Rob Bowron @Beta_Rank_FB
2019 Beta_Rank Projections
Updated: Aug 13, 2019
Rolling out the predicted or preseason Beta_Rank is always one of my favorite parts of the football off-season. Pre-season Beta_Rank performed pretty well last year and got an impressive number, 57%, of teams within +-5 of where they ended up in final Beta_Rank.
I will include the actual ranking below, but what goes into them?
Firstly, last year's Beta_Rank. If you were good last year chances are you will be good again. It's not perfect and teams do fluctuate, but its a good place to start.
Second I use Bill Connelly's returning production. I like these numbers more than returning starters and they are more predictive than using returning starters. If you think of the two variables we have already its what do you have coming back and were they any good.
Finally I add two years worth of recruiting rankings to account for who are your replacements for your production you are losing. At some level the prior Beta_Rank score and the recruiting score also account for coaching since recruiting and schematics are two of the most important factors in evaluating an NCAA coach.
Some math preferences from me. I mostly hate averaging data, but I do average the two year recruiting rankings for simplicity sake. I don't get a worthwhile bump in fit from running a lag there. I do some aggressive feature engineering on the variables that I include and it adds significantly to the fit and predictive power. I don't go far back in Beta_Rank lags because I like to keep the predictor fresh; even if it makes it a little less accurate overall. It's a preference, but going too far back gets into messy territory.
Messy territory like coaching changes. I don't track them and I don't have a field for them and to be honest if I was going to do it I would not just track head coaching changes which fans care about A LOT, but coordinator changes which fans pay way less attention to. I often think the head coach change effect is likely more to do with the usual change at coordinator that happens simultaneously than just the head guy effect.
How the data works together is interesting because not all the effects are similar. Offense is more tightly correlated with returning production than recruiting. Offense has the initiative in football and knowing the scheme is more critical than recruiting rankings. Teams with a ton of returning offensive production are often candidates for a jump. Recruiting is more correlated with defense than is returning defensive production. Defense is reactive and while scheme matters, getting great athletes who can react is often more important. Teams that recruit well are always candidates for improvement; though it may not show up in their record vs. the advanced stats.
Some things that jump out at me: Florida State and Tennessee are candidates for a big jump. They have just recruited so well that it's hard to be bad for long. The SEC had a great year last year and recruited extremely well again, so you should not be surprised to see a lot of SEC teams near the top. My model tends to believe a little less in the Huskies and a little more in the Ducks and Utes in the Pac-12 due to Washington's major production losses. Texas looks kind of back.
So without further ado, here is the Predicted Beta_Rank for 2019.