Efficiency Metrics: Opponent-adjusted yards per rush, pass, and play

Ralphie Halves · 06-21-18, 11:59 PM

I'll take a peek soon, but anything like this is certainly welcome here.

Bill Connelly is super smart, just can't pick winners consistently, LOL.

Wake has been a moneymaker the past few years. The books still have't caught up with the fact that Dave Clawson is a monster.

bigwilliestyle · 06-22-18, 12:43 PM

Great, let me know what you think!

Yeah, I think Wake is going to connect on some of the upsets they just missed last year. They posted some gaudy numbers against good opponents.

gojetsgomoxies · 06-22-18, 02:48 PM

aren't connelly and massey ratings (O and D only, not pass and run) adjusted for opponent?

gojetsgomoxies · 06-22-18, 02:50 PM

good stuff. thank you

can you add the column descriptions further down the page too?

Combato · 06-22-18, 03:31 PM

Thanks for the excellent data contained in this SS.

Do you have plans to update this CFB season as the season progresses?

bigwilliestyle · 06-23-18, 12:21 AM

Connelly and Massey-Peabody definitely adjust for opponent. In my opinion though, it’s harder to interpret what their rankings mean. For example, a team with an offensive S&P rating of 38 plays a defense with an S&P rating of 13. What’s the expected result? This question has an answer but it’s not readily apparent.

I’ll definitely look to make UX improvements on the site in the coming months.

And yes, I’ll be updating the site as the games happen so keep checking in as the season progresses!

Thanks for the comments!

gojetsgomoxies · 06-23-18, 12:44 AM

Originally posted by bigwilliestyle

Connelly and Massey-Peabody definitely adjust for opponent. In my opinion though, it’s harder to interpret what their rankings mean. For example, a team with an offensive S&P rating of 38 plays a defense with an S&P rating of 13. What’s the expected result? This question has an answer but it’s not readily apparent.

good stuff................ people seem to assume that a team that averages 40 points on offense plays a team that averages 20 points on defense (both adjusted for opponents) will score 30 points. but i'm not sure there's any real basis for that (and your open-ended question seems to agree with my thinking) other than simplicity and making some vague intuitive sense.

RGriebling · 06-25-18, 11:07 AM

Good Stuff. I will certainly be checking in as the season nears. Thanks for the hard work!

M.W. · 07-01-18, 12:47 AM

Originally posted by bigwilliestyle

I wanted to share this with the community in case anyone found it to be useful. I couldn’t really find all of this information in one, easy-to-reference spot, so I created it myself (scraping box scores and transforming the data in python). Great ranking systems like Bill Connelly’s S&P+ ratings and Brian Fremeau’s FEI ratings certainly exist and are much more technically-involved than these efficiency metrics, but in their complexity, they tend to become harder to interpret. Here, I aim to provide insight into the best teams in college football in terms that are relatable and easy to understand like adjusted yards per rush, pass, and play.

I’ve also generated 2018 projections based on the returning production at each school. (Watch out for Wake Forest?)

I’m looking to add detailed team pages and interesting data visualizations over the course of the season. I would appreciate any and all feedback!

*Adjustments are made for strength of opponents and to remove sack yardage from rushing and reallocate to passing (how the NFL accounts for sack yardage)

**I do the same analysis for the NFL and you can toggle between the two using the links in the top-left corner (still working through NFL projections)

http://parrystats.com

An obvious problem with your ratings is that you didn't back out garbage time. Clemson at #9 is a joke. And Wake at 13? UL at 15? You've got to make some major changes to your formula.

bigwilliestyle · 07-04-18, 04:43 PM

Originally posted by M.W.

An obvious problem with your ratings is that you didn't back out garbage time. Clemson at #9 is a joke. And Wake at 13? UL at 15? You've got to make some major changes to your formula.

I think the garbage time factor is a fair criticism. That said, I've yet to see any proof that garbage time significantly impacts these efficiency numbers. Additionally, what is garbage time? Is it just first-half clock kills and end-of-game kneel downs? Or is it any time a team is up more than 14 points in the second half? What if a team comes back from 21 down in the 3rd quarter? Do I count the drives when they were up or throw them out? The point I'm getting at here is that I'd prefer to include everything rather than introduce my own bias.

I don't think Clemson at #9 is a joke at all. Connelly's S&P+ ratings have them at #8. Clemson had an absolutely elite defense, but their offense left a lot to be desired.
Contrarily, Louisville had an elite offense and a terrible defense. (S&P+ agrees putting them at #16.)

Wake is the odd-ball. But if you dive deeper into their box scores, they performed a whole hell of lot better than their middling record.

I do sincerely appreciate the feedback; you've given me some ideas on things to experiment with and test for possible improvements.

bookie · 07-05-18, 12:28 AM

Thanks BigWillie...The addition I'd like to see is a sortable column by conference to make it easier to see how teams stacked up when they were matched in their own weight class.

M.W. · 07-05-18, 03:50 PM

Originally posted by bigwilliestyle

I don't think Clemson at #9 is a joke at all. Connelly's S&P+ ratings have them at #8. Clemson had an absolutely elite defense, but their offense left a lot to be desired.
Contrarily, Louisville had an elite offense and a terrible defense. (S&P+ agrees putting them at #16.)

Wake is the odd-ball. But if you dive deeper into their box scores, they performed a whole hell of lot better than their middling record.

Connelly knows that his numbers consistently underrate Clemson, partly because Clemson substitutes earlier and more freely than other teams and partly because his model gives SOS about half the weight it should. The Clemson/Wake and Clemson/UL games are good examples of garbage-time numbers corrupting your analysis. The UL game was 33-7 after 3Q and ended 47-21, while the Wake game was 28-0 after 3Q before ending 28-14. The final scores and stats greatly understated the difference in quality in each case.

M.W. · 07-05-18, 04:21 PM

I appreciate the politeness of your reply. I am looking at some other numbers that may be relevant or helpful. I can understand the UL ranking better than the Wake ranking because UL's net stats imply a dominance that was never realized on the field.