Bill Connelly's Georgia Tech Preview

Lotta Booze

Ramblin' Wreck
Messages
776
I think a big overprojection in his formula applied to GT specifically going into this year is the loss of Taquon. His formula is going to count that as a HUGE hit to our offensive production since he was the leading rusher and passer last year but I think most GT fans aren't likely to think we'll be taking a step back there next year with LJ or whoever wins the job. Quirks like that make it hard to apply one formula to all 130 teams and also include the infinite amount of caveats. The flip side of that is if a team lost a Cam Newton who was their leading passer and rusher then yeah, I think most people would expect a big step back offensively. Though it hopefully won't ring true for GT.

One point he does bring up in the article that concerns me is the slow start of Collins' Temple teams previously. Last year Temple lost to Villanova and Buffalo right out the gates in year 2 under CGC but like Bill mentioned finished strong later in the year with some good wins and giving UCF all they can handle. It's hard to imagine the team starting slow with all this JUICE but I don't know enough about that Temple squad to know what the issues were against Villanova. Clemson is Clemson but going into USF I'd hope CGC's familiarity with them previously will give us an edge in the game. Can't wait to see it!
 

Deleted member 2897

Guest
How about this: you produce a preseason ranking and record for all 130 teams and come back after the season and see how well you do in comparison. The truth is that the task is a very difficult one and beyond that results are very significantly affected by random variation. A 100% perfect model would still get a lot wrong.

Like we said, take the last two or three years, average them together, and be done with it. I’m not some paid sports journalist exuding expertise. He is not dumb. He is not evil. He does not have an alterior motive. He is just almost always wrong. He and other people who year after year predict us to be much worse than we ever are.
 
Messages
2,034
LOL... I am reading no pre=season prognostications because how do you predict a team that has no gauge to predict against. We do not know what Offensive system we will be running. We do not know who the starters are. On defense we have no idea who is playing where and what will change from last year. Here is all I can predict.
Wells is a vey good kicker so we are solid in place kicking
Harvin is a great punter so we are solid there
J Thomas is a dynamic return guy so we may get some special teams points.

Other than that forget it. You may all remember that in 2008 all the prognosticators had us at 4-8...How did that turn out.
 

1979jacket

Ramblin' Wreck
Messages
568
The article is well written. I didn't pay much attention to the projections. I know some of you don't agree but CPJ won some big games and his offense was an equalizer but was accompanied with mediocre defenses as the writer states. I only watched one Temple game - that being against Duke - and I agree with him that the offense lacked an identity. Meaning something that was a known attack style i.e. run/short pass/long pass about it that you could count on. The offense looked very scatter shot.

However I do agree that recruiting had gotten awful and Collins hopefully can make it happen. Projections for this team are impossible so arguing over that is ridiculous and amusing. Go Coach Collins and team. I think we are at the epicenter of football and with the right moves could be a recruiting success story but it will take time to make the change from CPJ.
 

GT14

Jolly Good Fellow
Messages
125
The math here is killing me.
The S&P+ numbers are based on last season, minus a weighted factor for the starters we lost, plus a weighted factor for the recruiting class we gained. There's no magic mojo formula adjustment for a new offensive scheme. He'd have made the same prediction if there wasn't a coaching change (and if we'd ended up with the same recruiting class, but that opens up another subject that's not on topic).

(I hope this pastes ok)
Code:
Opponent (Proj. Rk)           Date          Proj. Margin           Win Prob.
at Clemson (3)                29-Aug                 -35.9                       2%
USF (71)                            7-Sep                      -1.9                     46%
The Citadel (NR)            14-Sep                   20.7                     88%
at Temple (66)                28-Sep                    -8.1                    32%
North Carolina (61)          5-Oct                    -4.1                     41%
at Duke (65)                      12-Oct                   -8.9                    30%
at Miami (19)                    19-Oct                  -19.9                    13%
Pittsburgh (59)                 2-Nov                    -4.7                     39%
at Virginia (41)                  9-Nov                   -13.8                   21%
Virginia Tech (30)           16-Nov                  -11.6                    25%
N.C. State (47)                  21-Nov                   -7.3                    34%
Georgia (2)                        30-Nov                   -31.6                   3%
3.7
2% + 46% + 88% + 32% + 41% + 30% + 13% + 39% + 21% + 25% + 34%+ 3% = 370% = 3.7 wins or about 4 wins.
Not 1 win. Not 8. It's 4 wins. He did his math correctly.
He'd have predicted the same with CPJ. That's the nature of what he's doing.
We lost a lot of starting players. Other teams lost fewer starters. That leads to bad matchups.

As an example, last year he said we had a 50/50% shot against USF. We lost and it was close. Around here, most of the people thought we'd blow them out of the water. Similar for the other games. He was better at forecasting us last year than we were.

He's complimentary in the article. He thinks Collins can do great work. He explains exactly what he means by "identity", especially on offense. I didn't like Temple's offense, but I liked their defense and special teams. It makes sense to me.

Thanks a lot for posting this. I am amazed at the number of presumably engineers that don't understand expectation values.
 

RamblinCharger

Helluva Engineer
Messages
1,484
Location
Alabama
Right, we could win only 4 games. We could also win 10. In the last 25 years, we've won 10 or more games 3x as often than we've won 4 or fewer games. We've won 9 or more games 7x as often as we've won 4 or fewer games. So why does everyone always predict the low end option?

If we win 10 games I’ll streak during the UGA game. Gold colored glasses or not... goodness. We’re going from running a spread option to probably throwing the ball 25+ times a game. It’ll be an adjustment.
 

RonJohn

Helluva Engineer
Messages
4,492
Thanks a lot for posting this. I am amazed at the number of presumably engineers that don't understand expectation values.

I understand mathematical models, computer models, and probabilities. One of the foundations of all of those is a well defined understanding of cause and effect. You can build a computer model to determine how long it will take to fill a 5 gallon bucket if you set the flow rate of a garden hose to 1/2 gallon per minute. That computer model will be very accurate.(depending on the accuracy of the flow meter and how close the bucket actually is to 5 gallons) Computer models are at a maximum as good as the model and the data it is fed. This "computer model" from what @slugboy said:
The S&P+ numbers are based on last season, minus a weighted factor for the starters we lost, plus a weighted factor for the recruiting class we gained.
There is an attempt to adjust the S&P numbers from last year based on what was lost and what was gained. If your top running back is still on the team but the three best offensive linemen are gone, are his numbers going to remain the same or decline? If you have a 5 star QB recruit joining the team is he a Trevor Lawrence or a Justin Fields?

Computer models are only as good as the model and the data. The data in this case isn't scientifically derived. I would say on average, the predictions are probably close. It is as good as any method to get clicks and to promote discussion. However looking at an individual team or an individual game, the model isn't worth anything. I am not saying that because the GT prediction is low. I am saying that because it isn't a scientific model, and it doesn't use scientifically derived data. I would say the same thing if it predicted GT to be in the CFP.
 

GT14

Jolly Good Fellow
Messages
125
I understand mathematical models, computer models, and probabilities. One of the foundations of all of those is a well defined understanding of cause and effect. You can build a computer model to determine how long it will take to fill a 5 gallon bucket if you set the flow rate of a garden hose to 1/2 gallon per minute. That computer model will be very accurate.(depending on the accuracy of the flow meter and how close the bucket actually is to 5 gallons) Computer models are at a maximum as good as the model and the data it is fed. This "computer model" from what @slugboy said:
There is an attempt to adjust the S&P numbers from last year based on what was lost and what was gained. If your top running back is still on the team but the three best offensive linemen are gone, are his numbers going to remain the same or decline? If you have a 5 star QB recruit joining the team is he a Trevor Lawrence or a Justin Fields?

Computer models are only as good as the model and the data. The data in this case isn't scientifically derived. I would say on average, the predictions are probably close. It is as good as any method to get clicks and to promote discussion. However looking at an individual team or an individual game, the model isn't worth anything. I am not saying that because the GT prediction is low. I am saying that because it isn't a scientific model, and it doesn't use scientifically derived data. I would say the same thing if it predicted GT to be in the CFP.

I agree emphatically with you on your assessment of the model only being as good as the data. Crap in = crap out. But to your later points, isn't that was the formula tries to capture with the people leaving/people coming in?

I agree with you that it's not a theoretical model. You can derive and calculate how quickly a bucket will fill with water which is why a computer model for that system is easy. What he has here is an empirical formula. You can't derive or calculate how many wins a team will have, but based on past data, this is the most accurate model he has come up with.

I largely agree with you on the difference between a Lawrence vs. Fields argument, but of course we can't know the outcome a priori. The data will never be extremely clean, but this does the best with what is provided.
 

stinger 1957

Helluva Engineer
Messages
1,283
The one thing I've worried about with the way CGC runs his practices is kids making a lot of mistakes when they play Not positive this could be a problem, but makes sense that it could be. Don't know why his teams at Temple slow start, just wonder if this is the reason, too many mistakes.
 

smathis30

Ramblin' Wreck
Messages
732
Right, we could win only 4 games. We could also win 10. In the last 25 years, we've won 10 or more games 3x as often than we've won 4 or fewer games. We've won 9 or more games 7x as often as we've won 4 or fewer games. So why does everyone always predict the low end option?
7 returning starters. We’ve on average been predicted to get 3rd/4th in the Coastal (and have on average finished there
I understand mathematical models, computer models, and probabilities. One of the foundations of all of those is a well defined understanding of cause and effect. You can build a computer model to determine how long it will take to fill a 5 gallon bucket if you set the flow rate of a garden hose to 1/2 gallon per minute. That computer model will be very accurate.(depending on the accuracy of the flow meter and how close the bucket actually is to 5 gallons) Computer models are at a maximum as good as the model and the data it is fed. This "computer model" from what @slugboy said:
There is an attempt to adjust the S&P numbers from last year based on what was lost and what was gained. If your top running back is still on the team but the three best offensive linemen are gone, are his numbers going to remain the same or decline? If you have a 5 star QB recruit joining the team is he a Trevor Lawrence or a Justin Fields?

Computer models are only as good as the model and the data. The data in this case isn't scientifically derived. I would say on average, the predictions are probably close. It is as good as any method to get clicks and to promote discussion. However looking at an individual team or an individual game, the model isn't worth anything. I am not saying that because the GT prediction is low. I am saying that because it isn't a scientific model, and it doesn't use scientifically derived data. I would say the same thing if it predicted GT to be in the CFP.
.... it is scientifically serviced though
 

YellowJacketFan2018

Helluva Engineer
Messages
9,022
Location
Southeast Tennessee
The one thing I've worried about with the way CGC runs his practices is kids making a lot of mistakes when they play Not positive this could be a problem, but makes sense that it could be. Don't know why his teams at Temple slow start, just wonder if this is the reason, too many mistakes.
I expect there would be a lot of mistakes since Georgia Tech is changing to a radically different offense...........................................just sayin':cigar:
 

Deleted member 2897

Guest
7 returning starters. We’ve on average been predicted to get 3rd/4th in the Coastal (and have on average finished there

.... it is scientifically serviced though

Since the ACC went to Divisions, our average finishing position in the Coastal is 2.7. We've been 1st on 4 different occasions. We've been 2nd twice (including last year), and 3rd 5 times.

We've only finished in the bottom half of the Coastal Division (4th or worse) 3 of the 14 years.
 

Cam

Helluva Engineer
Messages
1,591
Location
Atlanta, Georgia
Here's Pete Fiutak of College Football News preview of the 2019 Georgia Tech football team. He thinks Georgia Tech will go 6-6.

https://collegefootballnews.com/2019/05/georgia-tech-football-preview-prediction-players-2019
Pretty clean write up. I love these reviews from an outside perspective. I particularly love seeing all these media pieces, tweets, and quotes that praise the staff that Collins put together (Fiutak says they are "terrific"). It really is great that we were able to pull together such a great group of like-minded coaches that all share Collins' vision. I've been especially pleasantly surprised by the amount of praise Marco Coleman has been receiving as a coach from the players, considering it's his first year at this level and second year coaching overall.
 

RonJohn

Helluva Engineer
Messages
4,492
I agree emphatically with you on your assessment of the model only being as good as the data. Crap in = crap out. But to your later points, isn't that was the formula tries to capture with the people leaving/people coming in?

I agree with you that it's not a theoretical model. You can derive and calculate how quickly a bucket will fill with water which is why a computer model for that system is easy. What he has here is an empirical formula. You can't derive or calculate how many wins a team will have, but based on past data, this is the most accurate model he has come up with.

I largely agree with you on the difference between a Lawrence vs. Fields argument, but of course we can't know the outcome a priori. The data will never be extremely clean, but this does the best with what is provided.

I do agree that this is probably as good a system as can be made. I would even say that on average it is probably decent. I only push back because with the way some people post, it appears that they believe that since it is mathematical and done on a computer that the results are accurate.
 
Top