Predictive rankings don't like us as much as the record would indicate | Syracusefan.com

Predictive rankings don't like us as much as the record would indicate

Millhouse

Living Legend
Joined
Aug 16, 2011
Messages
29,206
Like
34,386
Since all these ratings are so highly correlated with each other, I'll just look at basketball reference

14-6 against the 13th toughest schedule should be good enough. 13-6 if you toss d2

The computers hate that we lost by 22 on average.
We win by 11 on avg (tossing charminaid)

It would be a shame if this year is decided by our effort at the end of bad losses

Tennessee comparable schedule 15-4 record. Their srs ranking is 8th. We are 72nd

Thank you for this great board
 
I dont think Tenn is a good example. They have no bad losses, 3 top 15 Net wins (one in dominating fashion) and crushed every one in Q2/Q3/Q4. I assume that trend carries through to BPI, KenPom ect.
 
I dont think Tenn is a good example. They have no bad losses, 3 top 15 Net wins (one in dominating fashion) and crushed every one in Q2/Q3/Q4. I assume that trend carries through to BPI, KenPom ect.
if they have no bad losses and all these good wins and our schedules and records are so similar, we must be doing something right that they aren't.
 
if they have no bad losses and all these good wins and our schedules and records are so similar, we must be doing something right that they aren't.
what? They lost 2 less games, lost those games by far less and beat better teams. Maybe i misunderstood what this post is supposed to be saying.
 
These rankings predict well but they don't tell what kind of season they had. Is the committee giving bids based on predictions or past results? I hope these comparisons are useful, I find that comparisons can shed light.
 
St Johns 13-7
Slightly worse schedule
14th in SRS

Clemson 13-6 worse record
31st

We're getting killed for games where we quit. Maybe that's something that should be punished.

Florida State worse schedule 12-8, ranked higher
The thing that is saving Clemson is going on the road and beating Alabama (at the end of the season that might be this entire conferences best non-conference win). They also beat South Carolina who is tied for 3rd in the SEC at 17-3 and beat a 15-5 Big 12 opponent in TCU at a neutral site. Without those they would be dead in the water based on their sub-par ACC performance so far.
 
Is the committee giving bids based on predictions or past results?
They go by results. Another way to frame what you're asking is: If SU was 18-2 right now, even if their wins over top teams were by 1 point and their metrics were still horrible, would they be a favorite to be a top seed in the tournament? Yes, yes they would be.

The rankings are just one component of overall tournament selection. But few here seem to understand that because Syracuse is a bit of an anomaly this season where their relative efficiency, SOS, and record, don't line up neatly. Because it is SU, some fans here can't process that. But they will forget all about it next season when its Tulane or Clemson or East Nevada State or whoever.

They are were they are in metrics because they've been destroyed by better teams relative to how other teams have played.

Win games, go dancing.
 
These rankings predict well but they don't tell what kind of season they had. Is the committee giving bids based on predictions or past results? I hope these comparisons are useful, I find that comparisons can shed light.

What is “predicting well”? Within a certain margin of a predicted spread X% of the time? If they were dead on 80% of the time, it would kill gambling (and honestly would kill why we watch in the first place). But I’m not sure people appreciate how many wrong predictions are allowable in this “predicting well” concept.

Back a decade ago, I used to follow Rob Vollman on hockey analytics. He always stressed that you can and will get results which are outliers in statistics - so it was key to put some actual thought into whether or not the results made sense. Maybe based on the data used, Syracuse is an outlier this year. Hopefully the committee views it that way and tosses analytical results in the bin if there’s solid reason to think they don’t represent reality.

Having said that, I’ve been worried they do predict reality well, this team so far is fools gold and we’ll drop off a cliff at some point. So…I guess we’ll see.
 
Not every record is built the same even if you have a better schedule. Seems pretty simple. ST. John’s, Clemson, Bama all have better wins than us. We have sht the bed against every good team and we didn’t exactly blow out some of the mid majors we played either(Cornell/Colgate/New Hampshire/Canisius).
 
The thing that is saving Clemson is going on the road and beating Alabama (at the end of the season that might be this entire conferences best non-conference win). They also beat South Carolina who is tied for 3rd in the SEC at 17-3 and beat a 15-5 Big 12 opponent in TCU at a neutral site. Without those they would be dead in the water based on their sub-par ACC performance so far.
Big 12 is so overrated - this years Big 10
 
Big 12 is so overrated - this years Big 10
Fair enough. I don’t know enough about their conference, I only know that TCU has beaten NET #1 Houston and NET #18 Baylor. So that helps Clemson for sure.
 
Point differential is a garbage thing to measure because it promotes running up the score. After 10 points in either direction - win or loss - they should toss the rest of the point differential out the window.
 
Not every record is built the same even if you have a better schedule. Seems pretty simple. ST. John’s, Clemson, Bama all have better wins than us. We have sht the bed against every good team and we didn’t exactly blow out some of the mid majors we played either(Cornell/Colgate/New Hampshire/Canisius).
Two hypothetical teams have the same record and the same strength of schedule.

Say, one has better wins.

What must it say about the relative quality of their losses?

This is zero sum.

If our losses were only be an average of 10 points instead of 22, our SRS would be about 3 points higher (60 points divided by 20 games) , we'd be ranked 50th

If they also scored 3 more points per win, the SRS would be ranked 34th

I know the committee doesn't use SRS but it sure seems like they could've made things look better and still have the same record.
 
What is “predicting well”? Within a certain margin of a predicted spread X% of the time? If they were dead on 80% of the time, it would kill gambling (and honestly would kill why we watch in the first place). But I’m not sure people appreciate how many wrong predictions are allowable in this “predicting well” concept.

Back a decade ago, I used to follow Rob Vollman on hockey analytics. He always stressed that you can and will get results which are outliers in statistics - so it was key to put some actual thought into whether or not the results made sense. Maybe based on the data used, Syracuse is an outlier this year. Hopefully the committee views it that way and tosses analytical results in the bin if there’s solid reason to think they don’t represent reality.

Having said that, I’ve been worried they do predict reality well, this team so far is fools gold and we’ll drop off a cliff at some point. So…I guess we’ll see.
Of course it wouldn't predict 80% for any particular game. But over the course of many games, it does predict well. Florida state on the football side had an ordinary record through part of 2022 but their SRS said they were good and they won every game the rest of that year and the next until the qb got hurt
 
Point differential is a garbage thing to measure because it promotes running up the score. After 10 points in either direction - win or loss - they should toss the rest of the point differential out the window.

I don't love it, because I miss the days of putting your walk-ons in with a couple of minutes left in a blowout.

But it's clearly not "garbage." A team that can only beat you by 2 is not as good as a team that can beat you by 40.
 
Two hypothetical teams have the same record and the same strength of schedule.

Say, one has better wins.

What must it say about the relative quality of their losses?

This is zero sum.

If our losses were only be an average of 10 points instead of 22, our SRS would be about 3 points higher (60 points divided by 20 games) , we'd be ranked 50th

If they also scored 3 more points per win, the SRS would be ranked 34th

I know the committee doesn't use SRS but it sure seems like they could've made things look better and still have the same record.
Better to lose close to mediocre and good teams than get blasted by good teams. Bama lost a close one to Purdue, Clemson to Duke and St. John’s to UConn. Our SRS and Net are pretty similar because they both heavily weight point differential.
 
Last edited:
Of course it wouldn't predict 80% for any particular game. But over the course of many games, it does predict well. Florida state on the football side had an ordinary record through part of 2022 but their SRS said they were good and they won every game the rest of that year and the next until the qb got hurt

in 2021-22 season the vegas favorite won 76% of the time
 
1. Due to vast (too vast for many anyway!) discussions on that other thread that is now 38,000 posts long, we have all figured out why we are getting hurt by NET. We have all learned more about NET because we are one of the outlier teams this year.

2. Thankfully, its my general assessment (opinion), that the NCAA selection committee very much focuses on the quality of the end result in traditional form (W or L) and does not worry overly much about margins, or standalone NET's or RPI before it.

3. I have no issue with KP as a predictive model (margins based). Its generally good for what it is meant to do and gets too much criticism based on things its not even doing.. But a predictive model (which NET has high elements of) should not be the NCAA ranking selection system.
It's an odd ranking system to use, because the Committee does focus largely just on W or L, and no by how much. So they are incongruent with each other.

4. RPI, in theory, is more aligned with the selection committee because it was purely about W and L, and margin is irrelevant. Problem is it had some holes as well, and was gamed, and spat out some garbage.

My simple solution would be as follows:
Run NET as you do know = 50% of your ranking
Run RPI = 50% of your ranking

You will see the outliers in RPI (on the positive side) are often the opposite side in NET.
I think balancing the two out might be more effective.
 
Better to lose close to mediocre and good teams than get blasted by good teams. Bama lost a close one to Purdue, Clemson to Duke and St. John’s to UConn. Our SRS and Net are pretty similar because they both heavily weight point differential.

They also did much better than us in the sub 200 games -- that were all wins for both. Not going to dig up and find the post, but I ran the numbers and Alabama gained more than half the "point differential" difference between the two teams in their 5 easiest games. It was a bigger factor than our bad losses (that was also a big factor of course)
 
Last edited:
I know some haven't seen the numbers that are buried in the mega thread of stat geeking.
But the OP is probably incorrect in saying that the biggest reason for our seemingly weak SRS/NET/KP rankings (whichever one you choose) is the margin in our bad losses. That's still a very big factor, but the biggest factor compared to other teams appears to be our "poorer" performance in our 5 or 6 easiest games of the year. Just let those teams hang around for too long. I was able to show that when comparing us to Pitt and Alabama.

It would be a shame either way, if that is what lets us down, perhaps a bigger shame. But did want to get the proper info on what the biggest factors are.
 
Ironically we have no bad losses, we’ve just lost bad

Depends if you use metrics or eye test. But optically I agree with you and if FSU can keep playing well it should take care of itself either number wise or with the committee (if we are actually there in the end)
 

Similar threads

    • Like
Orangeyes Daily Articles for Friday for Football
Replies
4
Views
479
    • Like
Orangeyes Daily Articles for Friday for Football
Replies
7
Views
714
Orangeyes Daily Articles for Monday for Football
Replies
6
Views
500
    • Like
Orangeyes Daily Articles for Thursday for Football
Replies
5
Views
468

Forum statistics

Threads
167,889
Messages
4,735,614
Members
5,930
Latest member
CuseGuy44

Online statistics

Members online
265
Guests online
1,011
Total visitors
1,276


Top Bottom