I have always been fascinated by the confluence of math and sports. Over the past several years, I have developed and refined a set of mathematical tools that I use to help me predict and better understand the worlds of college football and basketball. With the 2024 college football season just days away, it is time to once again open up the analytical tool box and see what we can learn.
Before we delve into the upcoming season for the Michigan State Spartans and beyond, reviewing the results of the past season is a valuable exercise.
General Lessons from the Past
In the summer months leading up to the college football season, fans and sports analysts have limited information, and yet predictions abound regarding rankings, win totals, conference champions, and playoff participants. But how accurate are these predictions, historically?
I have written extensively on this topic elsewhere. The simple answer is that on average, the experts do a good job, but there is also a significant amount of variance in their predictions. More quantitatively, I have derived these two rules of thumb:
- The preseason rankings are accurate to plus or minus 25 positions (one standard deviation)
- The preseason win total predictions (i.e. over/under lines) are accurate to around 2.1 games (one standard deviation)
This information is helpful when looking at the preseason rankings and win predictions totals in July or August. Both data sets resemble a bell curve. As such, approximately one-third of the teams will finish the season more than 25 position above (or below) the preseason ranking and more than two games above (or below) of the preseason prediction.
But math also allows us to learn a bit more about the contributing factors that result in this deviation. The difference in ranking position is only semi-quantitative. But it is possible to dig deeper into the deviation in the win totals.
Once again, I have explored this topic in significant depth previously. The short version is that the difference in the number of wins between the final tally and the preseason expected win total is due to three factors: changes in the strength of each team's schedule, luck, and the expert's ability to accurately predict the strength of each team.
These three factors do not contribute equally to the final difference between prediction and reality. Here is another rule of thumb:
- Changes in schedule difficulty contribute plus-or-minus half of a game (one standard deviation)
- Luck (good or bad) contributes a little over one game (1.1 game)
- Differences in the perceived strength of a team in the preseason contributes plus or minus 1.75 games, on average.
While some teams are more lucky than others and while some teams wind up playing an easier or hard schedule than expected, the most significant impact on the accuracy of preseason predictions is simply that the experts do not always have a firm grasp on how good or bad each team will be.
2023 Preseason Polls in Retrospect
With this information in hand, Table 1 below shows the teams listed in the preseason top 25 (and beyond) by expert football prognosticator Phil Steele along with each teams post-season AP rank, final record, Preseason expected win totals (based on my calculations last summer) and the impact of changes in ability, schedule, and luck on each team's final regular season win total.
 |
| Table 1: 2023 Preseason Phil rankings compared to the year-end results |
The first five columns of are shaded depending on whether each team overachieved (shades of green), underachieved (red), or performed as predicted (white). The rightmost four columns are shaded based on the range of values in each column from high (green) to low (red).
The table is sorted based on the preseason rankings, similar to the information available to college football fans in this off season.
As a example, Michigan State started the 2023 season ranked No. 48 by Phil Steele. My detailed mathematical analysis of the full college football season and the Spartans' schedule resulted in an regular season expected win total of 5.33, which correlated strongly with the preseason win total over/under of 5.5 wins.
As Spartan fans know well, Michigan State finished the season at 4-8 or 1.33 games below expectation. What Table 1 reveals is that based on the Spartans' overall performance, things actually could have been a bit worse.
Based on my analysis, Michigan State's actual season-averaged ability should have resulted in the Spartans winning 2.58 games fewer than expected. By this measure, the Spartans should have finished with a record of just 3-9. Michigan State benefited from a little luck (+0.86 wins) and a slightly easier than expected schedule (+0.39) in order to get that fourth win.
Overachievers and underachievers
The data in Table 1 can be used to understand the results from the 2023 season in more detail.
Of the teams that started the season ranked in the top 10, only half of them lived up to expectations (more or less): Georgia, Michigan, Alabama, Texas and Florida State. Meanwhile, Ohio State, LSU, and Penn State finished just outside of the top 10 and thus had mildly disappointing seasons. The same can be said for Notre Dame and Tennessee which both started the season just outside of the top 10.
The notable underachievers in the preseason top 15 were USC, Clemson, and Utah. Clemson finished the season with four losses and ranked No. 20 in the final AP poll while the Trojans and Utes both finished the season unranked and with five loses.
In contrast, there are several teams that opened the season with lower expectations, but which ended the season ranked in the top 11. Washington, Oregon, and Mississippi began the season ranked in the teens and ended in the top 10. Even more impressive was Missouri and Arizona who both finished the season ranked in the top 11 despite starting out ranked No. 44 and No. 65 by Phil Steele, respectively.
Other lower ranked underachievers included No. 17 Wisconsin, No. 18 Texas A&M, No. 20 TCU, and No. 25 Texas Tech, all of which finished unranked with at least six losses. Those teams were replaced in the final AP top 25 by overachievers such as No. 45 Oklahoma State, No. 46 Louisville, No. 52 Kansas, and No. 54 SMU.
Digging Deeper Into the Data
Looking at the polls both before and after the season will reveal the overachievers and underachievers. But the data on the right side of Table 1 explains why the teams faired the way that they did.
For most of the teams in the preseason top 10, there was not a big change in the actual ability from the beginning to the end of the season. Most of those teams were roughly as good as expected. But what separated the very good teams from just the good ones was luck. Georgia, Michigan, and Alabama all graded out close to +1.0 games in luck. Texas finished with +0.74 games of luck, combined with a bump in both ability (+0.27) and schedule (+0.61).
The fact that the very good teams are also "lucky" makes sense. As Table 1 shows, no team started the season with an expected win total over 11 games. In order for a team like Georgia or Michigan to finish the regular season undefeated, it is necessary to win a few toss-up games. That is what separates the elite teams from just the very good ones. While "luck" is a generally accepted term in sports for winning more games than expected, it can just as easily be called "grit" or "execution."
Washington and Florida State followed a slightly different trajectory. My analysis suggests that Washington was not significantly better in ability that the preseason No. 14 ranking. Florida State was actually almost a full game worse than expected in raw ability.
Those teams were able to finish ranked in the top six at the end of the season due to an extraordinary amount of luck (+2.34 for Washington and +1.92 for Florida State). The Seminoles were also aided by a significantly easier than expected schedule (+1.53). To a lessor extent Mississippi (+1.50 in schedule and +1.03 in luck) followed a similar path.
For other 2023 preseason top 10 members, they simply were not as good as expected. Ohio State (-0.79 games in ability), LSU (-1.89), USC (-2.75) and Clemson (-3.19) all fall into this category. Farther down the standings, Tennessee (-1.61), Utah (-2.95), Wisconsin (-3.62), North Carolina (-2.14), and Florida (-2.31) were all also simply not as good as predicted.
A few other teams underachieved slightly just due to a bit a of bad luck. Penn State (-0.40 games of luck), Notre Dame (-1.18), and Oregon (-0.17) fall into this category. A few other unlucky teams of note include Texas A&M (-1.25), TCU (-1.71), and Kansas State (-1.71).
In these three cases, Texas A&M and TCU were also both significantly below expectation in ability which resulted the Lone Star State teams both finishing the season unranked. Kansas State, however, was actually better than expected on the ability front and thus the Wildcats finished the reason near where they started despite the poor luck.
The team which appears to be the opposite of Kansas State is Iowa. Both teams entered the 2023 season and ended the season towards the bottom section of the top 25. In Iowa's case, the Hawkeyes were significantly below expectation in ability (-1.95) but they made up for it with extremely good luck (+2.85), a value which was No. 1 in the nation.
The two other notable power five which benefited the most from luck in 2023 were Oklahoma State (+2.47) and North Carolina State (+1.49) (in addition to Washington and Florida State, both of which were mentioned earlier.) The Cowboys and the Wolf Pack were both expected to win around seven regular season games in the preseason and both squads finished the regular season at 9-3 and ranked in the top 25 despite neither team grading out significantly better in ability at the end of the season.
The teams at the bottom of Table 1 all started the season ranked outside of the top 40 by Phil Steele but in most cases finished the season ranked. In order to overachieve in this manner, each team needed to excel in at least two of the three metrics discussed so far.
The four power five teams which overachieved the most in 2023 were Missouri (+4.65 wins over preseason expectation), Arizona (+4.04), West Virginia (+3.76), and Louisville (+3.14). All four of those teams grades out to have a much higher ability that expected. Missouri, Arizona, and to a lessor extent Louisville all also benefited from an easier than expected schedule. Louisville had the most luck of the four teams and Arizona excelled despite being slight underwater in the luck category.
Looking Ahead to 2024
With this information in mind, it may be possible to make some initial predictions as to which teams might be overrated or underrated in 2024. Teams which benefited or suffered due to a significant amount of luck in 2023 may be more likely to regress to the mean in 2024.
I will have my eye on No. 3 Oregon, No. 7 Notre Dame, No. 9 Penn State, No. 11 Missouri all of whom were a little better than their final records last year indicated, based on my analysis. No. 17 Kansas State, No. 20 Texas A&M, and No. 21 Arizona are also teams to watch.
On the other side of the coin, I believe that there is reason to be skeptical about the potential of teams such as No. 6 Mississippi, No. 12 LSU, No. 13 Utah, No. 14 Clemson, No. 18 Oklahoma State, No. 22 USC, No. 23 North Carolina State, and No. 25 Iowa.
With this introduction in place, it is time to turn our focus to the Big Ten and specifically onto the Michigan State Spartans. Next time we will take a close look at the Spartans' schedule. Stay tuned.
Tweets
It is time to kickoff Dr. Green and White's (@PaulFanson) annual math-driven college football preview series. In Part One, we take a look back at the 2023 season to understand what the experts got right, what they got wrong, and most importantly, why.
https://michiganstate.rivals.com/news/dr-green-and-white-preseason-football-analysis-2023-retrospective
Comments
Post a Comment