Connelly's Pre-Season SP+ Rankings

Cam

Cam

Helluva Engineer
Messages
1,591
Location
Atlanta, Georgia
Bill Connelly is a big stats guy who did a pretty good job predicting the early performance of teams last year with his 2019 preseason rankings. He recently released his 2020 preseason rankings based on his algorithm. He has us ranked 58th overall, but what's most interesting to me is he ranked our defense 35th. This is based on:
1. Returning production. Last week we published my initial returning production rankings for 2020, based on players graduating, leaving for the pros, transferring, etc. Estimating improvement or regression based on these percentages and applying it to last year's SP+ ratings accounts for more than 50% of the overall numbers below.

2. Recent recruiting. After determining how much of last year's team is being brought back, the next step is to determine the caliber of the players who will be filling in the missing returning production. To do that, I use a weighted mix of recent recruiting rankings. They primarily pull from the past two recruiting classes, but I have begun to incorporate older classes as well, to a lesser degree. No matter what the "stars don't matter" crowd will tell you, recruiting rankings are extremely predictive, and these projections are more effective because of them.

3. Recent history. While last year's SP+ ratings are taken into account with the returning production estimates above, I've found that involving previous years' performances as well gives us a nice way of estimating overall program health. It stands to reason that a team that has played well for one year is less likely to duplicate that effort than a team that has been good for years on end (and vice versa), right? Recent history accounts for less than 10% of the overall projections; it is a minor factor, but the projections are better with it than without.
Click to expand...

It's also not explicitly stated what the effects of transfers INTO the program have on his rankings and if that's included in recruiting. I'm wondering if that counts as a production transfer from one team to another. Because I'd imagine the additions of Cochran, Johnson, and Ezzard should dramatically improve our 104th ranking on offense. Either way, it's encouraging to see how much some advanced stats back up the feeling that our defense will be much, much better in the near future.
 

Deleted member 2897

Guest
slugboy said:
Yes, he was called an idiot for projecting us to go 3-9 last season


Sent from my iPhone using Tapatalk
Click to expand...

And deservedly so. There was no reason to go 3-9 last year. And he had a history of predicting us to finish 10-12th in the ACC for years...and then we’d finish 4th and 5th. I think he’s a big idiot.
 
Cam

Cam

Helluva Engineer
Messages
1,591
Location
Atlanta, Georgia
bwelbo said:
And deservedly so. There was no reason to go 3-9 last year. And he had a history of predicting us to finish 10-12th in the ACC for years...and then we’d finish 4th and 5th. I think he’s a big idiot.
Click to expand...
I mean... there was at least SOME reason to go 3-9 last year. Over 50% of his algorithm is predicated on returning production, of which we ranked like 119th last year. And our recruiting strategy on offense in the past threw a lot of numerical metrics out the window. For what it's worth, nobody ever predicted GT under Johnson accurately. Not even GTs fans.
 

Deleted member 2897

Guest
Cam said:
I mean... there was at least SOME reason to go 3-9 last year. Over 50% of his algorithm is predicated on returning production, of which we ranked like 119th last year. And our recruiting strategy on offense in the past threw a lot of numerical metrics out the window. For what it's worth, nobody ever predicted GT under Johnson accurately. Not even GTs fans.
Click to expand...

Right. But we had 1 year of the last 25 where we were under 0.500 in ACC play. He regularly predicted us to.
 
slugboy

slugboy

Moderator
Staff member
Messages
11,726
I posted some of his numbers in the College and Pro Sports forum, but to have a good season it looks like the easiest path is to improve more on offense than expected.
We’re currently projected as the #104 offense. If we improve our blocking to even average levels—not even average P5, but just average–we should be better than that. Our strength of schedule is a hill to climb, though


Sent from my iPhone using Tapatalk
 
CuseJacket

CuseJacket

Administrator
Staff member
Messages
19,626
Also relevant to predicted improvement this year:
Note: In-season, SP+ also takes special teams into account. But I have not yet established a solid way to project special teams ratings, so for these projections it is assumed that everyone's ST rating is 0.0.
Click to expand...
 
smokey_wasp

smokey_wasp

Helluva Engineer
Messages
5,486
AlabamaBuzz said:
Sounds about right.
Click to expand...

Everyone except Duke, Syracuse and Gardner Webb (who obviously doesn't get a ranking). I am still cautiously optimistic for next season, but there just aren't many slam dunk wins on this schedule, so there isn't any margin for error if we want to get to 6 wins. Gotta win the ones we should, plus the toss-ups, plus maybe 1 we shouldn't. Doable, but challenging.
 
Last edited:

a5ehren

Jolly Good Fellow
Messages
484
Transfers in are not included in returning production.

He's addressed it before, basically saying that:
1) There is no way to know where they fit on the depth chart before the fall
2) They're usually transfers for a reason, so it isn't a big deal. If you're counting on (for example) an OL from Vanderbilt to help you dramatically you will probably be disappointed.

In special cases (think Justin Fields) he will use that a reason to disagree with the model in his writing, but he doesn't touch the model to account for it.
 

link3945

Jolly Good Fellow
Messages
101
a5ehren said:
In special cases (think Justin Fields) he will use that a reason to disagree with the model in his writing, but he doesn't touch the model to account for it.
Click to expand...

Which is exactly how you should use a model. A model should not be an end-all-be-all. It should be a quantitative way to get a starting point on every team, based on past experience. From there, you see what the model is saying, and look at why it thinks these things. That can tell you a lot about who may or may not over-achieve.
 
Top