# Coaching Trees and Coaching Graphs

The 2018-19 NBA season has started, Spoelstra apprentice David Fizdale has taken charge of the Knicks, Pop acolyte Mike Budenholzer is in Milwaukee, and journeryman assistant Igor Kokostov is leading the maybe-not-terrible Suns. In other words, it’s an ideal moment to gaze at coaching trees.

As quickly becomes apparent from the data, however, full NBA coaching trees are better thought of as gnarled bushes, with head coaches and assistants bouncing around the league, forming multiple, inter-woven connections. In this post, I present some of these networked connections and try to draw out interesting aspects. To do so, I rely on data scraped from Basketball Reference, giving all head coaches in NBA history, along with the team-seasons they coached as both head and assistant.

I used the scraped coaching data to put together a directed network dataset, consisting of a list of nodes (all the coaches in the dataset) and directed edges (connections between two coaches, pointing from head to assistant, formed when one coach served on the staff of another). With respect to these edges, I only used full season tenures – that is, I only included a coach-assistant pairing if there weren’t any mid-year coaching changes on the team they were coaching. I processed and analyzed all network data with R’s network and sna packages, and produced all graphs using a combination of ggplot2 and ggnetwork.

Without further ado, Figure 1 shows a bird’s eye view of the full NBA coaching graph, with edges representing all full-seasons head-assistant links (as occur in the Basketball Reference profiles). As promised, the central component of this graph is large and tangled – of the 359 coaches on record, 229 are all mutually connected to one another by head-assistant ties, another 10 are connected in isolated pairs, and 120 have no coaching staff connections with other head coaches (these latter isolates, which include a bunch of fairly random dudes but also the likes of Red Auerback – big red dot on left – could potentially be an artifact of the dataset, which includes sparse assistant coaching records from the ’70s and earlier).

To go beyond this bird’s nest and get a better look at meaningful coaching “trees”, I also restricted connections to more durable, chronologically ordered head coach/assistant coach pairings – a connection only counts if the assistant hadn’t yet had his first head coaching gig, and it only counts if the assistant worked under the head coach for multiple full seasons. The following graphs show such restricted networks. Figures 2 and 3 present all these connections, with isolates removed, and added labels to prominent coaches – in Figure 2 these labels represent the coaches with the greatest number of mutliyear/sequenced ties to other coaches, while in Figure 3 they represent the coaches with the highest scaled betweenness values, i.e. coaches that are most pivotal in tying different coaches together.

A few notes here:

• Pat Riley and Larry Brown have had a lot of guys on staff (the number of head coaches who have served under Brown balloons up to 18 under the full graph shown in Figure 1).

• Among active coaches, Pop is extremely linked in as one might expect, but so are Rick Carlisle and Mike Brown. Carlisle, specifically, has coached and won a ton of games as NBA coach, but he had extended stints under four different head coaches before taking the reins of a team.

• Probably not coincidentally, dots that are redder tend to be bigger and have more lines coming out of them. That is to say: coaches who are successful at winning games tend to stick around longer, and more of their assistants tend to spin off into head coaching gigs.

Finally, the following charts all focus in on the sequential, multi-year coaching links surrounding some prominent coaches. A few more notes here:

• Pop’s tree (Figure 4) is impressive, with Bud, Brett Brown, Mike Brown, and Jacque Vaughn. Perhaps even more impressive is who he spent time under: the great and prolific Larry Brown (Figure 5) and Don Nelson (Figure 6).

• The Popovich/Brown/Nellie clique (non-technical) can be thought of in comparison to the Pat Riley (8)/Van Gundy bros/Spoelstra/Fizdale/Thibs clique, extremely prolific and accomplished in its own right.

• Riley stands in distinction from another great Lakers coach, Phil Jackson (9), whose coaching tree is minimal (the late Tex Winter, not shown here, also has some head coaching experience, but he got this experience before becoming Phil’s longtime assistant).

• Ricks Carlisle (10) and Adelman (11) also have interesting graphs: the former for its fecundity and range, and the latter for its great parentage (Nellie and Dr. Jack Ramsay) and sparse offspring (a lone, very green-shaded Byron Scott).

Concluding note: there’s much else to potentially explore with this coaching network data. I hope to facilitate that exploration soon with some interactive graphics, as well as a bit of further network analysis. In the meantime, please let me know if you’d like the network dataset I’ve created from Basketball Reference records, or the code I used to scrape, assemble, and graph the data; I’m happy to share.