Florida had the benefit and curse of playing in the first game of the year in week 0. After Florida’s win over the Miami Hurricanes, Gators’ QB Feleipe Franks was widely criticized. Some of the criticism was valid, but a lot of the criticism was excessive and personal. With everyone else’s week one in the books (except the Labor Day Monday game), I decided to see just how poorly Franks’ performance compared to other QBs who faced power 5 opponents.
Taking the rankings from https://www.sports-reference.com/cfb/years/2019-passing.html I used this metric as a baseline for overall performance by the quarterback in week one. Franks ranked 44th overall.
Here are the top 50 performers:
As we know, however, a QBs performance is highly influenced by the opposing defense. To account for this variable, I utilized the 2018 team defensive rankings taken from https://www.sports-reference.com/cfb/years/2018-team-defense.html
Here are the top 20 teams from last year (Power 5):
The P5_Re-Rank is the new ranking once non-P5 teams were removed. These teams needed to be re-ranked. Since non-P5 teams play easier schedules, their defensive rankings are not an indicator of how good the defense actually is- and since the quality of the defense is a key measure in this analysis, I could not have this metric skewed.
Then, each QB who played in week one against a P5 opponent was extracted from the overall list. Their week one ranking was re-ranked from the overall group to just those that faced P5 opponents. It was clear upon running an initial analysis that QBs who faced non-P5 opponents did better than those who did.
It is easy to see that point in the chart above. The group on the horizontal (x) axis to the left were those who faced non-P5 opponents. They averaged a higher passer rating than QBs who did face P5 opponents. Ironically, a 5-star recruit playing against a non-P5 opponent had the worst day of all (Hunter Johnson of Northwestern).
I also wanted to see if the QBs recruiting ratings coming out of high school were predictive of success in week one. So, this metric was included as well. Now, the variables were set. I was looking to see how well each QB did in week one while controlling for the quality of defense faced. I just threw the recruit-rating in to see if it had any predictive power for how the player performed.
There were two important assumptions made: Power 5 defenses are generally better than non-P5 defenses, and that last year’s defensive rankings are indicative of this year’s defensive strength. Of course, there is fluctuation, but this assumption is necessary to quantify the level of opposition QBs faced in week one. On to the findings…
The first look was to see if opponent defensive strength (stored as variable P5_order) was correlated with QB performance. In the subsequent regression analysis, it was statistically significant (p = 0.04).
As the chart shows us, the higher (worse) a defense was ranked in 2018, the better the QBs performance against them was. Ok, great. Though not a perfect correlation (at all), it was still strong enough and statistically supported to apply it to the analysis.
Recruit Rating (RR)
As stated, this was more of a curiosity. What I found was that the rating of the QB (composite) was statistically significant in its correlation to performance against P5 defenses in week one (p = 0.04).
As this scatterplot shows, the higher a QB was rated, the more likely he was to have success in week one. That little dot in the lower right-hand corner is the aforementioned Hunter Johnson. Of note, if a player was unrated, I assigned a .7900 rating and 2 stars. That is why you see the line of players to the left.
After all was said and done, Franks ranked number one overall when controlling for the strength of opposing defense faced in week one. When adjusting for P5 opponents, Franks had the 8th best performance while facing the 13th best defense. I simply added these two ranks scores together, giving a score of 21 points (fewer points are better because the lower something is ranked, the better it is). Here are the final standings:
|Feleipe Franks||Florida||Miami (FL)||13||8||21||1|
|Jarren Williams||Miami (FL)||Florida||14||11||25||2|
|Levi Lewis||Louisiana||Mississippi State||2||27||29||3|
|Zach Smith||Tulsa||Michigan State||5||30||35||6|
|Sam Howell||North Carolina||South Carolina||38||3||41||7|
|Ryan Willis||Virginia Tech||Boston College||32||12||44||9|
|J’mar Smith||Louisiana Tech||Texas||33||16||49||12|
|Colin Hill||Colorado State||Colorado||40||10||50||14|
|Anthony Brown||Boston College||Virginia Tech||48||4||52||16|
|Woody Barrett||Kent State||Arizona State||29||25||54||17|
|Chris Robison||Florida Atlantic||Ohio State||31||26||57||18|
|Josh Adkins||New Mexico State||Washington State||25||32||57||18|
|Gresch Jensen||Texas State||Texas A&M||26||31||57||18|
|Drew Plitt||Ball State||Indiana||45||15||60||21|
|Jorge Reyna||Fresno State||USC||37||28||65||27|
|Spencer Sanders||Oklahoma State||Oregon State||63||2||65||27|
|Dan Ellington||Georgia State||Tennessee||43||23||66||29|
|Jordan Love||Utah State||Wake Forest||57||9||66||29|
|Tyler Vitt||Texas State||Texas A&M||26||41||67||31|
|Hank Bachmeier||Boise State||Florida State||52||18||70||34|
|Jake Luton||Oregon State||Oklahoma State||54||17||71||36|
|Cephus Johnson||South Alabama||Nebraska||50||24||74||37|
|Brady White||Memphis||Ole Miss||61||29||90||40|
|Jake Bentley||South Carolina||North Carolina||59||35||94||41|
This analysis does not prove or even claim that Franks is better than any other QB, or that we can draw conclusions based upon one game. The aim was to investigate the validity of the pervailing narrative regarding Franks’ alleged “horrendous” opening performance. The findings of this analysis strongly contradict that narrative and, at a minimum, offer some context regarding drawing hard and fast conclusions based on small sample sizes.
Each player was assigned a random ID number and the analysis was initially conducted using only ID numbers in order to avoid potential biasing of the outcome. Correlational analyses were conducted to validate the inclusion of the predictor variables but did not influence ranking or scoring. The intention is to conduct this analysis weekly. As such, going forward raw sums of rankings will not be used. Standardized scores will be used to ensure equal weighting of the variables.