NCAA tournament 2019: What did the bracket teach us about the NET?

The Selection Committee seemed to use the new metric much like the RPI in previous seasons. The new number also appears to have a stronger grounding in college basketball-specific advanced analytics.

After writing my quick post-Selection Show reaction piece and eating dinner on Sunday evening, I spent most of the rest of the night playing with numbers. I pulled up the official NCAA seed list, my own final projected seed list, the final NET numbers and nitty gritty report, WarrenNolan.com’s version of the RPI and the selection-time KenPom rankings. Then, I opened a spreadsheet and got to work.

Predicting the bracket with the NET

For starters, if you were to project a bracket solely based on the NET, you’d do a slightly better job than if you used the RPI, but you would still do far worse than if you attempted to replicate the Selection Committee’s process.

To test this out, I examined how the top 50 teams on the seed list — everyone on lines one through 12, which is the typical end of the window for at-large contention — and compared the true seed list with my projections, the NET and RPI. (Plus, the gap between seed numbers and NET rankings start to become so large by the time you get to line 14 that the analysis would be nearly meaningless.)

• Based on the RPI alone, 17 of these 50 teams were projected on to the correct seed line, with a further 11 being off by a single seed line, traditionally viewed as acceptable in bracketology because of the Committee’s historic tendency to move teams from line to line to correct bracketing issues or balance the field. That means 22 teams were off by more than a seed line, with the gap growing to eight lines in the cases of Liberty Flames, Ohio State Buckeyes and Ole Miss Rebels.
• The NET did significantly better, but its performance still wasn’t great. The seeds of 22 of the 50 teams were spot on, with a further 12 within a line. Of the 16 that were off, a six-line difference (St. John’s) was the maximum, with only a quartet incorrect by four or more lines. Seven such teams were off by four-plus lines in the RPI projection.
• Of the 49 teams I correctly placed on the top 12 lines (sorry, Temple Owls, again), 37 would have ended up on the same seed line as on the real seed list, and of the 12 that were incorrect, seven were within a line. And I had no misses greater than two lines.

So how is the Committee using the NET?

You might hate hearing this, but they seem to be using it in the same way as they used the RPI. Selection Committee chair Bernard Muir made this pretty clear in every single interview he gave last night. It’s being used as a sorting device, not a “throw it into a computer so it will spit out a perfect bracket” tool. Many of the other factors the Committee uses to select and seed teams are still very much in play, so the quality of wins and losses, winning percentage in Quadrant 1 and 2 games and scheduling will all continue to play a role in the NCAA tournament selection process.

Is the NET better than the RPI?

In my opinion, yes. Because it certainly seems to be more grounded in college basketball analytics, as designed. Switching our attention back to the larger NCAA picture, I compared the Selection Sunday NET of the 68 teams in the field and the top eight NIT seeds (the top four of whom are explicitly labeled as the first teams left out) with the day’s RPI and KenPom rankings.

The average difference between a team’s NET and RPI rankings for these 76 teams was a little more than 16 places, while the average gap between the NET and KenPom plunged to just 5.5 spots.

The NET manages to treat teams relatively equally when compared to KenPom, with a nearly even split between teams with higher rankings in one or the other metric. Plus, there aren’t quite as many extreme outliers as in the RPI. Only the UNC Greensboro Spartans, the first team out of this year’s field, and North Dakota State Bison have gaps of more than 20 spots between their NET and KenPom ratings, with the Spartans 21 spots lower in KenPom and the Bison 23 places higher.

But there’s much more variability when comparing the NET to the RPI. Just 28 teams from our 76-team sample boast RPIs that rank within five places of their NET, compared to 46 such teams in KenPom. Plus, there are 22 teams with NETs 20 or more spots either better or worse than their RPIs. And the North Carolina State Wolfpack, 33rd in the NET but 97th in the RPI aren’t even the biggest outlier. The SWAC champion Prairie View A&M Panthers, a First Four No. 16 seed, have an RPI (140) that’s 65 places higher than their NET of 205, while the Georgia State Panthers NET of 121 is 74 spots lower than their RPI of 47.

Did the Committee use the NET to build the bracket fairly?

I also took a little time to analyze each region and specific pod to see if anything interesting popped up, which naturally happened. Some of this information I’m saving for my forthcoming picks post, since it would apply more there, but I will share some data on how the regions break down now.

Based on the average NET ranking of the teams in each region (averaging the rankings of the teams playing the First Four games), there’s a sizable difference between the strongest and weakest regions.

• The Virginia Cavaliers’ South region is the strongest bracket with its participants’ averaging a NET ranking of 48.88.
• The Gonzaga Bulldogs’ West (52.56) and North Carolina Tar Heels’ Midwest (54.38) are fairly even, but still stronger than the weakest region.
• That’s the Duke Blue Devils’ East bracket, which has a NET average of 60.34. However, some of this gap is due to the presence of the NC Central Eagles, ranked 302nd, 80 spots lower than the second-worst team in the field by the rankings, and 222nd-ranked North Dakota State.

I decided to dig a little deeper, so I totaled the NET rankings of each pod within the four regions and the results were a little surprising. The lower the total, the stronger the pod.

You would expect that the quartet of pods anchored by No. 4 seeds would be at the top of this table in some order, and the No. 1 seeds’ pods would all rank near the bottom. But this is clearly not the case! While three of the four No. 1 seeds bring up the rear, Virginia, again, is the outlier, with the East region’s second seed, the Michigan State Spartans, finding themselves in a weaker pod, at least according to the NET, than the Cavaliers.

At the top of the table, the Virginia Tech Hokies’ East pod is significantly weaker than those of the other three No. 4 seeds. In fact, the LSU Tigers’ assignment as a No. 3 seed in the same region is slightly more difficult by the numbers. There’s even a fair mix in the middle eight spots on the table, which should theoretically be filled by the two and three seeds exclusively. Instead, there are No. 1 and 4 seeds (Virginia and Virginia Tech) mixed in with trios of teams from seed lines two and three.

That’s the kind of information that will lead me to doing some research in the offseason. Before that, however, we have 67 NCAA tournament games to watch and pick. I’ll post my selections, with more NET information relating to them, on Tuesday.

from SBNation.com - All Posts https://ift.tt/2TI5ofX