Basketball is in many ways a simple game. Score more points than the other team, you win. Score fewer, you lose.
All of which makes analysis of NBA plus-minus data an informative -- albeit complicated -- endeavor. Revisiting an exercise from last spring, I thought it'd be interesting to comb through both ESPN's real plus-minus and NBA.com's simple plus-minus data to see what it tells us about both the Bucks' roster. For a refresher on what RPM is, here's what we wrote last spring:
You can read a lot more about what RPM is here and here, though at a high level it's essentially a way to take the raw +/- data you now find in box scores and (using a huge, complicated regression model) strip out the biases that come from playing with different teammates and against opponents of varying quality. It's also intended to be predictive, though glancing at the data from the past two years suggests it can swing considerably over time. And so it goes.
In short, no one will claim RPM is perfect, or for that matter simple. ESPN doesn't provide standard errors for its RPM estimates, and for guys with low minute totals they would likely be quite high (possibly to the point that their RPM figures are mostly worthless). It's also somewhat abstract; a couple analytics guys who are far smarter than I am told me at the Sloan conference in March that RPM isn't something they would regularly cite in conversations with a GM -- though outlier situations such as Middleton's are always worth mentioning. In that regard raw on/off data is still an easier concept for most to grasp, and by netting out how a player does compared to his team without him we can at least implement some crude method of control.
[To further complicate matters, RPM also uses a box score-derived "prior," which is essentially a way of giving the model a starting point not dependent on pure plus-minus data. Rather than assuming all players are league average to start, RPM uses a stat similar to the box score plus-minus (BPM) you can find at Basketball Reference to help parse the uncertainty inherent in lineup-related data. Thus, RPM is ultimately a mixture of lineup regression and box score-derived data.]
Either way, neither raw plus-minus nor RPM should be used as a be-all-end-all of a player's contribution, and it's inherently problematic to use as a way of ranking players league-wide. Still, they are very useful metrics to have in your toolkit for assessing a player's overall impact. Below is a table showing net RPM as well as the Bucks' offensive and defensive ratings when each player is on the court, off the court and the net difference between the two.
Next, here's an RPM chart with offensive RPM on the horizontal axis and defensive RPM on the vertical axis -- green dots for guys who are positive on both ends, orange for players who were pluses on one side and negatives on the other and reds for guys that hurt on both ends:
Khris Middleton: A year ago we wondered whether Middleton's massive RPM improvement was "real"; a year later his defensive data remains rather noisy (it's gone from -3.50 to +4.11 to +0.73 in the last three seasons), but his offensive impact appears very real. Not only did Khris' offensive RPM continue to rise (from +1.22 to +1.96 to +2.98) but the Bucks were an utter trainwreck with him off the court as well (95.8 points/100 possessions on the bench, 104.7 with him on the court). Still, the fact that Middleton was the Bucks' best raw plus-minus guy and still was a slight negative on the season (104.6 ORTG, 104.7 DRTG) speaks volumes about the Bucks' overall struggles.
Giannis Antetokounmpo: While Antetokounmpo's aggregate numbers aren't eye-popping, he was still a net positive by RPM (+0.83 ORPM, +0.23 DRPM, +1.06 total) and in net on/off terms (-2.6 pts/100 on court vs. -5.9 off court). Perhaps more importantly, Giannis' numbers from January on were significantly better than those he posted in the first two months. In calendar year 2016, Giannis lineups outscored opponents by 1.0 points/100 while Bucks lineups without Giannis were disastrously bad (-13.4 pts/100), and starting units with Giannis were particularly impressive. Moreover, the fact that Giannis finished the season with a positive RPM at all is rather remarkable considering he was -1.56 when we last looked at this data in late November.
Greg Monroe: Surprised to see this name in the "good" section? Monroe is one of just three Bucks who finished with a positive in both offensive and defensive RPM, which might raise some eyebrows given the Bucks were demonstrably worse defensively with him on the court (107.5 DRTG on, 103.2 DRTG off). Some of this is likely an artifact of DRPM figures always being higher for big guys -- Monroe ranks just 61st out of 77 centers -- though in Monroe's defense he's not that much worse than John Henson (+0.93) and notably better than Miles Plumlee (-0.79). Either way, Moose's offensive contributions are undeniable, as he led the Bucks in points per minute and Milwaukee overall scored 6.4 more points with him on the court. In other words, moving Monroe this offseason might be the best move fit-wise, but it would still involve trading a productive player.
O.J. Mayo: No one would have pegged Mayo to lead the Bucks in DRTG (102.3) and finish second in DRPM (+0.80), especially given the prevailing narrative this season was his generally abysmal performance on the offensive end. But before breaking his ankle and vanishing over the final month of the season, Mayo was quietly the perimeter glue of the Bucks' defense and ranked third on the Bucks in overall net differential (3.2 points better on vs. off). The tea leaves seem to suggest the Bucks won't look to bring Mayo back, but maybe they should. Long live Uncle Juice!
Rashad Vaughn: Vaughn's historically bad rookie season has been well-chronicled, and not surprisingly it also bears itself out in his team-worst RPM. The Bucks were terrible offensively with him on the court and marginally worse defensively as well. Most rookies are bad; let's hope Vaughn is a different guy in the fall.
Johnny O'Bryant: There was a month or two early in the season when JOB was (remarkably) the Bucks' leader in net rating terms. Predictably, that did not last. By season's end JOB had fallen back to earth in raw plus-minus terms (-5.7) and his RPM was once again abysmal (albeit slightly better!) for a second straight season. He was the Bucks' worst player in ORPM terms (-3.71) but more respectable defensively (-1.23), which feels about right.
Greivis Vasquez: You can't say "bad" and "Bucks" in the same sentence without mentioning Vasquez's aborted first season in Milwaukee, can you? The Bucks were terrible offensively and defensively with Vasquez on the court, and all of his plus-minus-based metrics agree.
Jerryd Bayless: Bayless was one of just four Bucks to finish with a positive offensive RPM (barely), which speaks to an impressive season in which he shot a career-best 43% from deep and filled in admirably at both guard spots when healthy. But his defensive metrics suggest his renaissance only came on one end. Bayless' raw DRTG and defensive RPM were the worst of any Buck outside of Chris Copeland -- a rather damning stat on a team full of poor defenders.
I'm not sure that speaks to Bayless' faults as a man defender so much as his inability to make any impact as a team defender, though that's also the kind of sneaky badness that is often overlooked. While he can move his feet and stick with guys off the bounce (remember his work in OT vs. Lebron James?), his short arms and microscopic rebound, steal and block rates are nothing new. I'm not opposed to the Bucks re-signing him in spite of all that, but his limitations on the defensive end should give us pause before assuming he's a good option as a major-minute guy going forward.
Michael Carter-Williams: MCW is in many ways the opposite of Bayless. While his gambling, inconsistent style is often frustrating in on-the-ball situations, his length and athleticism give him unique abilities as a help defender -- think contested rebounds, highlight reel blocks at the rim, and the team's highest steal rate as well. All of that manifests itself in positive RPM and on-off impacts on the defensive end -- an important factor when considering his overall value moving forward -- though his effect on the offense was overall negative and got dramatically worse as the season wore on.
Jabari Parker: Rookies rarely help teams win basketball games, and the 14/15 version of Jabari Parker was no exception. His RPM numbers were poor on both ends (-1.69 offense, -2.66 defense) and the Bucks were dramatically better when he wasn't on the court. That's a big reason why many of us were queasy with banking on Parker's return actually helping the Bucks win more games. Of course we were excited to get Jabari back and restart his NBA learning curve, but it's still a leap from putting up numbers to getting wins.
The good news is that Jabari's numbers were better across all fronts this season; the bad news is that he's still a ways from being a clear "plus" player. The Bucks were slightly better with Jabari on the court than off, though his improvements in both ORPM (to -1.13) and DRPM (to -2.12) still left him squarely in the negative. The former seems like a good bet to flip to positive at some point in the near future; if the latter does it'd be major news.
John Henson: Perhaps not surprisingly, the Bucks have posted their best defensive numbers with Henson on the court the past two seasons. Alas, they've also been dreadful offensively in those same scenarios, which speaks either to the context of Henson's playing time -- sharing the court with less talented bench units doesn't help -- or the unfortunate inability of Henson to positively impact a team's offense. Henson's ugly ORPM numbers (-4.16 last year, -2.59 this year) suggest it might be more of the latter, which is of course interesting considering Henson's low-usage, high-efficiency style wouldn't seem that contrary to an effective offense.