Professional and College Basketball Forums banner

61 - 66 of 66 Posts

·
Registered
Joined
·
3,188 Posts
Discussion Starter #61
Two things seem apparent to me:

1 - I'm surprised that the math wizards who run some of the metric sites haven't reverse engineered the NET to show exactly how it is being calculated. Would think they have enough data points by now to do this.

2 - In theory, if you were using NET, the at-large bids would end at NET 48. Right now that puts 2 at-large bids for the A10, and 1 for the AAC, and 2 for the ACC(!).

There will be a lot of nonsense (different nonsense every year) when they pick someone in the 50's, and leave out someone in the 40's. Hello this year's Syracuse.
1. There's not enough data points because there's no raw data being displayed anywhere in any kind for NET, and that means it's not just one unknown formula, it's actually THREE.


Take RPI as an example first: The RPI was published at the end of every season by the NCAA and it included raw data (aka, not 63rd in RPI, but RPI of .5655). Win percentage is obviously known. And you can figure out SOS by doing some work; and all Division I teams data is available. Opp SOS would be a lot more work but could be tabulated.

- If you had all four numbers, you can figure out the formula: RPI = (Win Pct + SOS + SOS + Opp SOS)/4

- If you don't have Opp SOS, you can figure it out from the other three numbers if you have the formula.

- If you're missing Opp SOS AND the formula, it would take two reports to figure it out, because you'd have two points for every team, and know the results between the two points, so how Win Pct, SOS got from Report A to Report B, so the formula would have to get Opp SOS and RPI from A to B, for every team exactly the same way. Easy to reverse engineer.


BUT WITH NET, we can get a raw number for four things: Net Efficiency, Win Pct, Adjusted Win Pct, Score margin.

That missing component is it's own formula we have ZERO data points for: The Team Value Index. We don't know the rankings within the TVI that awards "points" for beating 1 or losing to 353. We don't even know who is 1 and who is 353, or the algorith used to run it.


And we don't know the formula of how those first four very different numbers, and the TVI are adapted to make a similar looking set of numbers to combine with TVI for a final number.

By that, I mean Dayton:
Win Pct: .9130 (based on 0 to 1)
Adj Win Pct: .8990 (based on 0 to 1)
Net Efficiency: 0.226 (generally based on -0.5 to +0.5, but theoretically could be between -3 and +3).
MOV: 7.6957 (based on -10 to +10)

So you'd either need to convert Net Efficiency and MOV to something based on 0 to 1 like win pct; or I suppose you could simply use each team's ranking in the category.

Sorry to go full nerd on you, but it's a boring day at work.
 

·
Registered
Joined
·
5,620 Posts
1. There's not enough data points because there's no raw data being displayed anywhere in any kind for NET, and that means it's not just one unknown formula, it's actually THREE.
....................................

BUT WITH NET, we can get a raw number for four things: Net Efficiency, Win Pct, Adjusted Win Pct, Score margin.

That missing component is it's own formula we have ZERO data points for: The Team Value Index. We don't know the rankings within the TVI that awards "points" for beating 1 or losing to 353. We don't even know who is 1 and who is 353, or the algorith used to run it.
..................................

And we don't know the formula of how those first four very different numbers, and the TVI are adapted to make a similar looking set of numbers to combine with TVI for a final number.
But you do have "data points" and "raw numbers"...the first 4 items you noted. OK, you don't have the TVI, but it would seem with enough examples from the real world (hell, some of these guys run tens of thousands of simulations to come up with percentage possibilities of all kinds of things) you should (I think) be able to start coming to grips with your 5th item and the formula/algorithm that drives the NET number we see.

But...I do acknowledge your better understanding of it, and admit my confusion.
So I'll just keep having the Spiders improve everyone else's NET !!!!
 

·
Registered
Joined
·
3,188 Posts
Discussion Starter #63
But you do have "data points" and "raw numbers"...the first 4 items you noted. OK, you don't have the TVI, but it would seem with enough examples from the real world (hell, some of these guys run tens of thousands of simulations to come up with percentage possibilities of all kinds of things) you should (I think) be able to start coming to grips with your 5th item and the formula/algorithm that drives the NET number we see.

But...I do acknowledge your better understanding of it, and admit my confusion.
I’m no math teacher, but it’s really simple reason why we can’t crack it (if only I can explain it clearly). When you solve for X, you move known data to the other side of the equation until you’re left with “X = ____”.

We know that WP, AWP, NE, MOV, TVI (with applied formula) = NET
So you’d move the unknowns to the same side: (WP, AWP, NE, MOV Formula) = NET - “TVI with formula applied.”

What you want to achieve is to make a formula that spits out a list of teams in the same order as NET, and we’re on the right track, right? And if we run next weeks’ data and it matches next week’s order of NET results, we solved it, right?

Wrong. Because our formula isn’t SUPPOSED to equal NET. It’s SUPPOSED to equal NET - “TVI with formula applied.”

If our formula’s results match the NET results, TVI must be "Zero," or "the same for everyone." Or have an inverse relationship with NET (TVI goes down, NET goes up).

But that can't be true because of common sense. No system is going to think everyone's tied, and always output zero. And we know if your TVI is better, your NET is better, because the NCAA said so.

You don’t need Google to make you an algorithm if you can get the exact same data from Win Pct, Adjusted Win Pct, Net Efficiency and MOV. It’s way more likely that the effects of what we DO KNOW carry the LEAST weight, and NET is basically “slightly adjusted TVI”
 

·
Registered
Joined
·
5,620 Posts
At the risk of being more stupid, don't we have a 5th "known"...the NET.
So, in looking at a team who moves from 35 to 30 between 2 releases of the NCAA NET, we could:
1 - look at the 4 known factors
2 - look at the movement of the NET
3 - decide what the TVI must be to make the (4 factors plus a TVI) = the new NET.

We might not know what the algorithm is that creates TVI, but calculating a few times using my example above might give us an answer as to whether it is a constant.

Oh well, I'll give up on all this. Thanks for the response.
 

·
Registered
Joined
·
3,188 Posts
Discussion Starter #65
No worries.

The issue is that in all likelihood, the TVI is the most important factor.

What we do know (win pct, adj win pct, NE and MOV) is all just clever ways of saying "they won."

In order to out-score your opponent, you need to get more points per possession than they do, because basically, both teams are going to have the same number of possessions (not always exact, but within 2 possessions per game, because you literally take turns with the ball!). So if you're positive in Net Efficiency in one game, you basically won.

If your MOV is positive, you won. If your win pct goes up, you won. So all we really know about this, is that the teams that win are at the top.

If the NCAA gave us TVI, we could probably simply ignore the rest. The TVI is going to be the most important component.
 

·
Registered
Joined
·
3,188 Posts
Discussion Starter #66
But let's end math class and go back to ripping all the terrible P5 teams who don't deserve bids!
 
61 - 66 of 66 Posts
Top