Over the past month and six articles, we have examined the details of MSU and college football recruiting from about as many different angles as I could imagine. We surveyed the stars, considered the classes, analyzed booms and busts, and scrutinized schools, states, and position groups. It has been a fun ride.
But, there is one topic that I have barely mentioned, and that is the simplest and most important metric in all of college football, or any sport for that matter: wins. We can talk about stars, All-Conference teams, and the NFL, but at the end of the day, it’s “just win, baby.” So, the final question that I wold like to ask is how does the strength of a team or roster, on paper, correlate to actual on-field success?
In all of previous pieces in this study, I have focused on individual classes, somewhat in a vacuum. However, a college football team is built based on a series of several consecutive classes. Any given team is likely composed of players from typically five different recruiting classes, from the current year (true freshmen) back to five years prior (red-shirt seniors).
I already have a way to quantify the relative “quality” of each class using my NFL Draft Potential metric that I introduced in Part One and have used throughout this series. The challenge now is to figure out how to weigh the contribution of each class to create an averaged “quality” of a given roster.
In order to accomplish this, I tabulated the number of wins earned each year for all Power Five schools from the 2007 to 2019 seasons. I then calculated the total NFL Draft Potential for each class and team for the full range of data in my database, from 2004 to 2019. I started by taking a simple average of the five classes that in theory would make up the roster of the team and year in question.
For example, MSU’s 2019 team (which won 7 games) was made up of players from the 2015, 2016, 2017, 2018, and 2019 classes. If I take the straight average NFL Draft Potential score for the those five classes, I get a value of 2.53, which has units of projected NFL draft picks per class.
I performed the same calculation for last 13 seasons for all 65 Power Five schools. The data is best handled by dividing it into bins of average NFL Draft Potential per class and then taking the average number of wins for all the teams/years in that bin. But, this initial pass assumes that each class contributes equally with an effective contribution of 20 percent. As a final refinement, I adjusted the weighting factor for each class in order to improve the overall correlation. The final result of this series of calculations is shown below in Figure 1.
I must admit that the first time I saw this graph I was absolutely shocked at how strong the correlation was. There is a very clear and direct relationship between the strength of recruiting and the number of wins. I have also noted the position on the graph of the four recruiting tiers that I identified in Part Two.
For reference, each data point in Figure 1 represents a bin of roughly 30 teams from different years that share a similar weighted average roster NFL Draft Potential. In addition, the optimized set of weighting factors by class are shown below in Table 1. As we can see, the senior (or red-shirt junior) class contributes the most weight to the average, and that significance decreases only slightly (less than two percent) down to the true freshman class. The red-shirt senior class has the smallest overall impact, which does make sense as a fair percentage of these players likely already left the program.
Getting back to Figure 1, based on the slope of the line, every additional one player of NFL Draft Potential per class is worth an average of about one win per year. More specifically, Tier Four recruiting teams, which average 1.0 to 2.0 of NFL Draft Potential per class, on average win six games a year. Tier Three teams usually have about one more potential draft pick per class and generally win an average of seven games.
Tier Two teams with 3.0 to 4.0 of NFL Draft Potential per class win an average of eight games. Finally, we have the Tier One teams, which as was pointed out in Part Two, are actually separated into sub tiers of teams that average between 4.0 and 5.0 of NFL Draft Potential per class (and win an average of nine games a year) and a small, super elite group that averages a little over six players of NFL Draft Potential per class and that wins an average of 11 games a year.
When it comes to teams like MSU, which recruit at a Tier Three level (with an average NFL Draft Potential score of 2.34 since 2007), this essentially means that the Spartans should expect to win about seven games a year with this level of recruiting. For a team like Michigan, who recruits at a Tier Two level (with an average NFL Draft Potential score of 3.75 since 2007) that corresponds to about a 8.5 wins per year. For Ohio State (with an average NFL Draft Potential score of 4.77 since 2007), the expected number of wins jumps up to almost 10 (9.7).
While the average data paint a clear picture, it is also quite important to consider the variance of the data in each bin. In Figure 2, I present the same data shown in Figure 1, only this time I am including error bars that represent one standard deviation from the average.
As we can see, while the correlation is very good, the standard deviation at each data point is also very large at about two-and-half games. To this point I have two basic thoughts.
First, the mere fact that the correlation is so good in the first place is amazing. The estimation of roster strength is based solely on the list of committed recruits on signing day for each team and their projected eventual NFL draft rate. A lot can and does change with a recruiting class as players transfer, get drafted early, get injured, or leave the university. The numbers used as input do not reflect the actual situation on any of the team’s rosters. It is a projection of a projection is many ways. A high level of variance is to be expected. Yet, on average, the trend is very strong and on some level, it can be predictive.
Second, I believe that this also speaks to the general level of uncertainty in college football in the first place. Coach Dantonio used to say that it is a game of inches, and in many ways that is true. Games are often won or lost based on a single play or bounce of an oblong ball. The exact same roster of players might win five game or eight games due to a bit a good or bad luck here or there.
Wins are, of course, very nice, but what about making the playoffs or winning championships? A quick analysis of the roster strength of the teams that either made the college football playoff or the BCS championship game (prior to 2014) shows that they averaged an NFL Draft Potential Average score of 4.46 players (plus or minus 1.1). That is solidly in the Tier 1/Top 10 recruiting territory, which makes sense. As for the eventual champions, they averaged an NFL Draft Potential Average score of 4.91 players (plus or minus 0.9). In order to reach that score, a team essentially needs to average a Top-five class for four to five consecutive seasons.
That said, there are deviations from these averages. How much hope does a Tier Two or Tier Three recruiting team have to reach the Playoffs or win a Championship? Based on the data, we can estimate those rates as well using the historical data. That correlation is shown below in Figure 3.
As for the odds to make the College Football Playoffs, even teams in the Tier Three category at least have a one percent shot. Notably, MSU’s 2015 team is the team with the lowest average roster NFL Draft Potential of 2.29. The other Playoff/BCS teams in this group are Washington in 2016 (2.47) and Oregon in 2010 (2.77). No Tier Three level team has won a National Title in the past 13 years.
For teams with Tier Two level recruiting, the odds vary from about five to 10 percent to make the Playoffs or BCS title game. The most recent Playoff teams with an average roster score below 4.0 include Clemson’s 2017 team (3.92), Notre Dame’s 2018 team (3.81), and Oklahoma’s 2017 team (3.79). Only Clemson’s 2016 team (3.80) won the National Title with an average roster NFL Draft Potential of under 4.0.
For the Tier One level recruiting schools, the odds to make the Playoffs steadily increase from 16 percent at the lower end up to 26 percent at the high end as the average roster score increases from 4.0 up to 6.0. As for the National Title odds, just being in this range seems to satisfy a threshold where the odds are a consistent seven to eleven percent.
Finally, on the far right end of the plot are a group of teams with super-elite rosters with scores over 6.0. Amazingly, these five teams are exactly Alabama’s last five teams, four of which made the playoffs and two of which won the National Title. While this is certainly somewhat of a special case, that is the data that we have. Alabama is in a class by themselves when it comes to recruiting. But even with all that talent they don’t win every year.
Putting All The Pieces Together
Now that we have a clear picture of the both the average value of stars and the average number of wins that they produce, it is time to assemble the final bit of data for this project. In the previous six parts of this study, we looked at teams, state, and position groups that were above or below average based on the calculated relationship between recruiting rating and NFL draft rates. The same basic idea can be applied to the data in Figure 1 correlating NFL draft rates to wins.
Essentially, the line shown in Figure 1 provides a projected number of wins that a team with a given roster can expect. As we did with NFL draft projections, we can also compare the number of actual wins for each team and year to the value predicted by the average in order to measure if a team over- or underachieved in the win column relative to the rest of the Power Five.
I summed up the total difference between actual wins and projected wins for all Power Five team from the 2007 to 2019 seasons. I then compared that to the difference in actual and projected NFL Draft picks for the 2007 to 2015 recruiting classes. The result is shown below in Figure 4.
Throughout this project, there has always been a question about the role of both coaching and evaluation when it comes to finding and developing talent. While there is no simple or clear answer to this question, I believe the data in Figure 4 provides some insight.
As I have done several times before in this project, I divided the data above into its four quadrants, and each quadrant tells a slightly different story about the schools who reside there. Let’s start in the upper left-hand corner, which includes teams such as Oklahoma State, Virginia Tech, and Northwestern. In this part of the graph, the teams are underperforming when it comes to putting players into the draft, but overperforming when it comes to wins.
While this is a complicated issue, the simplest explanation is that the school is a bit weak in the area of talent evaluation and/or development of high school prospects. NFL teams don’t care about wins in the draft. They care about raw talent. However, these schools are winning anyway. This implies that despite a slight deficit in talent (either raw or developed), the coaches are able to coach the teams to victory. They are making the most of the talent that they do have.
The opposite situation is true in the lower-right hand corner of the graph, which includes teams such as Florida, Miami, Notre Dame, Michigan, Arkansas, UCLA, and Illinois. That list sure looks like a group that underachieves on the field and the math tends to agree. The NFL draft data suggests that these teams are above average at recognizing and/or developing talent, but the coaching needed to actually win games is lacking, on average, over the past 13 years.
The upper right-hand corner is the best of both worlds. These teams have a skill for finding and developing high school talent AND they are able to win games with that talent at an above average rate. This quadrant contains teams that have generally seen a lot of success over the last decade or so, including Ohio State, Clemson, Alabama, Wisconsin, Penn State, Iowa, Stanford, Oregon, Oklahoma, Georgia, LSU, and Michigan State.
In the lower-left hand corner of the graph are the have-nots. These teams are below average both in finding talent and in coaching that talent once it arrives. This quadrant features such underachievers as Texas, Tennessee, Ole Miss, Auburn, Colorado, and USC as well as several Big Ten teams including Minnesota, Rutgers, Indiana, and Purdue. As for the worst coached team in the Power Five over the past decade, that “honor” goes to Kansas.
As far as just raw coaching ability, Wisconsin seems to be in class by themselves. Wisconsin recruits in the Tier Three zone in general and they overachieve in the NFL draft rather modestly (+4.78). Yet, over the 13 years of this study, the Badgers have won over 40 more games than expected, based on their projected roster strength. That is over three wins per year.
The next tier of teams are generally plus-30 in wins over this span and that includes TCU, Utah, Oregon, and Clemson. Clemson is actually not quite as big of a recruiting juggernaut as I expected. It seems that the main secret to their recent success is more due to outstanding coaching. As for Michigan State, the Spartans lie in the next tier down with the likes of Stanford, Ohio State, Oklahoma, and Oklahoma State. The data in this chart essentially covers the entire Mark Dantonio era. Time will tell if Mel Tucker’s staff will have the same level of success.
Looking Towards 2020
Up until now, this analysis has focused only on the past, but before I sign off I though that I would make an attempt to peer into the future. The mathematical tools and correlations that I have developed and explored this month have helped us to understand the past, but we would hope that the same methods can tell us something about the future.
In the previous sections, I estimated the strength of any given team’s roster based on the previous five recruiting classes. Naturally, it is trivial to perform the same calculations on each team’s roster in 2020 and to project the expected number of wins for each Power Five team. This results is a projected preseason Top 25, which I show below (along with the remainder of the Big Ten) in Table 2.
This table comes with the same caveats as above in this it does not consider the actual roster, each team’s actual schedule, or whether that team tends to over or underachieve in recruiting. It is also implicit that we can expect over two games of variance for the average number of wins shown above. The teams with good coaching will likely perform better, while the teams without will likely do worse. It certainly provide an interesting starting point for perhaps future analysis.
This data suggests that both Ohio Sate and Michigan have good enough rosters to be Top 10 teams. Penn State is Top 15 and Nebraska is Top 20. As for MSU, the Spartan’s rank No. 31, just three spots below Maryland. Before any Michigan fans out there get too excited, I would invite them to review Part Three and the high bust rates that have already occurred in their 2016 and 2017 classes.
Based on the recent historical data, Georgia and Alabama would project to have the best odds to both make the playoffs and win the National Title. The Buckeye’s odds to at least make the playoffs, based on a comparison to Figure 3, are better than the pack below them at about 25 percent. The remainder of the Top 11 in Table 2 would also project to have National Title Odds of about 10 percent. Hopefully we will get a full 2020 season and we can one day compare these predictions to reality.
The time has come to bring this project to a close. After seven posts, 74 Figures and Tables and approximately 25,000 words, we have covered a lot of topics. But, the overall purpose was really understand what we can actually expect when looking through a team’s commitment list on signing day.
As a first pass, we can estimate how many of the players on the list will likely get drafted into the NFL and we can make a fair and quantitative comparison of the class to classes in other years and on other teams in the nation and conference. We can then dig deeper into the home state, position, and team that the player committed to and then adjust our expectations a bit based on these parameters.
Based on today’s investigation, we can also estimate how many wins that class should be able to accumulate. For a team like MSU, the data above is a bit sobering. In order to really compete for a National Title, MSU will likely need to improve its recruiting to at least a high Tier Two level. Recent history has shown that it is nearly impossible to win at the highest level without consistently bringing in Top 10 classes.
That said, MSU was maybe just a bad pass interference call or two in 2013 away from getting a shot with a roster score of only 2.32. It turns out that team had both underrated talent and coaching and managed to stay very healthy all year. Can MSU do something similar again under Coach Tucker? That will most likely not be easy and probably isn’t likely. But, probability is not destiny, and as always, I remain optimistic.
That is all I have for now. I am sure that I will soon revisit this data set and methodology to see if I can extract any additional nuggets of data and information. Thank you to those with the fortitude to make it to the end of this current string of sports analytics ramblings. Until next time, enjoy, and Go Green.