Apparently margin of victory is capped at 10 points, which is ridiculous. And efficiency isn’t adjusted for the quality of the opponent.
OTOH, Kentucky is ranked 61, so I can get behind that. The SeaofBlue folks are not happy.
Apparently margin of victory is capped at 10 points, which is ridiculous. And efficiency isn’t adjusted for the quality of the opponent.
OTOH, Kentucky is ranked 61, so I can get behind that. The SeaofBlue folks are not happy.
"Better you than us." -- NCAA Tournament Selection Committee
A lot of consternation on this board for a metric that does something awesome: leave UNC out of the top 20. Once they learn how to count, and run out of fingers and toes, the folks at Inside Carolina are going to be really upset.
As I understand it, all metrics are meaningless this early in the season, which makes me wonder why the NCAA bothered to release numbers right now when they were under no obligation to do so. They should have waited until the new year, when most of the college football bowls are over, and most college basketball teams start conference play. That said, I appreciate all transparency, especially the kind that shows this level of organizational stupidity.
Enjoy the process, everyone. I can't believe no one in his thread has mentioned the one team in America worth following: Houston Baptist. They beat Fordham and Wake Forest, and lost to Virginia and Wisconsin. This puts them at #147, which is in the top half of today's list of all 353 teams, just ahead of Final Four team Loyola Chicago (#148) and the following power conference teams: Washington State (#155), South Carolina (#162), Utah (#169), Boston College (#170), Baylor (#185), Wake Forest (#205), and California (#246).
Brevity is, as usual, right on the mark with the above line. Folks are getting worked up about the NET but it is too early to know if it works or is an abject failure. The NCAA smartly is not including any preseason bias/ranking in the NET. Once we hit 10-15 games played by all the teams, then I bet the NET will start to look a little more ordinary. But, at this point, it is actually kinda difficult to tell the difference between Kentucky, Liberty, Radford, Kent St, and Stoney Brook... I mean, they are all 5-1.
That said, as I noted months ago when this sucker was first announced, capping the margin of victory at 10 points is a silly idea that will eliminate a lot of your analytical ability. I suspect that even once we have more data, there will be some head scratching results in the NET. I am hoping the NCAA Selection Committee will be able to see beyond some of the flaws, as they did when they were using RPI as a key measurement.
-Jason "this is better than RPI... but it still isn't close to many other measuring sticks commonly used" Evans
Why are you wasting time here when you could be wasting it by listening to the latest episode of the DBR Podcast?
I agree with the above analysis in theory, but there are a few outliers that I'm not sure will be cured by time. I"m not really sure time is going to correctly sort Kentucky versus Radford, Liberty, etc, under this system...As others have mentioned, the cap at ten for margin is indeed silly...a case can be made for 20 or so, but a ten point game can be a close game with late free throws...which is not what Duke UK was.
And while I have my differences with Nate Silver, his statistical work is very precise and elegant...and he's trashed this system since he found out how they're doing it. Time will tell...
In this poll, Kansas gets leapfrogged by EVERYBODY.
Getting jumped over is getting to be habitual in Lawrence.
Q "Why do you like Duke, you didn't even go there." A "Because my art school didn't have a basketball team."
This is the part that baffles me. They did not need to do this now. You would think that the NCAA would have looked at the results and said, "Nope. Not this week."
I assume they tested NET by running data from prior seasons, and I assume it reflected something close to reality in those tests. These results clearly do not, so why not wait until NET normalizes before releasing it, especially in its first year? I know some already were asking for results before this week, but I think most would understand the simple explanation that the data are not sufficient this early in the season for NET to produce meaningful results.
Last edited by JetpackJesus; 11-27-2018 at 06:51 PM.
I was asked via DM to expand what I meant by my comments above.
Here goes:
I’m a big believer in data and fact, and have a lot of respect for Nate Silver.
My main point was really more about flaws in the formula, and that a sharp statistician like Mr. Silver could have used NET as a foil to discuss the differences between data (raw analytics) and information (transforming data into a story).
It was a bit of a stretch to compare machine learning and artificial intelligence to the blunt tool that is NET (although I would contend NET may make a good case study in inherent bias).
NET, at best, is just data (or as others have pointed out, it may just be too early for the output of the formula to have any tangible meaning). I’m not sure it will ever produce information.
A brief aside on data and information.
An example of data might be, an ostrich lays the biggest bird egg.
If you’ve never seen an ostrich egg, that data isn’t super meaningful.
If I know you’ve seen a chicken egg, or assume it is a common enough reference point and add an additional data point, an ostrich egg is equivalent to 24 chicken eggs, now you have information.
(I resisted temptation to create an example using basketball players and everyone’s favorite unit of measurement.)
Disclaimer: I’m not a statistician or mathematician, and one of my favorite mantras is “often wrong, never in doubt”.
With a little luck, maybe someone found this useful or mildly entertaining.
Let’s Go Duke!
The NET rankings are starting to look more reasonable/representative (and playing Texas Tech and St John' looks like it will be much better for our SOS than expected pre-season):
1 Michigan
2 Virginia
3 Duke
4 Texas Tech
5 Tennessee
6 Kansas
7 Gonzaga
8 Nevada
9 Michigan St.
10 Auburn
11 Wisconsin
12 Buffalo
13 Ohio St.
14 Houston
15 North Carolina
16 Nebraska
17 Oklahoma
18 NC State-seems a bit high
19 Louisville-seems a bit high
20 Indiana Big Ten
21 San Francisco-high
22 Marquette
23 Villanova
24 Virginia Tech-a little low
25 Cincinnati
26 Arizona St.
27 Mississippi St.-a little low
28 Furman
29 Florida St.-a little low
30 St. John's (NY)
31 Purdue Big Ten
32 Syracuse
33 Kentucky
34 UCLA
35 Iowa St.
36 Utah St.-high
37 Arizona-high
38 Liberty-high
39 LSU
40 Colorado
41 Butler
42 Maryland-probably a little low
43 Washington
44 Iowa
45 Lipscomb
46 Kansas St.
47 Northwestern-high
48 Florida
49 San Diego-high
50 North Texas-high
51-60 include TCU, Minnesota and Creighton, all of whom it's hard to see as behind Northwestern, San Diego and North Texas. But, that's really nit-picking at this point
Pfftt...State will prove doubters wrong. The NET is just very forward thinking.
it makes sense. I still wouldn't be concerned about outliers until the end of january or so...and was kind of dissapointed that Nate Silver railed on it when to a large degree, the naivete of the rankings were due to a lack of preseason rankings...something that makes rankings look like crap early, but is likely a good thing down the road.
Now, I'd argue that the NCAA shouldn't have bothered to release them...but such is life.
April 1
I believe Mr Silver would agree with me here in that humans still play an important (the most important?) role in most ML/AI models, namely in designing/creating the features of the model. AI/ML is not just chucking a ton of data into some black box that figures it all out. Thus the importance of human decisions to, as Jason Evans relayed, cap the win margin at 10. That's a major decision made by a human. I'm not sure what goes into NET vs Nate Silver's models versus Ken Pomeroy's models, but I bet Nate's and Ken's are better...
I don't think you'll find disagreement among dorks that the latter 2's models will be better...but they're trying to accomplish different things. Two of them are predictors, one of them is attempting to seed teams both on absolute strength as well as reward for doing certain things that the NCAA likes to see...like winning games. So far as the results aren't as far out of whack as some of the RPI ones were, I'm fine with bumping a team up a seed line or two if they won a bunch of big games, even if they were close, over a team that lost the close games and beat everyone else a bit more.
So saying "we won't reward you for winning by more than 10" does two things:
1) it never encourages you to pour it on (though 10 is likely low)
2) you get more value by playing teams that are closer to you in ranking
Say I'm being compared to some other team. We both play Cupcake state. Team light blue wins by 15, and we win by 20. The system considers that the same even though there is some amount of predictive value in comparing those results. Now, instead we both play a slightly heavier weight team...like pound cake polytechnic...they win by 7 and we win by 10. Now we look better than they do...and that distinction would have been lost if we had played cupcake.
So even if it discards some predictive value, it encourages teams to play teams more equal in strength. I have no problem with that. I also have no problem rewarding more than 2 "points" worth of distinction between a 1 point win and a 1 point loss. You want to reward teams for actually winning the game, even if it has less predictive value.
I'm fine with these things. You should be rewarded for them. This system is getting far more flac than it should.
April 1
I agree, let's let the NET get same data before we just kill it. It's already not looking nearly as terrible as some have made it out to be an we're still in December.
That's fair, though I'm skeptical it will make substantial differences though. How many games will a team play between 10 and 15 points? I wager it's <5 on average. You'll definitely get a change in ranking of 1 or 2, but given the committee has the final say anyway, and it's still within the margin of error for regional balance and other bracketing rules, I'm not sure it will make that much of a difference in the end. At 10 or 15 points, the system SHOULD eliminate the huge outliers that plagued RPI.
April 1
I'd agree if PR for the NCAA were high on my list of concerns, but I like watching rankings "take shape" as more data is accrued and wouldn't have minded if the NCAA had released NET in early November after *1* game had been played by each team. Even releasing it in late November as they did probably has been instructive about the power of sample size.
Anyway, I hope the NCAA releases the formula behind NET at some point but am not very hopeful on that count.