While the Hoya faithful patiently await schedule news, the NCAA announced adoption of a new metric for ranking teams that will repeal and replace RPI as a key tool for NCAA Tournament Selection. The NCAA Evaluation Tool (NET) has some interesting bases but the algorithms appear to be hidden for now.
Of course, with Georgetown’s historically weak non-conference schedule last season (2017-18) drawing attention due to the opponents’ ultra-high RPIs, the Hoya faithful are compelled to investigate this new selection tool.
Moreover, in the expectation of Georgetown being competitive in the Big East (e.g., 6th place is typically 9-9), this new machine-learning-based number could very well be the determining factor for the Hoyas dancing in March.
Here are some writers and bloggers discussing NET:
The NCAA announced Wednesday that, effective immediately, they would be using the new NCAA Evaluation Tool (NET) to rank teams instead of the RPI... Between this new system and the NCAA’s adoption of quadrants last season, it seems like we might finally be on the right path to getting the NCAA Tournament field seeded more accurately than ever.
The NCAA Evaluation Tool, which will be known as the NET, relies on game results, strength of schedule, game location, scoring margin, net offensive and defensive efficiency, and the quality of wins and losses. To make sense of team performance data, late-season games (including from the NCAA tournament) were used as test sets to develop a ranking model leveraging machine learning techniques. The model, which used team performance data to predict the outcome of games in test sets, was optimized until it was as accurate as possible. The resulting model is the one that will be used as the NET going forward.
The NET was built to create a ranking system that was as accurate as possible while also evaluating team performance fairly. To ensure fairness, certain types of data were omitted from the model. Of key importance, game date and order were omitted to give equal importance to both early and late-season games. In addition, a cap of 10 points was applied to the winning margin to prevent rankings from encouraging unsportsmanlike play, such as needlessly running up the score in a game where the outcome was certain.
“What has been developed is a contemporary method of looking at teams analytically, using results-based and predictive metrics that will assist the Men’s Basketball Committee as it reviews games throughout the season,” said Dan Gavitt, senior vice president of basketball for the NCAA. “While no perfect rankings exist, using the results of past tournaments will help ensure that the rankings are built on an objective source of truth.”
Just spoke w/ Dan Gavitt. Some things to know:— Matt Norlander (@MattNorlander) August 22, 2018
—NET algorithm powered by artificial intelligence, thus isn’t something that’s “readable”
—NCAA owns NET metric
—NET will start w/ weekly unveilings in late-Nov/early Dec
—NCAA won’t apply NET metric to previous yrs for comparison
The NCAA announced that NET will weigh all games evenly regardless of when they take place during the season, and the margin of victory will have a cap of 10 points to discourage running up the score against inferior opponents...
RPI takes only the winning percentage of a team and its opponent, as well as the winning percentage of teams the opponent has faced, into account.
NET figures to provide a more scientific ranking due to the inclusion of margin of victory, game location and both offensive and defensive statistics.
NB: “Scientific” is usually reserved for a test with a control group, but let’s not split hairs when we can’t see the algorithm or tests...
In late July, use of the NET model was approved following several months of consultation between the Division I Men’s Basketball Committee, the National Association of Basketball Coaches, top basketball analytics experts and Google Cloud Professional Services.
The system was optimized through a trial period during the late portion of the 2017-18 regular season and NCAA tournament. Results from several games were used as test sets to create a system that leveraged the machine’s learning techniques.
This metric will be impacted both by the predictive nature of metrics like KenPom as well as purely results-based metrics like the RPI. The difference is subtle but important. Predictive metrics are generally based on things like efficiency and are not as impacted by something like a buzzer-beater going in and changing the outcome. Results-based metrics are, obviously, as they change the result of the game even if it shouldn’t impact how good you think either team is... Why is it important to include both?  Because we want those buzzer-beaters to matter, right?  Winning needs to matter, otherwise there’s no point in playing the game. But losing a nail-biter is not the same as getting whipped by 25. That should matter, too. I’m glad both will be factored in.
NB: Well... getting whipped by 25 only counts as 10 with NET, right?
Ultimately, it would be nice if “the NET” wasn’t vague and unreproducible... But regardless of those concerns, my biggest concern is that this is shuffling the deck chairs and ultimately won’t matter much at all if the Selection Committee mentality does not change. That mentality includes overvaluing volume of big win opportunities, so that teams that played lots of tough teams, and won a couple of them, were perceived as better than others who did not get as many chances. As I pointed out last year, the Cinderella runs of the last decade would have largely not occurred if selection continues the way it is going.
Last season’s change to the quadrant system was a big win for Wichita State, which has been college basketball’s best true road team the last five seasons... Wednesday’s changes to feature efficiency measures more prominently makes Wichita State a clear winner. The Shockers have consistently been regarded favorably by advanced metrics, only to be spurned by the NCAA Tournament selection committee and the RPI.
Preliminary thoughts are that (1) the metric’s lack of readability equates to a lack of transparency and (2) that a cap of 10 points seems a bit small in a game where 3.3 baskets could make a tie, but judgment will be reserved until more details emerge.
From the imminent Georgetown schedule release (September?) until late 2018 the RPI calculations and KenPom scores will have to support our schedule analysis, as (if Jeff Goodman is to be trusted!) we’re expecting first numbers in late November or early December:
NCAA’s Dan Gavitt just told @WatchStadium that the first NET will likely be released in late-November or early-December on a weekly, and then daily basis.— Jeff Goodman (@GoodmanHoops) August 22, 2018
Lastly, an old article about betting and predictive metrics, that might draw some ideas with the law changing around the country:
The over/under at one Las Vegas sportsbook opened at 121.5. Bettors recognized the mistake quickly and repeatedly fired $500 limit bets at the errant number. With smirking wiseguys lining up, it took only two or three bets for the book to realize something was askew... In the first hour after being posted on the oddsboard, the total grew nine points to 130.5, with the book bumping up the number several points with each sharp bet. Eventually, the glitch was corrected, and the total would close at 156.