Need in help in developing an algorithm for the 2014 college football season

[HuskerOC]

+/- yards gained against an opponent vs your opponents +/- yards gained.


Obviously the more plus the better.

 

[TheRedPlectrum]

Weighting home v. away games differently as many have said.

I would also add a margin-of-victory component, but limited to two options. In my opinion, there's no real difference between winning by 20 and winning by 30, but there is a huge difference between winning by 2 and winning by 20. So, I would probably put the cut-off at 10. Wins by 10 or less v. by 11+. I'm only talking about a slight weighting, mind all. Something like 1.2 v. 0.8.

 

[Codaxx]

+/- yards gained against an opponent vs your opponents +/- yards gained.


Obviously the more plus the better.

Lot of stuff is out there based on play measurements. Your results vs your opponents expected results. How your offense/defense performs vs an average unit. The biggest issue for 90% of people is gaining access to all the information. Even just total yards requires loading 70 some games a weekend. To do any in-depth formula requires a ton of man hours. I don't think the OP was looking to dedicate that kind of time into this project

 

[Codaxx]

Weighting home v. away games differently as many have said.

I would also add a margin-of-victory component, but limited to two options. In my opinion, there's no real difference between winning by 20 and winning by 30, but there is a huge difference between winning by 2 and winning by 20. So, I would probably put the cut-off at 10. Wins by 10 or less v. by 11+. I'm only talking about a slight weighting, mind all. Something like 1.2 v. 0.8.

This is difficult. Team A is up by 30 in the 3Q vs team C, but ends up winning by 10. Team B is up by 3 vs Team C and scores a TD with 10 seconds left. Margin of victories are the same, but clearly Team A was more dominant. Why some throw out results when the game is out of hand.

 

[WhiteMamba]

WAT

 

[4down20]

+/- yards gained against an opponent vs your opponents +/- yards gained.


Obviously the more plus the better.

Gets closer, except you can't treat running for 400 yards against a bad team the same as running for 400 yards against Alabama to measure how good the run game is for a team.

 

[4down20]

Weighting home v. away games differently as many have said.

I would also add a margin-of-victory component, but limited to two options. In my opinion, there's no real difference between winning by 20 and winning by 30, but there is a huge difference between winning by 2 and winning by 20. So, I would probably put the cut-off at 10. Wins by 10 or less v. by 11+. I'm only talking about a slight weighting, mind all. Something like 1.2 v. 0.8.

What you are talking about is basically what is called garbage points/time. It varies by quarter as to when it kicks in. I don't remember the number, but if you are up by like 35 in the 2nd quarter, it's considered garbage time, and then goes down by quarter. 28 in the 3rd, 21 in 4th. Some where around that.

 

[4down20]

Lot of stuff is out there based on play measurements. Your results vs your opponents expected results. How your offense/defense performs vs an average unit. The biggest issue for 90% of people is gaining access to all the information. Even just total yards requires loading 70 some games a weekend. To do any in-depth formula requires a ton of man hours. I don't think the OP was looking to dedicate that kind of time into this project

All the information is free to anyone who wants to use it.

cfbstats.com - College Football Statistics

I use them and recommend them.

I just run 2 commands every week to update my rankings. Takes less than 5 minutes.

 

[KansasSooner]

Lot of stuff is out there based on play measurements. Your results vs your opponents expected results. How your offense/defense performs vs an average unit. The biggest issue for 90% of people is gaining access to all the information. Even just total yards requires loading 70 some games a weekend. To do any in-depth formula requires a ton of man hours. I don't think the OP was looking to dedicate that kind of time into this project

That's the method I use and most of the stats necessary can be found on the NCAA stat site. I do use the Draper-Smith stepwise regression procedure to include only those variables that are significant and help explain the variance according a to F-test statistic. However I have also found SOS does have an impact on what the expected results will be and it's really hard to get a good handle on SOS due to so many FCS teams being played and trying to use those results within a FBS only model.

That said, the best model I've been able to produce still has a residual error that is 25% of the response mean, too high to be a very reliable model even though the model is statistically significant at the 95% level.

 

[Jack_John_Mark]

The way my old program used to work is that it took points scored per yard gained and points given up per yard given up against each team played and then compared it to how the other teams fared against the same competition, and how their competition compared against their competition, and so on and so forth.

Eventually the program completed a round-robin where every team faced every other team in the nation and gave a score prediction.

I ended up winning the Bowl Pickems that year using the program with almost 90% correct picks. I don't know if it was extreme luck, or if the program had finally reached enough data at that point in the season to be accurate.

 

[Jack_John_Mark]

The program literally picked the O/U within 1-2 points of the Vegas O/U every single time, too.

Which made me suspicious that maybe I was on to Vegas' ways of determining lines.

 

[4down20]

The way my old program used to work is that it took points scored per yard gained and points given up per yard given up against each team played and then compared it to how the other teams fared against the same competition, and how their competition compared against their competition, and so on and so forth.

Eventually the program completed a round-robin where every team faced every other team in the nation and gave a score prediction.

I ended up winning the Bowl Pickems that year using the program with almost 90% correct picks. I don't know if it was extreme luck, or if the program had finally reached enough data at that point in the season to be accurate.

90% is good, 70-75% is about where most end up I've seen.

Last year was brutal for me. I did less than 50%. In the entire time it's been doing them I never had anything below like 60%, and that was rare in itself. Was a bit shocking, but that was pretty much the norm for last year with all the upsets.

 

[4down20]

The program literally picked the O/U within 1-2 points of the Vegas O/U every single time, too.

Which made me suspicious that maybe I was on to Vegas' ways of determining lines.

I pick the same or close to the line a good bit. Makes it hard to use for pick ems that use the spread. You have the team at -21, vegas has it at -20. Which do you pick?

 

[Jack_John_Mark]

I pick the same or close to the line a good bit. Makes it hard to use for pick ems that use the spread. You have the team at -21, vegas has it at -20. Which do you pick?

I always like my number over Vegas' number because Vegas isn't always honest to start out, and their line will move based on how people are betting.

So it's easy to log on to your bookie and see a line and think "damn, that looks good." But you don't know if the line has already moved 5 points or what has happened. That's why I like having my own number.

 

[WNY_FOOTBALL_DUDE]

Not really. If I knew some great simple end all formula like that, I'd already be using it.

If you don't want to propose any ideas for me, then why bother responding? This is one of the main reasons why I don't like debating with you, because you don't like to respond to the points presented, and your viewpoint always revolves around "anything that helps Bama." Good night.

 

[4down20]

If you don't want to propose any ideas for me, then why bother responding? This is one of the main reasons why I don't like debating with you, because you don't like to respond to the points presented, and your viewpoint always revolves around "anything that helps Bama." Good night.

I gave you helpful feedback.

But hey, what do I know about the topic right?

 

[WizardHawk]

90% is good, 70-75% is about where most end up I've seen.

Last year was brutal for me. I did less than 50%. In the entire time it's been doing them I never had anything below like 60%, and that was rare in itself. Was a bit shocking, but that was pretty much the norm for last year with all the upsets.

So we can treat your predictions the same as OD's now?

 

[4down20]

So we can treat your predictions the same as OD's now?

He probably did better. You can flip a coin and should get somewhat close to 50%.

 

[Codaxx]

All the information is free to anyone who wants to use it.

cfbstats.com - College Football Statistics

I use them and recommend them.

I just run 2 commands every week to update my rankings. Takes less than 5 minutes.

I wanted something deeper. I know people use total yards and play data. I wanted to look into doing something derived from drive data. I have not seen that.

 

[4down20]

I wanted something deeper. I know people use total yards and play data. I wanted to look into doing something derived from drive data. I have not seen that.

It has drive data.

in the play.csv, which lists and gives the stats for every play:

"Game Code","Play Number","Period Number","Clock","Offense Team Code","Defense Team Code","Offense Points","Defense Points","Down","Distance","Spot","Play Type","Drive Number","Drive Play"