Spread Quarterback RankingsIntroduction
A few games into this season, it really became clear to me that this “Spread and Shred” offense is massively quarterback dependent. So much so, that I began to believe that the performance of the defense is almost irrelevant. Now, starting in the second half of the Illinois game it became pretty obvious that the defense needs to be at least competent to make this all work, but I still feel that this offense is so quarterback dependent that the team’s fate lies almost exclusively with the play of the quarterback. The main metric to measure quarterbacks currently in use is the quarterback passer rating system. This system completely ignores a QB’s rushing stats and is therefore not particularly relevant to this offense. So, I set out to design a “spread quarterback rating” and correlate it to winning football games. Methods
With the help of FormerlyAnonymous and a few other posters, I compiled a list of teams that run a pretty similar offense to UM’s current system and used ESPN.com to collect quarterback stats for all of these teams. I ended up with a sample of 222 games played in similar systems. I’d be happy to email the excel sheet to anyone that is interested (or hopefully can do better than I did with the data). I classified the results of each game into four categories. 1 = loss by more than 10, 2 = loss by 10 or less, 3 = win by 10 or less, 4 = win by more than 10. I basically then just played with the numbers as much as I could to come up with a metric that had the highest possible R-Squared value correlating spread quarterback rating (SQBR) to game result. The final formula came out to be…
SQBR = (Pass Yds / Attempts ^ 0.5) + (Rush Yds / Attempts ^0.5) + PassTD *5 + RushTD * 4 – INT*11 – Sacks*7
Please realize this was done 100% empirically, so that’s why it seems weird. But serious, look at the passer rating formula and this is nothing. Results
As stated above, this was all manipulated to fit the data. The following graph plots game result vs SQBR.
The trendline shows an r-squared value of 0.3901. Obviously this is far short of the perfect correlation of 1, but as the following data will show, it’s better than anything out there. Below is the same data set plotted against conventional passer rating.
The conventional passer rating system correlates approximately half as well as the SQBR for game results.
For a little comparison, the following graph shows how the SQBR and passer rating systems perform for pro-style quarterbacks.
The SQBR actually performs better at predicting the winner with a pro-style quarterback than the conventional passer rating system does. This isn’t particularly surprising given that the SQBR was designed to predict outcome while the passer rating system wasn’t, but I think it’s interesting nonetheless. All in all, the highest correlation between outcome and rating is for the SQBR for spread quarterbacks.
Finally, some numbers. The mean SQBR for spread quarterbacks was 36.93. For quarterbacks that scored more than one standard deviation above the mean, they won 97.2% of the time while winning only 18.1% of the time when scoring more than one standard deviation BELOW the mean. Discussion
So what does it all mean? It means the better your quarterback plays, the better chance you have of winning. This obviously isn’t groundbreaking, but I think it’s interesting to see how much bigger a difference it makes in this system. Keep in mind that all of this is calculated completely independent of defense. The extremes are very extreme here. Only one game was lost with a SQBR more than one standard deviation above the mean and that was a shootout loss by Pat White and WVU to Louisville at the peak of their Petrino era in 2006. On the other side, only 18% of the games were won with a SQBR more than one standard deviation below. Contrast that with the fact that in pro-style systems, teams were able to win 32% of the time with a poor SBQR and 48%(!) of the time with a passer rating more than one standard deviation below the mean. Close to twice as likely to win with a bad QB performance in the pro-style system.
Let’s see some numbers from the guys we know and love.
And finally, what I think most can agree is where we want to be.
The discrepancy between White’s last season with Rodriguez and where we are now is honestly mind boggling. In 23 games, we’ve seen a total of five above average games, and no games more than one standard deviation above the mean. The good performances basically line up with on field results. Forcier against WMU, ND, and Illinois*, Sheridan against Minnesota, and Threet against Notre Dame. White on the other hand had nine out of twelve games above average and eight out of twelve exceptional games. Unsurprisingly, WVU lost two of the three games where White’s numbers were below average.
What can our guys do better to emulate White? I think looking at these tables, it’s pretty obvious. One, be more of a running threat. White ran often and he ran well, our guys generally do neither. Two, limit the negatives. White only threw three INTs the entire season and was sacked less than once a game. Needless to say, our guys don’t exactly live up to this standard.
A big positive? Forcier is definitely doing better than last year’s guys. He is a true freshman, and I think we can expect a big jump from him next year (or else…). At the same time, I would also argue that Gardner should be given every possible chance to compete for the job next year. It’s clear to me that you need the best QB play possible for this to succeed and I think it’s only sensible to explore all of your options. With the current trends and feelings out there, I don’t think Rodriguez can worry about getting space between Forcier and Gardner, he needs to win as soon as humanly possible.
So basically, I guess I’m saying we need to continue to get better quarterback play if we’re going to succeed in this system. Looking at this, I’m starting to feel like Gardner is really the key to success, not Forcier. If Rodriguez needs to win next year to save his job, he should stick with Forcier. If he has the 4-5 years we think he does, I’d throw Gardner in there next year and get it rolling. Secondly, I really don’t think defense is all that important. Obviously the last two and a half games have made that statement sound kind of ridiculous, but I believe with good quarterback play in this system, the defense needs to achieve only competency.
Please let me know of any suggestions you can think of to make this better. Again, I’d be happy to give anyone the raw data that is a better statistician or is interested.
*I realized right at the end of this that I don’t have fumbles in the equation. For whatever reason, it wasn’t in the ESPN data I got. Crap… I think it would only be better though with fumbles included.