[curious]

I have prediction models for various sports.

I am building a more efficient one for NCAAF.

I am trying to determine what the most predictiive metrics are.

For offense I have narrowed them down to:
Starting Field Position (Avg)
Turnover differential
Plays per TD
Drive conversion % (% of drives that resulted in a TD)
Yards per first down (avg....many think that the first down is the most important down of the drive)

Any ideas?

[20Four7]

I don't really bet NCAAF so I can't help you. However, WB curious.

[durito]

Hi Curious

Long time no see.


Be careful with TO differential. Past turnovers don't correlate very highly with future turnovers.

Take a look at yds per pass play and yds per rush (both offensive and defensive of course and compensated for opponents)

[Peep]

points?

[Justin7]

Yards per rushing attempt, yards per passing attempt, pass completion rate.

Both offense and defense.

[Rain Man]

In addition to the metrics already mentioned (points, yards per pass and run on both sides of the ball), I have found it very useful to formulate those metrics around three seperate groups of teams.

Top 25
next 30 - 35 ranking
everyone ranked higher than that

How you rank everything beyond the top 25 is up to you but I use a combination of cbs sportsline and my own grading system. If you use cbs at a minimum, I believe you can find a sharpening of your edge.

[Justin7]

Another way to approach it is to weight games.

If the absolute value of the spread is <=10, you have 2 teams in the same league. These stats are most useful. For spreads >= 30, these are least useful.

[SlickFazzer]

Originally Posted by Justin7

Yards per rushing attempt, yards per passing attempt, pass completion rate.

Both offense and defense.

Do you find these categories equally important for both college and the NFL?

[Wheell]

I break it down to success rates in various situations. For me it is actually pretty complicated, I am not sure how best to describe it without giving up proprietary information. If you are familiar with DVOA, I have a modified version of that for college.

[Justin7]

Originally Posted by SlickFazzer

Do you find these categories equally important for both college and the NFL?

Yes.

[dwaechte]

Originally Posted by Justin7

Yards per rushing attempt, yards per passing attempt, pass completion rate.

Both offense and defense.

I'm intrigued by your inclusion of completion rate.

It seems as if completion rate would only be useful because it would correlate well with success rate, so do you just use it for the college game because of the lack of availability of success rate statistics? Or do you also use it for the pro game where success rate statistics are generally available?

[Justin7]

The higher the completion rate, the lower the volatility on passing yards per attempt. Lower volatility means you have fewer 3rd and long situations - "drive killer" situations.

[curious]

Originally Posted by Justin7

The higher the completion rate, the lower the volatility on passing yards per attempt. Lower volatility means you have fewer 3rd and long situations - "drive killer" situations.

I'm wondering if my use of the drive conversion ratio is another way of looking at the same thing. If the reason you like completion rate is because it is an indicator of drives that were killed it seems reasonable that this is a back door to drive conversion ratio.