From High School Stars to the NFL, a Football Recruiting Analysis

For those that follow my posts, it should be clear by now that I have a fascination with numbers and that this fascination takes many forms. One topic that I have been thinking about a lot lately is that of college football recruiting. This time of year various fan-bases either celebrate or mourn the fact that a 17-yr old kid just picked up a baseball hat with a specific color scheme off from a table.  Because we have no actual football to focus our emotions on, the successes or failures on the summer recruiting trail can be very intense.

But, does it actually matter? How much better is a 5-star compared to a 3-star? Do some schools do a better job than others with lower ranked recruits? Should we be freaking out if our rival just picked up a 5-star? I have often thought about ways to try to quantify these things. I think that I finally figured out how.

First, I needed to find a consistent measure of both the quality of high school recruits and the eventual success of those recruits at the next level. For the former, recruiting rankings are a reasonable place to start.  Sites like Rivals will cluster recruits into groups based on their apparent talent.  As I explained in my previous post, I prefer the "Rivals Rating" system that divides the high school players into ~10 bins.  The table below shows how these ratings compare to "stars" and how many kids are typically in each bin.

For the measure of success at the college level, one could propose a wide variety of options. The measure that I decided to use is simply the NFL draft. My reasoning is that while this is certainly not a perfect metric, it is a reasonable proxy for college success, and more importantly, the data set is large, consistent, and easy to mine.  

The Rivals database has the same properties, although there are limitations there as well.  For example, it is only a snapshot of a team's recruiting on signing day of each year. If a team picks up a transfer, that is not going to be reflected in the database.  From a consistency point of view, it is not practical to manually update each team's roster based on these transfers, and the transfer market is a fundamentally different animal than the evaluation of high school students anyway.  So, I decided the it is best to use the Rivals database as is.

The first piece of this puzzle was to try to understand the statistics from the perspective of an entire recruiting class (or classes) as it maps to the NFL draft.  So, as a start, I first set up a database of all 1,016 players selected in the last 4 NFL drafts. I then setup a second database of the roughly 20,000 players in the Rivals database from the 2013 to 2017 classes.  I have plans to expand both databases back to at least 2007, but I can already draw several interesting conclusions from this set of data.  

Let's first take a look at things from the view point of the NFL draft, starting with the recently completed 2019 draft.  I tabulated the distribution of draft picks as a function of the Rivals rating. That chart is shown here:

This may come as a shock to some, but the NFL draft is actually not composed entirely of 5-stars.  It actually can't be, because as the table above shows, there are usually only ~35 five-stars in a given year, while there are over 250 total picks in the NFL draft.  Rather surprisingly, the distribution of picks is actually pretty even across all rating down to the 5.3 bin, where the numbers drop off.  It is also notable that the 2nd largest category is actually the walk-on category, which is only beaten out by the 5.8 bin (low 4-star).

But, maybe all the 5-stars get taken in the first round?  Well, that isn't true either.  The plot below shows the average Rivals rating as a function of NFL Draft round.  While the trend is that the average rating decreases in the later rounds, even the first round average (5.73) would be a terrible average class for Ohio State, and a bad one for Michigan.  MSU's average over the past 13 years is 5.63, which is better than all but two of the rounds.  The overall 2019 NFL draft average Rivals rating was 5.58, which is basically equivalent to a Maryland or Wisconsin recruiting class.

While the analysis from the point of view of the draft is interesting, it is more interesting to look at the data from the point of view of an entire recruiting class.  Many of the players in the 2015 class and beyond are still in college.  But, the vast majority of the 2013 and 2014 classes have completed their eligibility.  How did these recruits eventually fare in the NFL draft?  That data is shown below:

Now we are getting somewhere. The main problem with looking at the data from the point of view of the draft alone is that the different Rivals rating categories have drastically different sizes. So, you really need to look at the data in terms of percentages, and you need an entire recruiting class worth of data to see the full picture.

The first thing that jumps out to me from this plot is that stars do, in fact, matter.  5-stars are much more likely to get drafted. In fact, a 5-star recruit (6.1) is about twice as likely to get drafted than even a high 4-star (6.0).  The odds then trail off gradually from there.  But, even though the odds of a 5-star getting drafted are high, they are only at 60%. That means that 40% of the 5-stars do not get drafted. In other words, a 5-star is hardly a "can't miss" prospect. 

Another interesting and subtle point in this figure is the difference in draft percentages between players from Power 5 schools and Group of Five schools.  The first point is that very few 4- and 5-star players sign with Power 5 schools.  Interestingly, those that do have a very low draft rate.  In this data set, the gray bar in the 5.8 bin is actually only one player out of the 18 four- and five-star that signed with Group of Five schools. This may suggest that this cluster of players had some issues (injuries? character issues? something else?) that prevented them from signing with a Power 5 team in first place.

For the players with a rating from 5.7 to 5.5, the NFL Draft rate is basically the same for both Power 5 and Group of 5 teams.  However, at a rating below 5.5, something interesting happens.  The NFL draft rate continues to fall for the Group of 5 teams.  But, for the Power 5 teams, it stays steady, right around 5% for all bins.  There is actually no significant difference in the draft rate between a 5.5, 5.4, 5.3, or 5.2 rating player if they sign with Power 5 team.  This seems like a clear indication that those Power 5 coaches actually know what they are doing.  It seems that they may be able to spot an under-the-radar player better than a recruiting analyst can.  

Now that we have a good set of reference data for the draft rate overall for Power 5 schools, it should be possible to compare the NFL draft rate of individual teams to this average. I do not yet have a good enough data set for most college teams. But, I do have data for all 14 Big 10 schools.  In this case, I have the Rivals Ratings data back to 2007, when Rivals started using the rating system in a consistent way. It doesn't make sense to include the data all the way up to the current recruiting classes, since those kids are still in school.  But, as stated above, the 2014 class and earlier has generally all graduated, so for this analysis I used the 8 full classes between 2007 and 2014, which is about 150-200 student athletes per team.

The first set of data to show here is a comparison of the NFL draft rates for this time frame for MSU, Ohio State, and Michigan as a function of the Rivals Ratings. That data is shown below.  In order to put things into perspective, I labeled each bar with the total number of recruits each team has in that bin over that 8-year span.

This figure has A LOT of information in it. Starting on the left side, MSU actually does above average when it comes to putting elite high school talent (ratings of 5.9 to 6.1) into the NFL draft, but on noticeably lower volume than the other two schools (only four players total). Ohio State actually is slight below average in all three of these bins, while Michigan is WAY below average.  In this time frame, Michigan only put two of their seven 6.1-rated 5-stars into the draft and ZERO of their seven high 4-stars (6.0).  The Wolverines are also a bit below average in the 5.9 rating bin as well.

The story changes a bit for the low 4-star / 3-star range (5.8 to 5.6).  In this range, MSU is a bit below average for the 5.8 bin and just average in the other two.  In contrast, Michigan is actually a little above average in this range, while Ohio State is WAY above average for all three bins, even extending down to the low 3-star (5.5) bin.  

Then, there is MSU's performance in the low 3-star / high 2-star range (5.4 and 5.5).  Anecdotally, MSU is famous for finding players like Darqueze Dennard (5.4), Le'Veon Bell (5.4), and Kirk Cousins (5.5) and turning them into NFL talent.  But, I don't think anyone has ever really been able to quantify how impressive this is.  MSU is clearly WAY above average in putting players in these bins into the NFL, and the total number of players is large enough that it is no fluke. (You can actually prove that mathematically).  This is clearly a skill of the current coaching staff. They take low 3-star / high 2-star recruits and send them to the NFL at a rate similar to that of 4-star player from an average Power 5 program.  There is no reason to believe that this will suddenly stop happening.

With this analysis in hand, it occurred to me that there was one more way to crunch the numbers to estimate the overall success that a team has in sending players to the NFL. Since I know the average rate that a Power 5 team develops recruits into NFL players as a function of their Rivals Rating, I can easily calculate the expected value of NFL players that any program should produce.  

As a simple example, OSU signed twelve 5-stars (6.1) during this time frame.  The average NFL draft rate of 6.1 recruits is 60%.  So, OSU should have sent 7.1 players (60% x 12) of those players to the NFL draft. In reality, the Buckeyes only sent 6, so they are 1.1 recruits short of expectations in the 5-star category (-1.1). I can perform the same calculation for the entire class and come up with a total deviation in the expected number of NFL draftees to the actual number for all 14 Big Ten schools from 2007 to 2014.  That data is shown below:

From this chart, it is pretty easy to see the teams that do an above and below average job of sending players to the NFL, relative to the strength of their recruiting classes. OSU is over +13 in this span, which almost seems unfair. Not only do they bring in the highest rated players, they actually develop them as well a high rate.  Penn State did surprisingly well in this span as well (+9.9), but the 2nd best team in the Big Ten is actually Iowa (+10.4).  MSU is sitting pretty in 4th place at +6.8, just above Wisconsin (+5.7) and there is another surprise in 6th place: Illinois as +3.5.

Most of the rest of the Big Ten is pretty close to average, but three teams do a particularly bad job: Michigan (-1.4), Minnesota (-3.6), and Maryland (-4.4).  

But, this really makes sense if you think about it.  The teams that have done well in Big Ten play recently are the teams that both recruit well and develop NFL talent at an above average pace: Ohio State, Iowa, Penn State, MSU, and Wisconsin.  Of the 16 participants in Big Ten Championship game, 14 of those slots were filled by those 5 teams.  Meanwhile, the teams that seem to "recruit well" yet struggle to win (*cough* Michigan *cough*) tend to be below average in the NFL development category.

Just to give a bit more detail for each Big Ten team, here is a full table of the expected NFL draft differentials by ratings bin for all 14 teams. Note that for this table, I also included walk-ons in a bin with a rating of 4.9.  There is no clear way to estimate the Power 5 average rate at which walk-ons make the NFL, so I just added 1 point of expected value bonus to each team's total for each walk-on.  This helps both Nebraska and Wisconsin quite a bit.

Based on this table, you can really see where teams struggle and where they excel.  As stated above, OSU struggles a little at the elite levels, but really excels in the 5.7 and 5.8 categories.  Penn State excels with the 5.5 to 5.8 recruits.  Nebraska and MSU are weakest with 5.8s. Wisconsin is weak in the 5.7 bin.  Iowa is strong across the board. Maryland and Minnesota are weak from 5.5 to 5.8.  You can also see how strong MSU is in the 5.4 and 5.5 category relative to the entire conference.  MSU has put 7 more players than expected in the NFL draft from these bins.  Finally, Michigan is clearly the place where Top 5 classes get turned into Top 25 teams and where 5- and 4-stars go to fade into oblivion.

Now, I could easily see the argument from our friends in Ann Arbor that this analysis does not consider recruits after the 2014 class, so essentially none of Harbaugh's recruits are included. That's true. It is also true that Harbaugh has already put a 5-star recruit into the NFL draft in Rashan Gary. Ultimately, time will be the ultimate judge of whether or not Harbaugh is any better than his predecessors at identifying and developing talent.  

But, I will say this: for the 2015 to 2018 classes, Harbaugh brought in a total of eight players rated either 6.0 or 6.1 in the Rivals system.  Of those eight players, one has already been drafted (Gary) but two have already left Michigan (Singleton and Asiasi) and two more are not listed as starters in their 3rd year (Solomon and Anthony).  In order to just meet expectations, Harbaugh better hope that the other three (Peoples-Jones, Bredeson, and Ruiz) all get drafted.  

As I see it, Harbaugh is most likely on his way to continuing the proud Wolverine tradition of underachieving when it comes to putting highly rated high-schoolers into the NFL.  So, when you see a few 6.1 or 6.0-level Freshman on Michigan's squad this year or on their commitment list, just remember that history suggests that Mark Dantonio's 5.4 and 5.5-level recruits have about the same odds (better actually...) to be drafted.  That is what the last 10 years tell us. I see no reason why that would change now.