Another Damn Simulation

[besserheimerphat]

With Williams and Blum putting out a CF Preseason Ranking, I figured I would start putting together a conference record modeler. Obviously this is just for fun and I don't really think this is indicative of how the season will go. I started with their rankings, O/U from CBS, and the conference schedule. I used that information to create the distribution parameters from which I drew the random numbers that represent how well a team plays in a particular game. Whichever team gets the higher number "wins" the game, and the result of that game updates the distribution parameters. When you win, your random numbers skew higher and when you lose your random numbers skew lower. For everyone's first game, I equally weighted the preseason information. As games were played, I phased out the impact of preseason information. Then I simulated 5k seasons and plotted histograms of conference winning percentage. This way the unbalanced schedule gets modeled - who you play and when you play them matters.

I ran several different versions where the preseason-phase-out-period set to 1, 3, 6, and 9 games. Basically that means that by the end of the season only the records count, but how quickly the record overtakes the preseason information changes. I also ran one version with a "12 game phase out" which practically means even at the end of the season the preseason gets a 25% weight.

There wasn't much difference between the 6, 9 and 12 game phase outs - ASU, KSU, ISU and Tech were consistently towards the top. West Virginia and Arizona were at the bottom. As expected, phasing out the preseason information faster resulted in more variation.



With only 3 games of preseason influence, the impact of unbalanced schedules started to appear more strongly. Suddenly Kansas and BYU, who's first three games are against the bottom teams, are in the top 3 of the league. As they win those early games and their preseason ranking influence goes to 0, they look like great teams. Conversely, ISU, KSU and Tech take pretty big hits as ISU and KSU play each other in a toss-up and Tech plays three top-half teams. When the preseason information goes away so quickly, a single early loss can lead to a spiral.


And for you guys who think only results matter, here is the version where preseason info is only used for the first game. After that, it's all 100% driven by game results. Total chaos, with nearly every team capable of finishing almost anywhere. Again you can see the impact of schedule on Kansas and BYU, as well as Colorado and OSU.





As I have time I might try this again with other conferences. I can also try different weighting for both the CF Ranking and CBS O/U, or I can incorporate other sites' preseason information.

 

[CycloneRulzzz]

Sorry but title immediately takes me this

 

[BWRhasnoAC]

Nerd! (And I like it)

 

[madguy30]

So is ISU going to win or no?

 

[Cyclones8824]

Always with the scenarios

 

[ribsnwhiskey]

So is ISU going to win or no?

Yes and no.

 

[2speedy1]

I have absolutely no idea what I am looking at here.

 

[JM4CY]

With Williams and Blum putting out a CF Preseason Ranking, I figured I would start putting together a conference record modeler. Obviously this is just for fun and I don't really think this is indicative of how the season will go. I started with their rankings, O/U from CBS, and the conference schedule. I used that information to create the distribution parameters from which I drew the random numbers that represent how well a team plays in a particular game. Whichever team gets the higher number "wins" the game, and the result of that game updates the distribution parameters. When you win, your random numbers skew higher and when you lose your random numbers skew lower. For everyone's first game, I equally weighted the preseason information. As games were played, I phased out the impact of preseason information. Then I simulated 5k seasons and plotted histograms of conference winning percentage. This way the unbalanced schedule gets modeled - who you play and when you play them matters.

I ran several different versions where the preseason-phase-out-period set to 1, 3, 6, and 9 games. Basically that means that by the end of the season only the records count, but how quickly the record overtakes the preseason information changes. I also ran one version with a "12 game phase out" which practically means even at the end of the season the preseason gets a 25% weight.

There wasn't much difference between the 6, 9 and 12 game phase outs - ASU, KSU, ISU and Tech were consistently towards the top. West Virginia and Arizona were at the bottom. As expected, phasing out the preseason information faster resulted in more variation.

View attachment 152463 View attachment 152464 View attachment 152465

With only 3 games of preseason influence, the impact of unbalanced schedules started to appear more strongly. Suddenly Kansas and BYU, who's first three games are against the bottom teams, are in the top 3 of the league. As they win those early games and their preseason ranking influence goes to 0, they look like great teams. Conversely, ISU, KSU and Tech take pretty big hits as ISU and KSU play each other in a toss-up and Tech plays three top-half teams. When the preseason information goes away so quickly, a single early loss can lead to a spiral.

View attachment 152466
And for you guys who think only results matter, here is the version where preseason info is only used for the first game. After that, it's all 100% driven by game results. Total chaos, with nearly every team capable of finishing almost anywhere. Again you can see the impact of schedule on Kansas and BYU, as well as Colorado and OSU.


View attachment 152467


As I have time I might try this again with other conferences. I can also try different weighting for both the CF Ranking and CBS O/U, or I can incorporate other sites' preseason information.

 

[CtownCyclone]

look, man, I've got certain information, all right? Certain things have come to light. And, you know, has it ever occurred to you, that, instead of, uh, you know, running around, uh, uh, blaming me, you know, given the nature of all this new ****, you know, I-I-I-I... this could be a-a-a-a lot more, uh, uh, uh, uh, uh, uh, complex, I mean, it's not just, it might not be just such a simple... uh, you know?

 

[AdRock4Cy]

look, man, I've got certain information, all right? Certain things have come to light. And, you know, has it ever occurred to you, that, instead of, uh, you know, running around, uh, uh, blaming me, you know, given the nature of all this new ****, you know, I-I-I-I... this could be a-a-a-a lot more, uh, uh, uh, uh, uh, uh, complex, I mean, it's not just, it might not be just such a simple... uh, you know?

She kidnapped herself, man.

 

[RagingCloner]

With Williams and Blum putting out a CF Preseason Ranking, I figured I would start putting together a conference record modeler. Obviously this is just for fun and I don't really think this is indicative of how the season will go. I started with their rankings, O/U from CBS, and the conference schedule. I used that information to create the distribution parameters from which I drew the random numbers that represent how well a team plays in a particular game. Whichever team gets the higher number "wins" the game, and the result of that game updates the distribution parameters. When you win, your random numbers skew higher and when you lose your random numbers skew lower. For everyone's first game, I equally weighted the preseason information. As games were played, I phased out the impact of preseason information. Then I simulated 5k seasons and plotted histograms of conference winning percentage. This way the unbalanced schedule gets modeled - who you play and when you play them matters.

I ran several different versions where the preseason-phase-out-period set to 1, 3, 6, and 9 games. Basically that means that by the end of the season only the records count, but how quickly the record overtakes the preseason information changes. I also ran one version with a "12 game phase out" which practically means even at the end of the season the preseason gets a 25% weight.

There wasn't much difference between the 6, 9 and 12 game phase outs - ASU, KSU, ISU and Tech were consistently towards the top. West Virginia and Arizona were at the bottom. As expected, phasing out the preseason information faster resulted in more variation.

View attachment 152463 View attachment 152464 View attachment 152465

With only 3 games of preseason influence, the impact of unbalanced schedules started to appear more strongly. Suddenly Kansas and BYU, who's first three games are against the bottom teams, are in the top 3 of the league. As they win those early games and their preseason ranking influence goes to 0, they look like great teams. Conversely, ISU, KSU and Tech take pretty big hits as ISU and KSU play each other in a toss-up and Tech plays three top-half teams. When the preseason information goes away so quickly, a single early loss can lead to a spiral.

View attachment 152466
And for you guys who think only results matter, here is the version where preseason info is only used for the first game. After that, it's all 100% driven by game results. Total chaos, with nearly every team capable of finishing almost anywhere. Again you can see the impact of schedule on Kansas and BYU, as well as Colorado and OSU.


View attachment 152467


As I have time I might try this again with other conferences. I can also try different weighting for both the CF Ranking and CBS O/U, or I can incorporate other sites' preseason information.

 

[VeloClone]

So you are saying the more the preseason expectations continue to be factored in the more skewed the results are? Are you sure we aren't talking about the AP poll here?

 

[Cycsk]

With Williams and Blum putting out a CF Preseason Ranking, I figured I would start putting together a conference record modeler. Obviously this is just for fun and I don't really think this is indicative of how the season will go. I started with their rankings, O/U from CBS, and the conference schedule. I used that information to create the distribution parameters from which I drew the random numbers that represent how well a team plays in a particular game. Whichever team gets the higher number "wins" the game, and the result of that game updates the distribution parameters. When you win, your random numbers skew higher and when you lose your random numbers skew lower. For everyone's first game, I equally weighted the preseason information. As games were played, I phased out the impact of preseason information. Then I simulated 5k seasons and plotted histograms of conference winning percentage. This way the unbalanced schedule gets modeled - who you play and when you play them matters.

I ran several different versions where the preseason-phase-out-period set to 1, 3, 6, and 9 games. Basically that means that by the end of the season only the records count, but how quickly the record overtakes the preseason information changes. I also ran one version with a "12 game phase out" which practically means even at the end of the season the preseason gets a 25% weight.

There wasn't much difference between the 6, 9 and 12 game phase outs - ASU, KSU, ISU and Tech were consistently towards the top. West Virginia and Arizona were at the bottom. As expected, phasing out the preseason information faster resulted in more variation.

View attachment 152463 View attachment 152464 View attachment 152465

With only 3 games of preseason influence, the impact of unbalanced schedules started to appear more strongly. Suddenly Kansas and BYU, who's first three games are against the bottom teams, are in the top 3 of the league. As they win those early games and their preseason ranking influence goes to 0, they look like great teams. Conversely, ISU, KSU and Tech take pretty big hits as ISU and KSU play each other in a toss-up and Tech plays three top-half teams. When the preseason information goes away so quickly, a single early loss can lead to a spiral.

View attachment 152466
And for you guys who think only results matter, here is the version where preseason info is only used for the first game. After that, it's all 100% driven by game results. Total chaos, with nearly every team capable of finishing almost anywhere. Again you can see the impact of schedule on Kansas and BYU, as well as Colorado and OSU.


View attachment 152467


As I have time I might try this again with other conferences. I can also try different weighting for both the CF Ranking and CBS O/U, or I can incorporate other sites' preseason information.

“And for you guys who think only results matter, here is the version where preseason info is only used for the first game. After that, it's all 100% driven by game results. Total chaos, with nearly every team capable of finishing almost anywhere. Again you can see the impact of schedule on Kansas and BYU, as well as Colorado and OSU.”

This seems like the most likely of your calculations. Total chaos!

The good news is, we have proven our ability to win in chaos, namely, last year, winning 11 games, despite devastating losses to our linebackers. If we don’t lose a position group this year, I like our chances to do well in total chaos for the rest of the league

 

[besserheimerphat]

So you are saying the more the preseason expectations continue to be factored in the more skewed the results are? Are you sure we aren't talking about the AP poll here?

I get what you're saying and to an extent I agree. I think many sports talking heads hold on to their priors too long, and are too quick to dismiss contrary evidence to maintain their narrative.

But many professional analysts have shown that retaining some of that preseason information for the duration of the season improves model ability to accurately predict outcomes. Those preseason projections come from somewhere, and they have some value.

 

[Kettes]

look, man, I've got certain information, all right? Certain things have come to light. And, you know, has it ever occurred to you, that, instead of, uh, you know, running around, uh, uh, blaming me, you know, given the nature of all this new ****, you know, I-I-I-I... this could be a-a-a-a lot more, uh, uh, uh, uh, uh, uh, complex, I mean, it's not just, it might not be just such a simple... uh, you know?

 

[ElephantPie]

I've read a few recent articles by Bill Connelly and I found it interesting how much he brought up the results of 1 score games for Big 12 teams relative to the success of their season. Shouldn't be surprising in retrospect; especially for a conference that is more evenly matched top to bottom.
1. About ASU

2. About Big XII

3. About ISU

 

[HFCS]

With Williams and Blum putting out a CF Preseason Ranking, I figured I would start putting together a conference record modeler. Obviously this is just for fun and I don't really think this is indicative of how the season will go. I started with their rankings, O/U from CBS, and the conference schedule. I used that information to create the distribution parameters from which I drew the random numbers that represent how well a team plays in a particular game. Whichever team gets the higher number "wins" the game, and the result of that game updates the distribution parameters. When you win, your random numbers skew higher and when you lose your random numbers skew lower. For everyone's first game, I equally weighted the preseason information. As games were played, I phased out the impact of preseason information. Then I simulated 5k seasons and plotted histograms of conference winning percentage. This way the unbalanced schedule gets modeled - who you play and when you play them matters.

I ran several different versions where the preseason-phase-out-period set to 1, 3, 6, and 9 games. Basically that means that by the end of the season only the records count, but how quickly the record overtakes the preseason information changes. I also ran one version with a "12 game phase out" which practically means even at the end of the season the preseason gets a 25% weight.

There wasn't much difference between the 6, 9 and 12 game phase outs - ASU, KSU, ISU and Tech were consistently towards the top. West Virginia and Arizona were at the bottom. As expected, phasing out the preseason information faster resulted in more variation.

View attachment 152463 View attachment 152464 View attachment 152465

With only 3 games of preseason influence, the impact of unbalanced schedules started to appear more strongly. Suddenly Kansas and BYU, who's first three games are against the bottom teams, are in the top 3 of the league. As they win those early games and their preseason ranking influence goes to 0, they look like great teams. Conversely, ISU, KSU and Tech take pretty big hits as ISU and KSU play each other in a toss-up and Tech plays three top-half teams. When the preseason information goes away so quickly, a single early loss can lead to a spiral.

View attachment 152466
And for you guys who think only results matter, here is the version where preseason info is only used for the first game. After that, it's all 100% driven by game results. Total chaos, with nearly every team capable of finishing almost anywhere. Again you can see the impact of schedule on Kansas and BYU, as well as Colorado and OSU.


View attachment 152467


As I have time I might try this again with other conferences. I can also try different weighting for both the CF Ranking and CBS O/U, or I can incorporate other sites' preseason information.

The way Arizona and Arizona State looked coming into the league absolutely nobody would have picked ASU #1 and Arizona #16 in any of these early years of the new league.

It's like a million to one odds it would have turned out this way. ASU being the worst team and Arizona winning it all had to have been radically more likely looking at past data as they came in.

 

[besserheimerphat]

I've read a few recent articles by Bill Connelly and I found it interesting how much he brought up the results of 1 score games for Big 12 teams relative to the success of their season. Shouldn't be surprising in retrospect; especially for a conference that is more evenly matched top to bottom.
1. About ASU View attachment 152500

2. About Big XII View attachment 152501

3. About ISU View attachment 152502

I DM'd him on BlueSky about this topic - ISU won their one score games last year with a lot of late heroics, but as noted an unreal number of Rocco's INTs were Pick-6s. I could argue that those one-score-late-win games would not have been classified that way without the Pick-6. I know he does a lot of work on "luck," comparing things that are/are not repeatable from one year to the next. He said he didn't have anything on Pick-6s specifically.

 

[VeloClone]

I DM'd him on BlueSky about this topic - ISU won their one score games last year with a lot of late heroics, but as noted an unreal number of Rocco's INTs were Pick-6s. I could argue that those one-score-late-win games would not have been classified that way without the Pick-6. I know he does a lot of work on "luck," comparing things that are/are not repeatable from one year to the next. He said he didn't have anything on Pick-6s specifically.

The veracity of the vulnerability of a team who won a lot of one score games also comes down to whether the team was in a lot of nailbiter games or whether they were the victim of a lot of back door covers. That is, did they eak out a win or did they give up a meaningless score in the waning seconds of a game they already had comfortably won. Without looking, my memory suggests that last year ISU had a lot of the nailbiting heroics rather than being up by 2 scores and giving up a late meaningless score to make it look close. That is concerning for this year.

Similarly if a team back door covered a lot of losses to look better on the final score column it doesn't necessarily bode well for them turning those losses into wins the next time.

Kansas found every conceivable way to snatch defeat from the jaws of victory early last year. So that is why most people who were paying attention weren't shocked by their end of season run. They finally figured out how to get out of their own way for a few games.

 

[besserheimerphat]

Just looked it up, and last year we were 4-1 in games decided by <7 pts. In two of those games (wins against UCF and Utah) Rocco threw a Pick-6 in the second quarter. Take away those early 7 points swings and those aren't 1 score games anymore. The other Pick-6 was late in the fourth quarter against KU, which we lost by 9 - sure it impacted the score, but it didn’t impact the game like the others did.

So 3 Pick-6s on 9 INTs, two of which really impacted the game and resulted in one-score wins.