A few weeks ago, Matt D’Anna published a fine article on the “trade space” of the NBA. He did a fine job of arguing that communities with in the league exist to facilitate trades. His work used both the history of trades across the NBA as well as a secondary network of General Managers to more effectively model the trading communities. To his credit, his model suggested the possibility of the Derek Rose trade. As some of you know, I am very interested in modeling sports data as social networks. It is an annual hobby of mine to use a network to model the win/loss network of college football — this effectively predicts the playoffs. Making additional models of the NBA trade space would be right up my alley, so I got in touch with Marden (who scrapped the original data from the most awesome of site Basketball-Reference) updated it with recent trades, then segmented it into historical periods and used both modularity and centrality analyses to study the relationships across the league.
To begin, this is my rendering of the classic social network hairball for this dataset:
Each of my primary layouts is circular, with the top five nodes by Eigenvector centrality in the center of the circle. Nodes are colored by algorithm detected centrality.
This same network can be filtered by activity within conferences. As you can see, the community that is colored yellow is not represented among the most central teams in the league. And that intra-conference activity is substantial. Interconference activity is also heavy, some teams like Chicago are most active in inter, rather than intra conference activity.
The most central teams make some sense to me off handedly, although there was something strange — why is Atlanta so central? They are an old team, but not that active a trade player in recent years compared with the trade happy 76ers. In fact, in recent years the 76ers and Timberwolves have traded nearly two-thirds of a league worth of players.
In the time period after 2006, MIN and PHI have been bad and they have traded with almost everyone. So, why doesn’t MIN appear in our top five trading centrality teams? They are too young…
I decided to break down the history of the league to understand why teams like ATL have such high degrees of centrality. First, I needed to understand the eras of the NBA. I enjoy this sort of analysis. To get at eras, let’s look at some data.
Two line graphs: Offensive rating and effective field-goal percentage. Each has multiple encodings. On the upper graph, the color green indicates the percentage of three-point shots made, the thickness is the number of threes per game. On the lower graph, color indicates pace, darker red is slower, green is faster. Thickness is offensive rebounding percentage. The x-axis for both graphs is the season. Why so many null entires? Offensive rating is only calculated in the early 1970s. As for the lower chart, the NBA game before the late 1960s was really different. In the 1950s for instance, you were nearly as likely to end a possession with a made basket as much as a free-throw attempt. Although another line graph could be had here, you should know that folks just aren’t fouling like they used to. How does this help us break down the eras? As major players come and go the style of play changes. The Jordan dominance era (91–98) is marked by a dramatic reduction in pace and shooting with a strong emphasis on offensive rebounding. Immediately after the end of Jordan’s career with the Bulls, the NBA entered an era with terrible shooting and an even slower pace of play. I call this caveman ball. Each era will be described here and analyzed for centrality. Color = community of teams trading, placement of node = eigenvector centrality, size of label = beteweenness centrality.
The Prehistory of the NBA
Until the end of Russell’s career with the Celtics, it was simply a different game .This is the pre-history of the NBA. For the first few years of this era there was no shot clock and no one could shoot. Also, there were very few teams. This is the longest era in the NBA as these micro-eras, team menus, and rule sets don’t call for analysis as much of anything. Don’t get me wrong — Russell, Mikan, and Chamberlain were excellent. It would be interesting to see them play in a league that wasn’t such a mess.
This is a league with a handful of teams. The Hawks of Milwaukee, St. Louis, and Atlanta were by far the most active traders of this era. Isolating this era is important as the early network established by Atlanta in this era could have effects on the analysis for ages to come — thus the move for eras. For those of you playing along at home, the underlying data tracks trades based on the recent names of teams. So Warriors activity is tracked as GSW regardless of the team being in Philadelphia, San Francisco, or Oakland. Pistons activity is tracked as DET in both Fort Wayne and Detroit. The tiny nodes were a wave of expansion teams in strange spots like Seattle and Portland. The communities: the old guard, and everyone else.
The Expansion Era
After the Russell era, the NBA rapidly expanded and experienced an incredibly era of parity. Until the 2013–2016 stretch of Miami, San Antonio, Golden State, Cleveland titles, this was the only time when four different teams in a four year period won the title. The league is much larger in this era, which I place as after 69 and until the 79 draft when Bird and Johnson enter the league and shape the league until the Bad Boy Pistons and then Jordan and the first three-peat. This era included the ABA-NBA merger, which gave the league critical mass as a national league. Aside from the Spurs, the ABA teams (IND, DEN, NJN) tended to trade with each other, and core teams in CHI and the Clippers who as they moved about the country were in flux, as were the traveling Kings.
The Lakers, Suns, Bullets, and Cavaliers formed another community. The rest of the league formed a single trading block with high eigenvector centrality for ATL and betweenness for Seattle. This was the Lakers only real period of trading centrality, number two in eigenvector and number one in betweeness.
Bird and Magic changed the game. Magic won the title in 80, 82, 85, 86 and 87 with his teamfriends Kareem Abdul-Jabar and James Worthy. Bird won in 81, 84, and 86 with his buddies Kevin McHale and Robert Parrish. The 83 76ers superteam featuring Moses Malone won the sole non-Lakers/Boston title until 88, when the rough, tough Bad Boy Pistons won back-to-back. The game was fast and robust with improving offensive rebounding.
Cleveland and Seattle. First, Seattle had a long period of mediocre play and was in the process of trading away all the parts of a championship team as well a surge in the late 80s. Second, Cleveland was a very special case. The trading activity that brought them to this graphic was so pernicious that the league named new rules after their owner, Ted Stepien as he seemed to think that he could trade an unlimited number of assets to win now.
For eight years Chicago and Houston ran the NBA. I chose the Jordan era as it was a high point for the NBA, and if you scroll up you can see that the game circa 1996 is more similar to what we know now, but with more offensive rebounding. It would be fair to say that the first half of Jordan’s era was more impressive than the second. Why? Because the basketball was better. The NBA is a copycat league and as the Pistons and Bulls did their thing they slowed the pace of the league, the pace continued to slow and shooting continued to decline to the point that the slow Jordan Bulls would have been quite fast.
When I was younger, a commentator during a Magic game noted that in “the modern league” of 2001, only Doc Rivers (Magic) and Rick Adelman (Kings) even tried to push it in transition. The Caveman era featured ugly basketball, the 2002 Western Conference finals, the slowest pace since the 1950s, and dropping television ratings. This era also featured many trades by Orlando (rebuilding again), Mark Cuban (DAL), Denver and Philly (rebuilding as always), and Phoenix building the SSOL offense that would prove to the league that you don’t need to let the defense set before you shoot. This is perhaps one of the silliest ideas, that it is somehow less awesome for Steve Nash to do basic skills at a full sprint than it is to watch some over-hyped shooting guard run a flex-post to pretend to be a center and execute a fall-away shot. The 1996 Bulls are a historically great team, the 1998 Bulls were a team that won in a year that was already marked by statistical abnormality. The 99 lockout wasn’t good for basketball, but this doesn’t explain the decay of the
The dominant teams of the Caveman era: Spurs, Lakers, Pistons. Many of the remaining teams were designed to fight and defeat the Bulls (the Pacers and the Bucks) or the Lakers (the Nets). Orlando was particularly active in this time period. They emerged from this era with the Dwight Howard NBA finals team. And which team continues to be tiny: the Spurs. For any number of reasons, including Duncan through the relatively modest financial situation of the team, they were developing their own players, not making splashy trades with other squads. In the Jordan era, the Spurs had nine transactions, in the Caveman era 14. The average team: 45. In terms of z-scores, this is only -1.09, while Orlando’s trade-fest was a whopping 2.79.
Is my loving title for the time after the Caveman era of awful ended. This really could start at the end of the 04 season when the Spurs-Pistons matchup showed how ugly a finals could be. You might argue that this was the result of “excellent defense” and that it was a softening of player discipline that made this possible. I disagree. Pace increased, shooting percentages improved, and most importantly offensive rebounding took a dive. Why do I care about offensive rebounding? It isn’t the oreb that matters, but why it isn’t happening: transition defense is incompatible with sending multiple players to keep a possession alive. At the start of the caveman era the offensive rebound rate was over thirty percent, it was falling quickly as the league started to counter the ultra-heavy rebounding crews, you know, the groups that can’t run. Now one can run and offensive rebound, this was the Showtime era. It might be fair to say that the offensive firepower of this era was tied to poor defense. Offensive rebounds are now rare and shooters are excellent. Apparently, practice does improve shooting.
Trading activity in this era was wild. The Spurs with 17 had a z-score of -2.06, while the mega-traders: HOU, MIN, PHI, MEM, POR, NOH, OKC, BOS, and BKN had as many trades as Orlando in the previous era. In fact, the number of trades in this top core was so extreme that it pulled the average to seventy six transactions over ten years. The average in the Caveman era: forty six. Even correcting for the length of era difference, this is a substantial increase. Another important note, teams in this era that trade win. In my flower I featured MIN and PHI, despite HOU trading more than either of them. In the Futurball era the Rockets have made the playoffs seven times, appeared in the Western Conference Finals. Memphis has appeared in the playoffs six times with a Western Conference Finals and two second rounds. The Trailblazers have made the playoffs six times and won a series.
Aside from the Spurs, the teams near the bottom of the trading chart in this era were stinkers, or at least became stinkers in the second half of the era such as the Lakers, Jazz, Pistons, Magic, and Bulls. If trades are suggestive of other roster movement, NBA rosters are in flux like never before which could explain the current era. Could Golden State win another title? Sure, so could ten other teams. Unlike the NBA of the Showtime, Jordan, or Caveman eras, there are no apparent dynasties. This makes the game exciting. I could also be fairly easily convinced that the 2014 Spurs mark a new era of shooting.
Eigenvector vs. Betweenness centrality
As you can see in some of these graphics, there are cases when betweenness and Eigenvector centrality produce different results.
The red community has high betweeness, with lower eigen-centrality. The yellow community has higher eigenvector centrality, but lower betweeness. The green community just trades a ton and is in the center of everything. Eigenvector assumes that a node is as important as the nodes it is connected to, betweenness charts the number of paths through a particular node. Consider the Heat: MIA has a moderate eigenvector (they trade with important nodes), but their betweeness is quite low, they don’t trade with a lot of teams and paths through their squad are rare. Now for the fun — Steven Borgatti has argued that removing key nodes and recalculating is a way of demonstrating impact across a social network. Removing the core of HOU, MIN, and PHI:
The initial impact: IND, LAL, and DAL find themselves in their own little community, now disconnected. Boston now leads the league in eigenvector centrality, but OKC is the node through which the most common path crosses. The Nets, Bucks, Mavericks, Raptors enable many paths. Of course, the Spurs trade with no one.
Centrality and the Trade Space
It seems that the relationship between team success and trading is breaking down. In prior eras, high centrality teams were clearly not elite. These were teams that were either falling apart or just getting put together. Above all, your team won’t be getting a great reserve from the Spurs. Trend lines suggest that you want to be a team that is influential, but not between if you want to be successful. Golden State is the outlier here, every other team has recently won a title is outside the confidence interval.