Analyzing the Lagged Effect of College Football Recruiting

By Kevin Meers

nkimdiche
Robert Nkemdiche, Rivals.com #1 overall recruit

Player projection is tricky, a process that gets less accurate when we have limited knowledge of the players we are trying to evaluate. Given our particular lack of information on high school players, in comparison to what is available for professional athletes, it would be fair to expect some additional difficulty in projecting prospective college recruits. We have, at most, four years of tape and statistics for any high school player, and little ability to account for many potentially confounding variables. Because of these limitations, it would make sense if the highest-touted athletes out of high school were less likely to have a reliable impact on their team’s performance than we expect them to. With another class of high school football recruits recently committing to their programs of choice, I decided to examine the association between having a highly touted recruiting class and future team performance.

To tease out this relationship, I analyzed data from the 2006 through 2012 college football seasons via Football Outsiders and from the 2002 through 2011 recruiting classes via Rivals.com, totaling 842 observations. I measured team skill through F/+ and the strength of each incoming class through Rivals’ point system. (F/+ is a percentage stat that combines two college football team-level statistics, FEI and S&P+, provided by Football Outsiders. For more on the methodology, go here.)

Running an ordinary least squares regression of F/+ on lagged values of recruiting scores and F/+ from previous seasons, I found the following relationship:

recruits

This regression table shows each variable name with the regression coefficient to the right and robust P-values in parentheses below them. The notation N-1 refers to a given year’s freshman recruiting class, N-2 refers to the class a year before… you get the idea. The same notation goes for the F/+ lags.

This model shows a clear lagged effect for high school recruiting. The freshman recruiting class has no statistically significant relationship with F/+, but the sophomore and junior classes do have significant (at the 90% confidence level, at least) positive effects on F/+. These effects, however, are very mild: the difference between the average number one recruiting class and having zero recruiting points is about 7.5 percentage points of F/+: about the difference between Texas A&M and Notre Dame in 2012. The marginal effect of getting a top-rated high school recruit is 0.40 and 0.42 percentage points of F/+ during his sophomore junior years, respectively. That is a large impact for one player to have on a team’s expected performance, but is also fairly small given the large range of F/+ every season (generally between -30% and 50%).

How the team performed during the previous season is clearly the most significant predictor of current performance, and greatly overshadows the observed effect of high school recruiting. The coefficient on F/+ N-1 is below 1, meaning that the year N’s F/+ should be closer to 0 than in year N-1. This result provides more evidence for the idea that teams tend to regress towards mediocrity instead of having back-to-back seasons of dominance or incompetence.

I want to stress that this relationship between recruiting scores and F/+ may not be directly causal. Having high recruiting scores may be a signal for a different root cause like coaching or program quality (for example) that truly impacts team performance. Indeed, program quality would help explain the very high correlation in back-to-back recruiting scores (about 0.83). Further, there is probably some simultaneous causality in this model. If a team is good in one season, it may attract better players that offseason, making the team even better, and creating a positive feedback loop. That said, we should expect the schools that had highly successful recruiting this year (looking at you, Alabama), to have continued success for the next few years.

There are many extensions that could come from this data that we hope to examine in future studies. How much does team quality affect future recruiting scores? Is offensive or defensive performance affected more by high school recruiting quality? Please leave your suggestions in the comments below.

About the author

harvardsports

View all posts

8 Comments

  • I’d break out the regression to offensive and defensive categories. I predict you’ll find a stronger relationship on the defensive side of the ball.

  • Very nice analysis. But you should also take into consideration that many top players (possibly including many top recruits) don’t stay all four years? The best players go to the NFL in 2 or 3 years after their first year, so I would expect to see the drop as a result.

      • I agree, Dave. I find the analysis quite thorough in that regard.

        Additionally, however, I have colinearity concerns with the data. Have you run a correlation matrix and tested for VIFs?

        Furthermore, for robustness, you should consider that the data is not likely linear, as one would certainly diminishing marginal returns from recruiting class talent level, especially.

        I think that a Logit model may be more appropriate, which may be evidenced in part by the fact that the constant is significant, when the real data should be expected to regress through the mean, which almost certainly has to be positive due to the skewness.

        Finally, robustness could include utilizing FEI, S&P, and the separate Defensive, Offensive, & Special Team metrics as the other reply notes. Though, certainly separating these recruiting points may be more difficult, particularly since players are not always contributors in their primary recruiting position.

  • One more thing; try re-running this and only use BCS teams. Available evidence clearly shows that recruiting rankings are less predictive for the non-AQ conferences. I think this is for a couple of reasons. First much less recruiting analysis is done on the lower level athletes. No analysts are writing things like “Navy got a bunch of high 2-Star players, but Army got nothing but weak 2-Stars”.

    Second, at the lower levels the basic physical measurements are pretty much the same but the skillsets can vary quite widely. Successful recruiting at the non-AQ schools depends very much on getting those kids whose talents match what you are trying to do. A 2-Star OG who can block in space but not pull well needs to go to a team dedicated to zone running and not a power-based scheme. That’s info that can’t be gleaned from one-size-fits-all numbers.

    That is important at the higher levels too, just not as much. If the guys you recruit at BigState are simply much bigger and much faster then everyone else you can get away with gaps in the skill set. The only time you get exposed at BigState is when you play another BigState who has recruited great athletes as well as you have.

    Anyway, what you’ll find at the non-AQ schools is that recent performance levels(F +/-) influence future performance levels far more then recruiting. At the BCS schools gross recruiting numbers have a greater influence.

    Frankly I’m surprised HSA hadn’t figured this out already.

  • MGOBlog did a nice analysis on position impact and W/L records. They discovered that QB and the three defensive positions are the most important (DL,LB and DB). I have also read some cursory studies on the relationship between recruiting and team success; the correlation seemed to be at about 40% or so.

  • A red wine hardly needs this consideration; it will
    contain abrasive materials in it. It does not require many years of
    training or copying what others have done in the house.
    Copyright c 2013 Freelance Press Network There are many types of corkscrews available, though all serve the same but
    grown in different regions taste different. A wine carrier defeats both of
    these issues with one fell swoop.

Leave a Reply

Your email address will not be published. Required fields are marked *