Today's interviewee is Loren Maxwell, founder of the Georgia High School Football Historians Association and creator of the Maxwell Ratings, which were published for the first time for 2019 in GHSF Daily and on ajc.com this week.
Loren Maxwell, creator of the Maxwell Ratings
1. For those new to your rankings, or those who might need a refresher, on what do you based your preseason rankings, and what do they mean? "The goal of the model is to provide the best mathematical explanation for the season by assigning each team a rating that best correlates to their performance. Although there are many possible approaches, the one I use is based on logistic regression, which is a fairly common statistical technique. The data the model takes into account are the scores, the locations and the order of the games - I weight more recent games slightly heavier - along with each team's historical performance. As the season progresses, the past [previous season's] ratings count less until they are relatively insignificant. The ratings themselves represent the point differential between teams so that you could predict a margin between two teams by comparing their ratings." [For example, using today's Class AAAAAA ratings, Tucker with a 56.85 rating would be roughly a six-point favorite on a neutral field against Stephenson, which has a 50.91 rating.]
2. Some ratings don't match what many fans might expect for the 2019 season. For example, Marietta - with several major Division I senior prospects on its roster - is nationally ranked in some preseason polls yet only the 18th-best team in Class AAAAAAA, according to your rankings. What do you tell those disbelieving Blue Devils fans? "The model doesn't give any consideration to personnel. Although Marietta might have plenty of D-I talent, there is no on-field performance for this season yet, so their rating is based solely on their historical performance. This is similar to teams such as Colquitt County and Grayson, which are highly rated because of their recent performance although both teams have new head coaches. A similar situation happened last season with Marietta, but they didn't have the break-out season they were hoping for and instead performed close to their historical average. In that case, the model turned out to be correct, but if Marietta has that break-out performance this year, the good news is that their ultimate rating will reflect their on-field performance."
3. Are some teams' strength-of-schedule underrated because of out-of-state competition? How do you deal with that? "Unfortunately I don't track out-of-state teams, so while those games are considered in the model, their impact is limited because the model isn't aware of how strong those teams are inside their own state. As an example, if a team from Georgia beat an out-of-state team by 14 points, then the model would simply rate that out-of-state team as 14 points lower than the Georgia team they played because that's all the model would be aware of. However, the rating of out-of-state teams is not considered in the SOS rating."
4. Your model not only ranks teams, but also regions and classifications. What insights do your ratings provide in terms of the strength of each classification? Are there some classes that are almost even, and others where there's quite a gap? Is there evidence of too many classes?"Currently there is little difference in the ratings between AAAAAA (42.64), AAAAA (40.20) and AAAA (41.77), and then AAA (29.42) and AA (28.92). That might indicate four classifications is really the proper number for Georgia. However, I'd argue that there could be the same number of classifications, but the enrollment cutoffs on the lower end should be different. Because only 11 players can be on the field at a time, there's actually less difference between a 2,000-student school and a 1,000-student school than there is between a 500-student school and a 200-student school."
Produced by Georgia High School Football Daily, a free e-mail newsletter. To join the mailing list, click here.
