The Ewing theory was actually a real-world occurrence of Braess' Paradox.
Braess' Paradox is a phenomena in networks where adding additional pathways does not always result in increased efficiency. It is most famously observed in traffic, where the addition of an alternate route between points A and B can actually end up slowing things down. It has applications in all kinds of places, from electrical and data circuits to hydraulics, plumbing to various human interaction systems.
Some researchers a while back heard about Simmons & Lowe's theory regarding Ewing and they did an analysis, looking at the scoring efficiency of Ewing and his teammates and showed that what was going on was an instance of Braess' paradox. The short version is that while Ewing represented an efficient 'alternate route' to scoring, the distribution of shots changing when he was on/off the floor decreased the overall throughput.
In the case of the Boston Celtics, though, if Braess' Paradox is in play, it's a more nuanced thing than with what happened in the Ewing case. With the Celtics, the problem can be broken down into three states that cover everything:
A) Possessions with Kyrie on the floor.
B) Possessions without Kyrie, in games Kyrie did not play.
C) Possessions without Kyrie, in games Kyrie started.
The team's net rating in (A) and (B) were both very, very good.
The team's net rating in (C), however, sucked like smelly things in your septic tank. And that's the part that killed us.
That's the mystery of this entire last season and indeed, for the entire two seasons Kyrie was here. For whatever reason, this team just could not figure out how to play well in games Kyrie started, once he would go to the bench. The same non-Kyrie players would play great in games he did not start.
If that can be explained by Braess' Paradox, I'd be very interested.
People will probably shrug when I say this, but I actually believe this may have something to do with the significant disparity in Terry Rozier's production as a starter versus a bench player.
If you look over just his last two seasons, 17-18 and 18-19, Rozier's Per-36 production over that two year stretch was as follows:
As a Starter: 16.3 Pts, 6.6 Reb, 5.7 Ast, 1.6 Stl, 1.5 TO, 42% FG, 40% 3PT, 80% FT
As a Reserve: 14.6 Pts, 6.3 Reb, 3.9 Ast, 0.8 Stl, 1.0 TO, 39% FG, 36% 3PT, 77% FT
His net ratings for last season:
2018-19 (starter): Off Rating 118, Def Rating 107, Net Rating +11
2018-19 (reserve): Off Rating 100, Def Rating 109, Net Rating -9
The only stats that didn't get significant improvements when he started (versus off the bench) were rebounding (about the same) and turnovers (which increased. Other then that he put up better stats across the board in the games he started.
And there are roughly 60 total games started by Rozier over those two seasons, so the sample size is pretty decent.
The two stats that really jump out at me are the Offensive Rating (which rises from 100 to 118) and assists (from 3.9 to 5.7). This gives me the impression that Rozier for some reason did a much better job of running the offense when he in the starting 5.
Possible that when he played with that starters because of the talent around him he played more of a playmaking role, while when he came off the bench because there was a comparative lack of talent he went into ISO mode and played much less efficiently. Or maybe it was because the lack of talent on the second unit allowed defenses to focus more of their attention on Rozier, reducing his efficiency.
Either way that Net Rating disparity is huge, and considering that Rozier was typically the guy who would come in when Kyrie sits, it may well go some way to explaining why Scenario C was such a drastic change. The Celtics played great when Kyrie was in with the starters, and they played great when Rozier was in with the starters - but they played terrible when Rozier was on with second unit guys.
