Tuesday, January 31, 2012

Talent ID, USTA Player Development Folly

Identify the talented kids.  Bring them to a regional or national training center.  Get them the best coaching.  Put them through the best drills.  Train them over time to be champions.

That's the basic idea of talent identification within the USTA Player Development program.  It seems obviously the right thing to do.  But is the first step possible?

In this post I'll start off with some analogies that pretty clearly show the difficulty of projecting talent or skill into the future.  In another post I'll go into some of the science behind such human judgments, citing some relevant research and explaining the theory behind human judgment.

Imagine you're a National Football League general manager.  It's your job to acquire the most skilled football players for your team. You must look at players around the league who may become available as free agents or via trade.  Based upon their demonstrated performance at the NFL level you have to project how well they will perform on your team in the future, in this case probably from one to five years in the future.  This task is challenging, but not impossible.  I don't know all the criteria that GMs use to project future skill and I don't know their success rate in trades and free agent acquisitions.  I do know that they aren't either perfectly good or perfectly bad.  They have some rate of success between zero and one hundred percent.  If they're to keep their jobs the success rate probably has to be well over fifty percent or you could replace them with a coin.  Since GMs are frequently fired, perhaps replacing GMs with a coin would make sense for many NFL teams.

Let's increase the difficulty of the NFL GM's judgment problem a bit.  Each year NFL teams draft college players.  In this case, the GMs must predict how college level skills will translate to NFL environments.  They also must project over a longer time horizon, spanning from one to possible ten or even fifteen or more years in some cases.  Once again I don't know all the cues that the NFL talent evaluators use, but I know that they watch players in person and on tape.  I know that they frequently interview the players and the player's coaches.  I know that they use a variety of quantitative measures of speed, strength, size,  and even mental abilities.  Despite all these cues, the success rate for the NFL draft is considerably below one hundred percent.  No team goes even one draft year without making several mistakes.  To succeed in placing even half of a draft class on an NFL roster is probably considered pretty good, given how few roster spots are available each year.  So NFL player personnel departments spend tremendous amounts of time and money in making judgments about the future skill of grown men who have already mostly mastered the skill under consideration.  They still miss on a great many of these judgments.  They pick players who fail to pan out and they fail to pick players who go on to have outstanding careers.  This happens every year to every team.

Let's leave the NFL now and go back one level to college coaches.  They too must select players for their teams.  They do this by evaluating high school football players.  These are mostly sixteen, seventeen, and eighteen year-old young men.  They have played football for probably ten years already.  They have all gone through puberty.  They are not yet at their full size and strength, but they are within a few years of reaching their full physical maturity.  College coaches must judge how well these players high school performance will translate to the college level.  They must judge how a player will develop in the year or so before he comes to college and how much he will improve in the four or five years he spends on campus.   The younger the player, the tougher this judgment becomes.  Coaches who commit to juniors in high school have a higher error rate than those who commit to seniors in high school.  But the race to get the best players often induces coaches to commit to a "can't miss" player in that player's junior year.  Unfortunately they can and do miss.

Now let's jump to the talent identification problem facing the USTA Player Development coaches.  The goal of USTA PD is to produce world-class players.  The assumption is that the US Open, the USTA's big money maker, will be more popular with American players in the championship hunt.  The USTA's goal is to grow the game and the organization thinks the game will be more popular with Americans at the top of the professional game.  To develop these champions, the USTA PD tries to identify young kids with the potential to become world-champion players.  In light of the above discussions, how likely is the USTA to succeed in identifying children under age ten who will dominate professional tennis from age twenty to age thirty?

Projecting tennis skill, like football skill, a few years into the future is not easy.  The list of national junior or international junior champions does not completely overlap with future grand slam winners.  So skill in the late teens does not perfectly predict adult skill.  It's pretty good, certainly, but far from perfect just like in the football case.

The judgment of which ten-year olds will be great players at twenty is much, much tougher.  These kids are all yet to go through puberty.  They are a decade away from achieving their full size, strength and maturity.   Skill at age ten is not perfectly uncorrelated with skill later on.  I don't think there is a case in the last several decades where a player who subsequently became a world champion was not a very good player by age ten or twelve.  But the universe of all skilled players at age ten does not overlap very much with the set of world champions, that's for sure.

The probability that coaches can accurately judge the skill at age twenty of a child at age ten is vanishingly low.  The cues that coaches use simply are not valid, they are not stable over the time frame necessary to make accurate judgments.  Unless we discovered that skill was like eye-color or sex, that is completely genetically determined, the judgment task is essentially impossible.

By selecting small groups of kids and bringing them to regional and national training centers, the USTA Player Development is assured of making numerous errors.  They will invite plenty of kids who fail to become great and they will fail to invite kids who do become great.  By great, I don't mean world champions, unfortunately.  I really only mean great juniors or possible professionals.  I think world champions are Black Swans as Nassim Taleb would say, rare and unpredictable.

So rather than wasting resources on dozens or even hundreds of kids who show promise at a young age, the USTA Player Development system should spread its dollars far and wide to make sure the net captures as many players as possible.  By bringing more kids into the game they will by definition be growing the game.  If by chance some future champion resides in the US, then the wide net will have captured that player, too.

No comments: