Analytical approach to the inexact science of recruiting

Submitted by Kewaga. on

 

 

Background:

A blog post Ben Weiss wrote for a Northwestern fan site made its way to the desk of athletics director Jim Phillips, who then forwarded the link to Chris Bowers, the Wildcats’ director of football player personnel.

Inside Weiss’ article was a breakdown of Northwestern football’s recent recruiting efforts –— a deep dive into the Wildcats’ hits and misses, and so detailed in its methodology that Bowers called Weiss into his office and offered him a position.

 

Thought Process:

“Schools are looking, naturally, for the best players in the country that are somewhat likely to commit to their school,” Weiss said. “So what we’re saying is, ‘Look, you’re essentially trying to do this in your head.’ We have proof that this can work and this is how we do it, and this is what we do.”

Boiled down, Zcruit’s goal is to assist a program’s efforts by streamlining the process — by taking all the streams of data at their disposal and creating a formula for recruiting success, in the same way a university’s admissions office attempts to pinpoint the best and most likely fits for the student body at large.

 

Means:

Three baseline factors are taken into account. The first is demographic information: background information, such as where a recruit is from. The second is a prospect’s interactions with the school, such as how many visits he has made on campus, whether he attended any camps or when the scholarship offer was tendered.

The third is the prospect’s interactions with other schools. Is he showing any interest? When was he offered by another school, when did he visit, how many times did he visit? In the end, the data compiled by Zcruit creates a threshold, for lack of a better word, between whether a program should recruit a player — if the data suggests he’s gettable — or whether it should move on to another prospect.

 

Results:

The algorithms created by Zcruit have predicated which recruits would not sign with Northwestern with 94% accuracy; the same algorithms predicted which recruits would sign with the Wildcats with 80% accuracy.

 

Conculsion:

After all, an entire regular season is simply composed of a dozen 60-minute chunks. In comparison, exponentially more time is spent recruiting, from the period of early evaluation — when teams create their general board of prospects — through the final home stretch into signing day. So an analytical approach wouldn’t simply help programs sign specific recruits; it would streamline the entire operation, allowing coaches to not just save time but spend their time effectively.

 

http://www.reporternews.com/story/sports/ncaaf/2017/01/23/analytical-ap…

 

Hot take:

Up for anything that might give Michigan an edge in recruting 

schreibee

January 24th, 2017 at 12:41 PM ^

Yeah, 99% accuracy more like it. 

Let's start by removing the pool of ALL GOOD PLAYERS, and then winnow down from there.

After removing all the 5* & high 4* players, you take the rest of the pool and ask:

Are they on the Honor Roll? No? Not coming!

Do they have any family connection to NW? No? Put them in the unlikely pile.

Do they live within a 2-hr drive of Chicagoland? No? Put them in the even more unlikely pile...

Now using the same methodology for Michigan recruiting would be completely fruitless. How would you even begin to winnow out who M might want but is unlikely to get?

Does he have Bama logos on his Facebook/Twitter/Instagram? Send a letter, but don't waste any time til he expresses interest in visiting us.

Same for osu, except send someone by his school to recruit a teammate and try to get him interested. 

Is he from SoCal? Does he have Trojan logos all over his Facebook/Twitter/Instagram? Follow Bama protocol...

Zenogias

January 24th, 2017 at 10:00 AM ^

Haven't RTFA, but we really need to know how humans do as well. As A Lot of Milk notes above, a lot of humans can tell you who isn't going to to Northwestern. And I bet a lot of people can identify kids who are likely as well. None of which is to say an analytic approach can't be better, but you gotta establish what the baseline is. If humans are predicting at 95% and 85% (for example) then the algorithm isn't adding much.

tasnyder01

January 24th, 2017 at 10:01 AM ^

When were the predictions made? I mean, isn't Wiltfong around 80% accurate too? Its just he can change picks at the eleventh hour.

The model is supposed to save time by picking Early, right?

Also, kind of sounds like a self-fulfilling prophecy. If it does name people who wont sign, and does so early, wont that mean the coaches wont spend time on the recruit. . . Leading him to not sign?

Ecky Pting, Mathlete: Get on this, stat. (pun intended)

Markley Mojo

January 24th, 2017 at 10:29 AM ^

Recruiting is basically selling what your school offers, and there's a variety of tools aimed at customer relationship management (CRM) analystics.

I don't think you would use this to drive decision making, but it can be useful as a corrective against falling in love with a recruit that you're *sure* is going to sign with you. Keep track of risks and red flags, assess the likelihood that another school that hasn't yet made an offer will come through with one, assess the impact that would have on your own recruitment, etc. The goal is to try to automate some of the (human) things you've learned so that you won't have a blind spot in the future. Like having a card for two-point conversions, or a Madden player for clock management.

Also, it wouldn't be terribly predictive for Michigan, since we're still in the early days of Harbaugh. Worth a call to the Ross departments of marketing and operations, though. Get a few PhD students to look at it.

 

michelin

January 24th, 2017 at 12:03 PM ^

The NU prediction tool seems to have value.  But it is only a very crude prediction tool, and the task coaches face is not simply prediction.  It is a broader decision problem that must consider many other factors—some easier and some harder to analyze but all of them addressed by existing methods.   In the next 5 sections, I mention a few of these.  At the very least, these sections should illustrate how complex the recruiting problem is.  Ideally, however, I hope that a student here may apply one of the methods listed (possibly with financial assistance from the AD). 

 

1.       Resource Allocation and Cost-effectivess/cost utility Analysis

Successful recruiting requires us to consider not just the likelihood of the recruit committing but also the money and time needed to pursue him.  That pursuit decreases the resources that could be used on other players.    In addition to considering the chance a recruit will commit (NU method), we need to consider his quality and whether he fits position needs now or in the uncertain future, as well as the sometimes uncertain constraints of class size.  Operations researchers have extensively studied how to allocate limited resources within such constraints (e.g., money and time).  

michelin

January 24th, 2017 at 12:06 PM ^

2.       Advertising effects.

A hidden value may emerge from pursuing a low probability/low quality recruit or having a camp in a region rich in more valuable recruits.  The same applies to many other possible efforts to increase visibility—like a “sigining of the stars” event, a camp in Samoa, or even or a trip with the team to Rome.   Many methods in advertising attempt to quantify the value of increasing visibility and brand awareness using various types of advertising.

3.       Social network analysis

Having a close friend on one’s HS team who is committed or has interest in a particular school can powerfully influence some players to join him.  Sometimes then, it makes sense to recruit the friend to attract the target player.  It may also make sense to take a middling recruit from a “pipeline” school—one that has or will have other, more valuable players in the future.  Hiring a coach from such a school may also work.  One may even hire coach who is the parent or other relative of a valuable recruit.  Many social science methods exist to assess influence networks, and such methods could help design more effective recruiting strategies.

michelin

January 24th, 2017 at 12:08 PM ^

4.       Game theory

In recruiting, we must not just pursue recruits according to their value, the chance of getting them, and the resources required to do so.  We must consider the constraints in time, money and class size of our competitors.  We also must consider possibly misleading information about their “offers.”  To misdirect its competitor, a school may get their own best and solid commitments to remain “silent commitments” and express false interest in another school.  They may also “offer” a recruit merely to elevate his apparent value. Such attempts at misdirection can get competitors to drain their resources by pursuing poorer quality recruits or high quality ones that are very unlikely to flip. 

 

Possibly, we might identify such schools by analyzing why certain schools end up not getting players who (according to the NU analysis) had a high probability of commitment.  Such schools may be more likely to have rescinded offers.   Also, military, economic and many other analysts have developed methods based on game theory to help analyze a competition---not unlike our competition with other schools for recruits.   

 

5.       Content Analysis

 In recruiting, there is a lot of soft information not captured by the NU prediction tool.  We often are influenced by crystal balls, the words of analysts and even those spoken by the recruit and family.   When we merely hear these words, it is often hard to gauge their meaning  (or lack thereof).  So they can instill false optimism and unnecessary use of resources.  Or they may lead to false pessimism and insufficient resource use.   All too often, the words present a lot of meaningless noise.  But analytic methods have sometimes distilled meaningful signals from similar noise.  Advertisers and financial analysts may do content analyses of positive or negative or neutral words.  In recruiting, such analyses could include many factors that influence their significance, such as when the words were spoken (e.g., before or after a visit).

 

colin

January 24th, 2017 at 12:23 PM ^

Without a baseline, it's hard to say that 80% is actually adding incremental information. Given how important recruiting is, I doubt there's a lot of value add considering the far more granular data a given coaching staff is going to have. Get the coaching staff on board to rate commit likelihood on a regular basis and you can really build a tool of some use.

 

steve sharik

January 24th, 2017 at 1:46 PM ^

Factor #2 on whether or not to recruit a player:

 

The second is a prospect’s interactions with the school, such as how many visits he has made on campus, whether he attended any camps or when the scholarship offer was tendered.

Um, isn't that already recruiting him?

Lee Everett

January 24th, 2017 at 3:35 PM ^

I wonder if it's *quantifiable* how much relationship building is.  

I'm sure there are instances where a coaching staff can look at this data and think, alright, there's a 94% chance that this kid does not sign with us, yet we want to recruit him for other strategic reasons: establishing a pipeline in that school, getting in with a certain coach, making inroads with another coveted commit that is related to or teammates with or friendly with a recruit that seems like a numerical lost cause by himself.

Or, how much a team's draft board, so to speak, can outweigh this tool.  For example, let's say they've identified a 3* that is an 80% chance to sign with them, but there's a 4* that is determined to be far less likely, at the same position.  How much do you you invest in the 4* versus the 3*?  

I certainly think this is a smart and valuable tool but it'll be interesting to see how they employ it, strategically.

Steve in PA

January 24th, 2017 at 5:47 PM ^

I was working with TomVH on something similar a very long time ago.  I was only consulting him via email not actually working WITH him in case anyone would think I am in any way associated with TomVH.

What I found and was trying to model is that 5* players generall stay close to home.  Sure, there is outliers but if it comes down to 3 schools a 5* will more than likely pick the closest one.

4* and 3* recruits were much more willing to travel.  by that I mean they would sign with schools further from home.  That may have more to do with their offers than actual willingness to travel.

Coach's record (by extension reputation) accounted for ~20% of the decision.  Record over the last 3 years was 3rd in my criteria.  There was a lot of random noise that I was trying to filter out that couldn't be modeled (bagmen, etc).

I was working on weighting when TomVH left, I got busy with work, and that's where it got left.  There's probably some old posts where I was trying to predict where the recruits would wind up.  If I remember correct I was around 80% which probably isn't much better than random darts.

eth2

January 25th, 2017 at 7:37 AM ^

Thanks OP. This is some pretty fascinating stuff. Recruiting has changed so much for schools, prospects, and fans in recent years and technology and data analysis are increasingly at the forefront. Not surprised a school like NW who has a bent for academics, but lacks the budget of top football programs would be looking to focus their efforts as much as possible. I'm wondering how many other programs have developed similar proprietary methods.

With the analytical chops at MGoBlog, maybe a similar model could be developed.