Three-Man-Weave

View Original

Testing The Ewing Theory

-Matt Cox

See this content in the original post

Spot on, Andy. Screw what Michael says.

In college basketball, the ‘addition by subtraction’ phenomenon is alive and kicking…

Many moons ago, defending an instance of such convoluted logic was viewed as preposterous. After all, it defies basic math: Team X — Team X’s best player = Team Y (a lesser version of Team X). Yet somewhere along the way, the basketball world wised up and disproved this theory. I’m not sure when this paradigm shift officially transpired, but based on the widely used and understood application of ‘The Ewing Theory’, I’ll give ‘the Hoya Destroya’ all the credit (sorry, Bill).

How fitting that within the scope of the 2019-20 college basketball season, Patrick Ewing’s been the biggest beneficiary of the same phenomenon named after him. All it took for a struggling Georgetown team to get back on track was removing its sophomore star point guard James Akinjo from the equation and Voilà! The Hoyas immediately rattled off six straight wins, dominating Oklahoma State, SMU and arch rival Syracuse over a 10-day span in mid December.

Listen, I’m by no means a fan of Jim Boeheim, Syracuse’s curmudgeon head honcho, but I stood up and applauded at his refreshingly honest take on Georgetown’s situation.

“They got rid of a guy that wouldn’t pass the ball to anybody and just shot it every time, and that’s why they’re good now,” Boeheim said of Akinjo. “Patrick [Ewing] can’t say that but I can. He lost two games for them by himself.”

Thank you Jimbo for cutting through all the politically correct speculation that indirectly pointed the finger at Akinjo.

Now a month removed from Georgetown’s December tear, a 2-5 Big East record makes it harder to prove James Akinjo was truly a cancer to the Hoyas. I’ve yet to pick a side on this debate, but that’s not why I’m here. This example is one of many fascinating puzzles we here at the Weave attempt to solve on an annual basis in our offseason previews. If there’s one thing we’ve learned over the years, putting too much weight in the “Key Returners” and “Key Losses” variables can come back to haunt you.

Two of the most cerebral minds in college basketball analytics, Ken Pomeroy and Bart Torvik, tussle with the same conundrum in their preseason ranking models. I wanted to test whether or not their projections are susceptible to overvaluing certain star players, a pitfall I find hard to avoid when compiling my own preseason rankings qualitatively. By aggregating Ken and Bart’s archived preseason rankings and comparing them to the current team rankings gives us some reasonable measure for how applicable the Ewing Theory is for a handful of relevant case studies.

The teams selected for this analysis were those who lost a high usage player this offseason - for simplicity, such players are referred to as ‘Ewings’ from here on out. Using basketball-reference.com’s query tool, I pulled a list of players with the highest usage rates during the 2018-19 season.

  • Minimum of 29% Usage: Roughly 100 players met this threshold last season, headlined by Chris Clemons at the top.

To whittle this initial list down to a more appropriate and manageable subset, the following filtering criteria was used:

  • Minimum of 700 Total Minutes Played: This filter removes players who were injured for large chunks of the season or players who had their per game minutes capped around 20 minutes a game (Lafayette Dorsey for Pacific is one example).

  • Additional Hand-picked Exclusions: Other unique scenarios called for additional exclusions (these were at my discretion, so deal with it!). One examples from this batch was Jarrell Brantley from Charleston, who played second fiddle to an even higher usage ‘Ewing’ in Grant Riller.

The two-step filtering process shrunk my list of Ewings from 100 to 39 players, which are captured in the comprehensive chart below (see left column).

This chart is crowded, so hopefully the zoomed-in versions below are easier to digest. The color coded arrows in the middle three columns also deserve an explanation. Each set of numbers under the ‘Team Projection Trend’ section show an average of the KenPom and BartTorvik national rankings at three different points in time:

  1. 2018-19 season ranking (end of year)

  2. 2019-20 preseason ranking

  3. 2019-20 current ranking (as of 1/22)

The arrows in the ‘2019-20 Start Rank’ column indicate which teams were projected to be worse (red), better (green) or roughly the same (yellow) as where they finished last season. The arrows in the ‘2019-20 Current Rank’ column indicate which teams have outperformed (green), underperformed (red) or ‘met expectations’ (yellow) relative to the KenPom and BartTorvik preseason projections. Finally, the ATS record gives a sense of how well each team has performed relative to oddsmakers’ expectations.


Key Findings & Observations

As we’d expect, most teams ridding themselves of a Ewing were discounted in the KenPom and BartTorvik 2019-20 preseason projections. Out of the 39 cases, 28 teams (72%) were projected to be worse than their 2018-19 year end ranking, 6 teams (15%) were projected to be better and 5 teams (12%) were projected to be roughly the same.

Let’s divvy this up into two subgroups: 1) The 28 teams who were predicted to be worse and 2) the other 11 teams pegged to improve (6) or hold steady (5). The second subgroup, shown in the chart below, is an intriguing set of squads.

First of all, I couldn’t help but chuckle when TJ Starks popped up on this list. I’m certain Ken and Bart hit the manual override button in their respective models by incorporating a variable dedicated to Starks, forcing the algorithm to spit out an improved projection for Texas A&M.

Tangents aside, it’s fascinating that the six teams that lost Ewings with O-Ratings of 105 or higher - Purdue, Charlotte, Kent State, Cal State Fullerton, VMI and Washington State - did not fall off a cliff in the KenPom and BartTorvik 2019-20 preseason projections. It makes sense that removing an inefficient gunner from the offensive concoction and divvying his shots to more efficient shooters would warrant an upgrade. But removing a high usage and high efficient alpha?

Anecdotally, that should signal ‘regression’. Yet the analytic projections were clearly influenced by other stronger variables (and rightfully so, for the most part). While Purdue’s obviously struggled a bit this season, Matt Painter still boasts a fringe top-30 team (according to the metrics), a testament to his system’s stability. The same could be said of Rob Senderoff at Kent State and Dedrique Taylor at Cal St. Fullerton, both of whom have gradually risen the program standards at their respective schools over the years. Fullerton has hit some bumps in the road this year, but you can’t blame Ken and Bart for not factoring in the absence of Jackson Rowe for the first 11 games of the season. Dan Earl at VMI also employs a unique system, but the Keydets’ favorable projection was likely driven by the return of five key rotational players from a year ago (VMI ranks 83rd in KenPom’s ‘Minutes Continuity’ metric). Finally, Charlotte and Washington State, despite losing two tremendous talents in Jon Davis and Robert Franks, were beneficiaries of a coaching Ewing Theory - that is, two gargantuan coaching upgrades (sincerest apologies to Ernie Kent and Mark Price).

Taking a step back at that second subgroup of 11 teams, six currently reside within 30 spots of KenPom and BartTorvik’s aggregated preseason rankings - pretty damn impressive, fellas.


Now, let’s see how Ken and Bart fared on that first subgroup, teams projected to be worse than their 2018-19 year end ranking…

The boys did pretty damn good with this group as well. They essentially hit the nail on the head for nine of these teams’ projections (identified by the yellow arrow under the ‘2019-20 Current Rank’ column) and only four teams fell significantly lower than their initial estimates (identified by the red arrow under the ‘2019-20 Current Rank’ column). But, their models did fall prey to the Ewing Theory on the remaining 15 teams, who have significantly outperformed initial expectations (identified by the green arrow under the ‘2019-20 Current Rank’ column). For ease of viewing, I blew up those 15 examples below:

Attention all Ethan Happ haters: the chart above is the ammo you’ve been looking for. Wisconsin’s not far off last year’s pace, currently nestled inside KenPom’s Top-25, a slight uptick from their mid-30s projection. I’m still flabbergasted by how Campbell, Northern Colorado, South Dakota State, Hofstra, Arkansas State, Georgia State, Quinnipiac and Montana State haven’t skipped a beat this year, despite losing a do-everything offensive catalyst this summer. All eight of those teams are shattering preseason expectations, both KenPom’s and BartTorvik’s calculations, as well as the oddsmakers’ expectations. That subset is a combined 90-47 against the spread this season, which equates to an outrageous 66% win rate.

There’s no uniform explanation for why everyone under the sun - Ken, Bart, the oddsmakers, and all other preseason prognosticators - underestimated these teams coming into the year. Each case is influenced by its own unique set of circumstances.

The broader takeaway is that while there’s no ‘one size fits all’ explanation that can be applied to every team above, this quick and dirty analysis unveils some evidence that we tend to overvalue individual players in our preseason projections. Remember that one ancient Greek mythology saying? ‘Cut off the head of the snake and one will grow back in its place’. To basketball-ify that mantra, simply control find and replace ‘head’ with ‘Ewing’ and ‘snake’ with ‘team’.