the rookie. And, of course, at one point, the starting Sullinger would then come in for Bass.
Using the PPP stats as a guideline, I think it would be best to see Sullinger and Humphries hover around 30 minutes per game, and for Olynyk to receive about 20 depending on his effectiveness on a night-to-night basis. Bass would then receive any additional minutes, usually between 15 and 20.
What does this all mean?
Everything or nothing, depending on how much you value advanced statistics. Personally, I will use what I learned from this study and apply it to what I see on the court in upcoming games. I have never been a "stats geek," since I think the best way to analyze a team is with your own two eyes, with math added in to support your theories on a particular player, lineup, or team. In any case, these findings do bring up some compelling ideas, since they could explain some moves made by Stevens, or changes to be made in the near future.
One issue holding this study back is the small sample size for a few of the lineup combinations. For example, Boston has played 501 minutes with both Sullinger and Bass on the floor for a total of 973 possessions, which is a completely reliable sample size.
On the other hand, the Celtics appear to play well when Olynyk and Humphries are both on the court, scoring 112 PPP and allowing only 101.4 PPP, but the sampling is quite small. Those two have only been on the court together for only 108 minutes or 216 total possessions, and the majority of that time has been alongside Gerald Wallace, not Jeff Green.
Speaking about the small forward position, it's worth mentioning that I was originally going to include a "with Jeff Green" column for all the previous statistics used, but I found that there was no significant difference offensively or defensively if the small forward was Green or Wallace. If you're curious to see the full analysis with those numbers, click here. Keep in mind that the numbers including Faverani are often a misleadingly small sample.
In the coming weeks I'm expecting Brad Stevens to continue tinkering with his big man rotation, just like he did early in the season. The Boston Celtics are riding a six game losing streak and it's time for him to play around to find out what works best. Perhaps, Stevens' stat guru, Drew Cannon, will pass along some statistics similar to the data used in this article to influence some changes.