The On-Off debate about Luka

[___Rand___]

https://www.theringer.com/2024/5/8/24151888/luka-doncic-stats-dallas-mavericks-oklahoma-city-thunder

Across Doncic’s first five seasons, the Mavs were only marginally better with him on the court than on the bench, according to advanced stats.

The math, as they say, wasn’t mathing.

Luka Doncic’s Year-by-Year On/Off Stats
Year On/Off
2018-19 -3.7
2019-20 1.2
2020-21 3
2021-22 0.2
2022-23 4.9
2023-24 9.4
The raw stats have always been robust. Doncic averaged a league-leading 33.9 points this season, with 9.2 rebounds and 9.8 assists. For his career, he’s averaged 28.7 points, 8.7 rebounds, and 8.3 assists. And he’s always been an efficient scorer, not just a volume shooter. And yet his “on/off” stats—the difference in the Mavs’ success with Doncic on the court vs. off the court, a generally reliable indicator of a player’s true impact—did not reflect his elite play.

This all confounded and fascinated Ilardi—a clinical psychologist, neuroscientist, and basketball enthusiast with an analytics background, who has worked with the Phoenix Suns and the Houston Rockets and helped develop ESPN’s “real plus-minus” model with Jeremias Engelmann, back in 2014. Ilardi was also one of the pioneers in developing adjusted plus-minus models. (Raw plus-minus measures a player’s impact by tracking his team’s net scoring advantage—points scored minus points allowed—while he’s on the court. Advanced models like RPM adjust for the impact of the other nine players on the court.)

Ilardi had been periodically checking Doncic’s “on/off” stats. And the numbers were, well, weirdly lackluster.

“Over 6 seasons the Mavs have been nearly as good when he’s on the bench as they are when he’s on the court,” Ilardi tweeted on January 26. And this season, Ilardi noted then, Dallas was actually worse when Doncic played—posting an on/off net rating of minus-1.2 points per 100 possessions. “No other superstar shows this pattern,” Ilardi tweeted, punctuating his observations with a googly-eyes emoji.

Even as Luka’s on/off rating began to rise, he lagged far behind other MVP candidates, as Ilardi would note two weeks later, with a graphic showing Nikola Jokic (plus-18.1), Kawhi Leonard (plus-13.4), Shai Gilgeous-Alexander (plus-12.8), Giannis Antetokounmpo (plus-12.6), and Doncic (plus-3.4).

And then came the real kicker: a 20-year comparison, compiled by Ilardi, of every player who’s finished in the top five in MVP voting, listed by their career on/off ratings. At the top: Jokic (plus-11.8), Kevin Garnett (plus-11.3), LeBron James (plus-10.8), Joel Embiid (plus-10.5), and Stephen Curry (plus-10.4). And at the bottom of the 35-player list? Derrick Rose (plus-0.1), who was just below Doncic (plus-1.0), who was just below Joakim Noah and Jermaine O’Neal (both plus-1.1) and Carmelo Anthony (plus-1.7).

“Shockingly low,” Ilardi said of Doncic.

On/Off Leaders for Top-5 MVP Finishers From 2003-2023
Rank Player Career Net Rating
1 Nikola Jokic 11.8
2 Kevin Garnett 11.3
3 LeBron James 10.8
4 Joel Embiid 10.5
5 Steph Curry 10.4
6 Chris Paul 9.3
7 Dirk Nowitzki 9
8 Shaquille O'Neal 8.5
9 Tim Duncan 8
Damian Lillard 8
11 Paul George 7.6
12 Steve Nash 7
13 Kawhi Leonard 6.9
14 Jayson Tatum 6.8
15 Giannis Antetokounmpo 6.7
16 Blake Griffin 6.4
17 Kevin Durant 5.8
18 Dwyane Wade 5.1
19 Anthony Davis 5
20 James Harden 4.8
21 Kobe Bryant 4.6
22 Dwight Howard 3.9
23 Shai Gilgeous-Alexander 3.6
Isaiah Thomas 3.6
25 Russell Westbrook 3.5
26 Devin Booker 3.4
27 Allen Iverson 3.2
28 Peja Stojakovic 2.9
29 Luka Doncic (as of April 14) 2.7
30 Chauncey Billups 2.5
31 Tony Parker 2.1
32 Carmelo Anthony 1.7
33 Jermaine O'Neal 1.1
Joakim Noah 1.1
35 Luka Doncic (as of Dec. 23) 1
36 Derrick Rose 0.1
Steve Ilardi
Everyone on that list, of course, is a star by definition—but there are different tiers of star in this league. And Doncic, by any measure—eye test, box score test, Rorschach test—has to be a considerably better, more impactful player than, say, Peja Stojakovic (plus-2.9) and Isaiah Thomas (plus-3.6) … right?

“Yeah, it’s a compelling illusion!” Ilardi said via text, as part of our season-long discussion.

The thing is, Doncic isn’t merely a great player, but possibly one of the greatest. He’s a generational talent, an all-timer—“the best offensive player I have ever seen in the NBA,” Stan Van Gundy, the veteran coach and TNT analyst, told me. So none of this made much sense.

[___Rand___]

Starting in late December, Doncic’s on/off number started creeping upward, a trend that continued all season. From December 23 through April 14, his on/off net rating was a robust plus-14.4—a figure that would have been second only to Nikola Jokic among high-usage stars. For the full season, Doncic finished at plus-9.4, easily the best mark of his career.

[sp6r=underrated]

“Yeah, it’s a compelling illusion!” Ilardi said via text, as part of our season-long discussion.

This is an important point and if it continues for his career I'll become a Luka skeptic. But he's only 24. And some ATGs have had pedestrian on/off numbers early in their career.

In Luka's defense he made a big breakout this year and it may be a case that either it was noise (possible, on/off is noisy) or his team gave him rope early in his career to figure things out on the court recognizing he's a generation talent (very possible).

I consider the two possibilities I listed above much more likely than him being a pseudo-star.

[nikster]

The last 2 years already disprove this isnt an issue, no?

2022-23 4.9
2023-24 9.4

[___Rand___]

“Yeah, it’s a compelling illusion!” Ilardi said via text, as part of our season-long discussion.

This is an important point and if it continues for his career I'll become a Luka skeptic. But he's only 24. And some ATGs have had pedestrian on/off numbers early in their career.

In Luka's defense he made a big breakout this year and it may be a case that either it was noise (possible, on/off is noisy) or his team gave him rope early in his career to figure things out on the court recognizing he's a generation talent (very possible).

I consider the two possibilities I listed above much more likely than him being a pseudo-star.

That's the thing right - his impact on the other end of the court has been negative. That's why his numbers are what they have been. As the article notes, Luka's numbers and his defense began to improve in December. He finally decides to play defense, and coincidentally, also got slimmer!!!

So now the question is if he sustains this next season or will this be a half season wonder. Let's hope it is part of his maturing process, and this is an indication of his growing maturity to be a more intelligent impactful player.

[lessthanjake]

Starting in late December, Doncic’s on/off number started creeping upward, a trend that continued all season. From December 23 through April 14, his on/off net rating was a robust plus-14.4—a figure that would have been second only to Nikola Jokic among high-usage stars. For the full season, Doncic finished at plus-9.4, easily the best mark of his career.

That is really heavily affected by the Mavericks losing games at the very end of the season by 49 points and 18 points, when Luka wasn’t playing because the first round matchup was already set and the Mavericks didn’t really care about the games. For reference, if we look at that December 23 to April 14 timeframe, if you don’t count those games (which are just garbage games that the Mavs weren’t really trying in), his on-off per 48 minutes was more like +11 (and considering the Mavs had a 100.1 pace this season, the on-off per 100 possessions was likely essentially the same). Still really good, but notably less good than +14.4.

[Mavrelous]

Multiple reason:
1. Most of his career, 2nd best player on his team is his backup.
2. Had to drag KP with him in the rotation for a year and a half, KP was -5 on/off after coming back from meniscus.
3. Never had a good healthy roster, and his starters are very similar to his bench players, only ones are playing opposing starters and the others are playing opposing benches.

Luka has 35 PO games and 1340 minutes, that's half a season worth, with normal rotations, his on/off is 15.1, that's very respectable.

[TheSuzerain]

The last 2 years already disprove this isnt an issue, no?

2022-23 4.9
2023-24 9.4

Single season on/off is crazy noisy.

So, no, it hasn't been disproven.

[Lockdown504090]

Luka and 5 junkies will be decent so the mavs can and have used him in lineups like that so they can be meh without him. Brunson played a lot without Luka so they were still okay even though they aren’t a great fit stylistically. Kai is a better and smarter overall player than Brunson and always having one of lively and gafford on gives them better lineups when Luka is on the floor. Also, Luka has gotten better. He shoot better, and gets a lot of rebounds compared to a few years ago.

[JonFromVA]

The last 2 years already disprove this isnt an issue, no?

2022-23 4.9
2023-24 9.4

Single season on/off is crazy noisy.

So, no, it hasn't been disproven.

There's no "noise" in raw +/- or on-off, but if you're trying to explore cause & effect rather than guess or blame: lineup data is a lot more useful.

[TheSuzerain]

The last 2 years already disprove this isnt an issue, no?

Single season on/off is crazy noisy.

So, no, it hasn't been disproven.

There's no "noise" in raw +/- or on-off, but if you're trying to explore cause & effect rather than guess or blame: lineup data is a lot more useful.

there is massive noise in both.

[CoP]

The last 2 years already disprove this isnt an issue, no?

2022-23 4.9
2023-24 9.4

Single season on/off is crazy noisy.

So, no, it hasn't been disproven.

While I agree with you, this is two years now. It's still not as high as some other top players, but you can't deny that it has improved.

[bledredwine]

Just from seeing D Rose listed at .1, I can tell you that sometimes on/off is purely contextual and complete BS.
During his years, our bench crew crushed opposing team's bench crews and he had little support in comparison.

There must be something contextual to Luka.

And I'd love to see the numbers in the playoffs only, for Rose and Luka.

[Texas Chuck]

yeah so we are watching Jalen Brunson become the darling of the league. Him being on the court for the non-Luka minutes probably have nothing to do with this....

This is why +/- is flawed as a measure of player quality. KG looks great because the Wolves were terrible at team construction so the team fell apart when he sat. Doesn't mean he was a better player than Tim Duncan on Dirk who played on better constructed teams.

It's lazy and incomplete.

[zimpy27]

Yeah Brunson coming off the bench really messed with the +/- numbers..

Kind of odd that Ilardi didn't even mention this

[TheSuzerain]

The biggest improvement to Luka's on/off is that now the non-Luka minutes are also awful defensively in addition to the Luka minutes.

It's pretty clear that Luka is a prodigiously great offensive hub. But the signals are there that he's awful defensively, and it eats away at his impact.

[JonFromVA]

Single season on/off is crazy noisy.

So, no, it hasn't been disproven.

There's no "noise" in raw +/- or on-off, but if you're trying to explore cause & effect rather than guess or blame: lineup data is a lot more useful.

there is massive noise in both.

Literally 0.

Like any other raw statistic it can be inaccurate if someone simply entered the data wrong, but I wouldn't call that noise.

+/- has more year to year variance than box score stats like pts, rebounds, assists, etc, but that's to be expected from a statistic that has a much larger team component.

[TheSuzerain]

There's no "noise" in raw +/- or on-off, but if you're trying to explore cause & effect rather than guess or blame: lineup data is a lot more useful.

there is massive noise in both.

Literally 0.

Like any other raw statistic it can be inaccurate if someone simply entered the data wrong, but I wouldn't call that noise.

+/- has more year to year variance than box score stats like pts, rebounds, assists, etc, but that's to be expected from a statistic that has a much larger team component.

Ok, you don't know what you're talking about.

[nikster]

The last 2 years already disprove this isnt an issue, no?

2022-23 4.9
2023-24 9.4

Single season on/off is crazy noisy.

So, no, it hasn't been disproven.

It's 2 seasons. And of the last 4 years only 1 was close to neutral. You think it's less noisy if you look at his rookie and sophmore seasons when he was clearly a different player?

[ChipotleWest]

The last 2 years already disprove this isnt an issue, no?

2022-23 4.9
2023-24 9.4

And Brunson was there before that running the second unit, who turned out to be not a bad starting pg. With this stat you get punished if you sit and your backup plays well though.