THE BOOK--Playing The Percentages In Baseball

I just read part of this book last night.  Gladwell’s cherrypicking of examples is beyond infuriating.  His footnote regarding Harvard admissions is laughable for both its numerical errors and the its implication that Harvard looks only at standardized test scores and class rank as sole criteria for admissions.  He frequently overstates or otherwise butchers the results of scientific studies to fit the story that he would like to tell.  And this is all coming from a person (me) who basically agrees with the general viewpoint of the book.

Anyway, regarding birth month, another likely cherrypicked example given in the book is the Czech U-20 2007 soccer team that performed well in the championships that year.  Only 3 of the 15-20 players had birth months in the second half of the calendar year.  Gladwell: “At the national team tryouts, the Czech soccer coaches might as well have told everyone born after midsummer that they should pack their bags and go home.” The implication, of course, is that the Czech system is so rigid that essentially half the talent is being squandered and has no chance of becoming great “outliers.” Hmm… how then to explain the current composition of the Czech national soccer team (not U-20, but the very best Czech players of all ages)? This info is available on soccernet.com and other sites… anyway Jan-March: 5 players, April-June: 5, July-September: 2, October-December: 5.  Of course, perhaps the Czech system has changed dramatically in the last 5-10 years and has become far more rigid, but there are more likely explanations that have to do with Gladwell himself.

The irony is that Gladwell implies in another part of the book that 10,000 hours of effort at a task is far more important than innate talent itself in determining who will become an “outlier”.  I am not sure even he believes this, as it seems to contradict his first implication that Hockey talent is being squandered.  If innateness is a small factor, why worry if you focus all your efforts on people born in the first half of the year?  The results should be approximately the same.

I would love to hear your opinion about the Canadian Hockey stuff Tango, as it is something I know little about.

Finally, I would just like to say that I am a regular reader of this and other good analytical baseball sites because of the effort taken by Tango, MGL, and many others not to produce results based on biased samples, and to be very careful, and as “scientific” (for lack of a better word) as possible when performing studies on baseball.  If only Gladwell exhibited a fraction of this responsibility and restraint.  All I could think about when reading this book was: even most amateur scientists and sabermetricians I know would never be so bold, irresponsible, and arrogant to present data in this shoddy manner.

auntbea/1: you should take a look at my post.  In Canadian junior hockey, there is a clear bias towards players born early in the year.  However, players born late in the year are much more likely to make it from junior to the NHL.

Gladwell’s point about 10,000 hours, as I understood it, is that someone needs to put in that time at the highest levels in order to be a top talent.  He notes that players born in the second half of the year never get that opportunity just because of the way the age cut-off system works.

In anecdote after anecdote Gladwell implies and sometimes outwardly states that there is a threshold for greatness, above which more innate talent (be it IQ, height for basketball players, or musical talent or whatnot) confers essentially no advantage.  P 39: “The striking thing about Ericsson’s study is that he and his colleagues couldn’t find any ‘naturals’, musicians who floated effortlessly to the top while practicing a fraction of the time their peers did.  Nor could they find any ‘grinds’, people who worked harder than everyone else, yet just didn’t have what it takes to break into the top ranks.” P 79: “Once someone has reached an IQ of somewhere around 120, having additional IQ points doesn’t seem to translate into any measurable real-world advantage”.  P 80: “What Hudson is saying is that IQ is a lot like height in basketball… A basketball player only has to be tall enough”.  There are more anecdotes about nobel prize winners, etc.  Gladwell minimizes the role of innate ability at every chance he gets.

I did look at your post and I know there is a bias.  All well and good.  What I would like to see explained is if we should be especially concerned as members of a community (in this example, as fans of NHL Hockey) that we are missing out on the best players somehow because of this bias.  My inclination is that we don’t miss out on such truly great hockey talents very often because of the birthdate cutoff, as I was attempting to show from using the Czech National Team. For the “rank and file” of the NHL to have a slightly higher percentage of birthdates from earlier in the year does not seem, to me, to be much of a travesty.  From what we know of talent distribution for MLB and other top sport leagues, my first inclination would be to think that mostly marginal talents are being “squandered”, so to speak, but likely not many truly elite talents.  These will rise to the top despite early bias in their careers.  Sure it is a shame for the individual who could have been a borderline NHLer had he been born in January and not November, but does the league as a whole suffer if you replace a few players at the margins?  I am not convinced.  Much more interesting to me is Tango’s investigation into the bias against players from Europe v. their North American counterparts.

Another footnote from Gladwell: “A physically immature basketball player in an American city can probably play as many hours of basketball in a given year as a relatively older child because there are so many basketball courts and so many people willing to play.  It’s not like ice hockey, where you need a rink.  Basketball is saved by its accessibility and ubiquity.” While I don’t know enough about hockey in Canada to say whether this is accurate or not, it sure sounds wrong to me.  And furthermore, since Gladwell is essentially equating bias in hockey with bias in European soccer, i would like to hear his “explanation” for why soccer is harder to play across Europe than basketball is in a U.S. city.

auntbea/3 - i’m not sure I understand why you think that only marginal talents lose out because of birthday bias.  At age 8, you can’t identify future stars, but the bias manifests itself that early.  Plenty of players never get their chance at all.

From what we know of talent distribution for MLB and other top sport leagues, my first inclination would be to think that mostly marginal talents are being “squandered”, so to speak, but likely not many truly elite talents.

I like your idea, but I disagree.  Sure, we know the Grezky’s and Bobby Orr’s when they’re 10 years old, because they’re playing with 14 year olds and leading the league in points.  But what about guys like Mike Piazza, who are late bloomers?  Or more abstractly, kids who will have the physical makeup to be elite once they hit puberty, but just hit it a little later than the rest due to being at the wrong end of the age cutoff?

As much as it bothers me to see parents flip out about their 10 year old not making the “A” squad of the travel team, I absolutely believe that this can be critical to whether or not they have a chance at the pro’s (for like, .01% of the kids).  If you’re on the B team when you’re 10, you have less experience against good players, you’re not pushed as much, you don’t learn as much as you would from the A coaches, A teammates, and A opponents.  That can be the difference between the standout high school player (who did get on the A team at 10 years old) and the natural who can hop into a softball game with those varsity players, hit pretty much just as well, but won’t ever go anywhere with it.

Agree with auntbea/1.  Gladwell’s book is a great read, and I very much agree with his thesis.  That being said, the tendency to cherry pick is pretty annoying.

And he must be wrong about height and basketball, no?  I would think the percentage of 7-footers who make it to the NBA is a a couple orders of magnitude (literally) greater than the percentage of 6’7” guys who make it to the NBA.

Outliers, P 156 (on becoming a successful “outlier") “That’s an unstoppable combination.  That’s like being a hockey player born on January 1.”.

Curious, I looked at the top 50 point scorers and top 50 minutes leaders for this season in the NHL, and ignored anyone who was not born in Canada.  The results by quartile? According to my quick tally, there were 25 Canadian born players in the top 50 scoring leaders.  Jan-March: 4, April-June: 9, July-Sept: 9, Oct-Dec: 3.  For minutes leaders (almost all defensemen, so pretty much an entirely different sample even though the numbers look similar).  Jan-Mar: 5, April-June: 8, July-Sept: 8, Oct-Dec: 4.  For the minutes leaders, not a single Canadian-born player was born in January.

Outliers, P31: “So What do you do if you’re an athletic young Czech with the misfortune to have been born in the last part of the year?  You can’t play soccer.  The deck is stacked against you… Those born in the last quarter of the year might as well give up on hockey too.”

It is this kind of hyperbolic discourse about predestination by circumstance that is so irritating about the book.

But what about guys like Mike Piazza, who are late bloomers?

The implication in Outliers is that these guys would have no chance, regardless of their birthdates, as they would have missed out on all the crucial playing time and training necessary for success very early in their lives.

auntbea/7 - Let’s assume that an equal number of NHL-caliber players are born each month.  (That seems like a pretty reasonable assumption, does it not?) How do you explain that among junior players aged 16-20, there are almost four times as many players born in the first three months of the year than in the last three months?

I don’t understand your question.  I readily acknowledge there is bias in these leagues.  This bias, not incidentally, is reflected in most junior sports organized this way throughout the world.  My question to you is: “Does this bias affect the overall quality of the NHL in a meaningful and significant way?” I see little evidence of that.

Like i indicated earlier, other types of biases (against European players for example), might have important and lasting effects for future generations of NHL players.  I fail to see how a bias by birth month could have effects that last into the future.  The only affects are ones that should be readily apparent right now.  I am not convinced.

Of course it affects the overall quality of the NHL.  The birthdate bias already shows up among player registrations at age 8.  If you had separate leagues for players born in the second half of the year, then you’d have almost twice as many athletes to pick from.  Whether you think the issue is natural talent or hours devoted to playing, registering twice as many players at a young age will give you twice as many good players later.

We already know that the junior players born in Q4 are better on average than the junior players born in Q1.  So it would make sense to get more Q4 players, would it not?

If you had separate leagues for players born in the second half of the year, then you’d have almost twice as many athletes to pick from.  Whether you think the issue is natural talent or hours devoted to playing, registering twice as many players at a young age will give you twice as many good players later.

I don’t understand your thinking here at all.  Are you trying to tell me that none of the players in the second league would be taken from the pool of players currently in the first league?  So… double the coaches and the money and everything else?  By this logic, why not have a league for each day of the year?  Then you would have 365 times the number of players to choose from.

More realistically, let’s use the numbers you have on your blog post to get a better idea.  You have 31% of players in the NHL from Jan-Mar, and 19% from Oct-Dec.  So let’s say maybe 27% from Apr-Jun and 23% from Jul-Sep.  That means 58% from the first half of the year and 42% from the second half.  If you had two junior leagues, one for each half of the year, the idea is you would equalize that to 50/50, in effect replacing 8 players per 100 from Jan-Jun with ones from July-Dec.  My suggestion is that the 8 players per 100 added in from July-Dec would virtually all be rank and file NHL types, and not truly standout talents (for the NHL anyway).  I find it unlikely that the Gretzky’s born July-December would not have gotten the very best training and ice time throughout their young careers.  If I am correct, it makes very little difference to the overall quality of the league if you replace 8 players per 100 at the margins with ones who would be very slightly better had they been given the opportunity.  I assume, as in MLB, there are many players of borderline NHL caliber that just have not been lucky enough to make the league, all waiting in the wings, so to speak.

I continued a bit with my quick tallies.

Birthdates of all goalies that played enough to qualify for the “goals per game average” list: Jan-Mar: 7, Apr-Jun: 5, July-Sep: 4, Oct-Dec: 4. 

Extending point leader list to top 100 instead of 50 I get Jan-Mar: 7, Apr-Jun: 20, July-Sep: 14, Oct-Dec: 13.

For minutes leaders top 100 (probably a few player overlap with above list now) Jan-Mar: 12, Apr-Jun: 14, July-Sep: 13, Oct-Dec: 12.

All numbers for Canadian-born Hockey players only.

Canadian-born players in the 2008 NHL draft by year quartile and round:

1st Round Jan-Mar: 7, Apr-Jun: 4, Jul-Sep: 1, Oct-Dec: 6.

2nd Round Jan-Mar: 7, Apr-Jun: 6, Jul-Sep: 5, Oct-Dec: 2.

3rd Round Jan-Mar: 10, Apr-Jun: 5, Jul-Sep: 3, Oct-Dec: 1.

I am not sure i want to turn this into a real study, but it seems with these small sample sizes there seem to be some patterns emerging.  The higher the ceiling for the player, the less birthmonth seems to matter.  This is what you would expect if birhtmonth were more a tiebreaker than a determining factor.

I may not have followed all of this correctly, but is Gladwell talking only about physical ability in this chapter/section? I suspect intelligence would not be subject to the same biases that he is describing, because everyone starts school in the fall and ends in the spring whereas physical ability is cultivated on a more seasonal basis. In other words, everyone does school at the same time for the same length, but some people can get a “head start” in sports.

If I’m way off base here then I’ll just keep my mouth shut and read. Interesting discussion going on here.

auntbea/12 - “You have 31% of players in the NHL from Jan-Mar, and 19% from Oct-Dec.  So let’s say maybe 27% from Apr-Jun and 23% from Jul-Sep.  That means 58% from the first half of the year and 42% from the second half.”

I’m not sure where you got those numbers.  If you look at the chart labeled “percentage of canadian junior players born by quarter”, the split is 67/33 between the first and second half of the year.

/13 - The ratio of Q4 to Q1 NHL players over the last 25 years is approximately 2:1.  Your “quick tallies” from a portion of one year’s worth of data do not address this issue.

/14 - I think NHL scouts are aware of the difference between a player who just turned 17 and a player who’s about to turn 18.  Because they’re concerned with a player’s ceiling, they will not draft based on expected performance this season.  On the other hand, junior coaches will dole out ice time based on who’s the best player at the moment.  So we would expect the birthdate skewing to be less extreme in the NHL (and it is.)

I got the NHL numbers 31% (Jan-Mar) and 19% (Oct-Dec) from YOUR CHART!  Then I extrapolated the 27% (Apr-Jun) and 23% (Jul-Sep) from those first 2 numbers.

We know there is bias in the Canadian junior leagues.  However, my contention was that this bias would not be nearly so evident in the top ranks of NHL Hockey.  My quick tallies of NHL leaders seems to back this up.  The only way to really tell if the “Matthew Effect” (cumulative advantage) plays an important role in Hockey is by seeing if the advantage increases as one goes up in the Hockey ranks.  All the evidence I have seen shows exactly the opposite.  The higher up you go in Hockey in Canada, the less birthmonth seems to matter.

You could also simply say that the Junior coaches give more ice time to the players that are better simply because they are older (thus bigger, faster, and stronger, and more experienced).  So age bias in a junior league alone is not evidence of a “Matthew effect”. If Gladwell is right, then by the time the January players get into the NHL, they really ARE better than the December players, and thus should get more ice time even in the NHL, thus increasing their advantage.  There would be no reason to expect the relative numbers of Jan vs Dec players to even out for points and minutes leaders.

More quick tallies from this season’s NHL: top 50 canadian-born penalty minutes leaders by quarter.

Jan-Mar: 14, Apr-Jun: 9, Jul-Sep: 9, Oct-Dec: 4.

Almost none of these guys are leading the league in points and minutes.  Anyone surprised?  Most people would not consider these the most “hockey-talented” (so to speak) players in the league. 

Note one other interesting fact.  A much higher percentage of these players were born in Canada than that of the points and minutes leaders.  That speaks to Tango’s bias against European players.  Once again, the lower tiers of the league seem to display the birthmonth bias while the upper tiers do not.  Now why should we have two junior leagues again?

Bottom 100 Canadian-born players by minutes played.

Jan-Mar: 18, Apr-Jun: 17, Jul-Sep: 13, Oct-Dec: 10.

So, if we have two junior leagues instead of one, we swap a few bench warmers around?  Let me know if I am missing something crucial.

As of today in this year’s NHL:

Top 200 Canadian-born players by ATOI (average time on ice).
Jan-Mar: 22, Apr-Jun: 29, Jul-Sep: 29, Oct-Dec: 20.

Bottom 200 Canadian-born players by ATOI.
Jan-Mar: 37, Apr-Jun: 36, Jul-Sep: 22, Oct-Dec: 19.

I just want to clarify… in the preceding 3 posts the numbers were for the top or bottom number of players in each category, ignoring anyone not born in Canada.  Thus there are only 100 Canadian-born players in the “top 200 by ATOI” list, and only 114 in the “bottom 200 by ATOI” list.

I think you’re analyzing too small a group of players and coming up with the wrong result.  Here’s a quick quote from the guy who writes hockeynumbers.blogspot.com:

I did a quick regression with both:
- number of players vs. birth month (r2 = 82%, r=0.91)
- points per game vs. birth month. (r2 = 53%, r=0.73)
n = 1210

The younger group (Jul – Dec) scored on average 0.43 points per game, where the older group (Jan – Jun) scored 0.36 points per game (significant difference)

Too small a group of players?  Sigh.  You really just won’t get it will you?  I have already given you the breakdown by “average time on ice” for 400 out of the 735 players listed on espn.com. 

Like Gladwell, you seem painfully un-numerically savvy.  Of course there will be some correlation between points per game and birth month, even if only the bottom tier of players are biased by birth month.  Why is this?  Simple.  Even the bottom tier of the league, for which we all know there is bias, scores some points per game.

I see where you went wrong.  Outliers, P 268: “If Canada had a second hockey league for those children born in the last half of the year, it would today have twice as many adult hockey stars.  Now multiply that sudden flowering of talent by every field and profession.  The world could be so much richer than the world we have settled for.”

This argument is astounding for its stupidity.  Apparently Gladwell does not understand the concept of diminishing returns.  It would be amusing if it weren’t so insidious.  Here he pretends returns would not diminish at all (full returns), but in an earlier case (basketball players over 6’6") he pretends you get no returns at all for extra height.  In other words, he comes to whatever conclusion suits his worldview, despite the fact that anyone can see his conclusions have no basis in reality.  Gladwell is simply not a serious writer about these subjects and I am beginning to think he does much more harm than good.

I don’t think you need to resort to insults because you disagree with me.

I’ve presented 25 years of birth data that shows a 2-3:1 ratio between Q1 and Q4 birthdates in the NHL.  You’ve gone through 30 games of data from 2008 and I don’t think it’s enough to show that the birthdate bias results in no impact on the caliber of the NHL. 

I’ll run the numbers back to 1980.  I think that will help answer the question.

None of the data that you have presented, none of it, breaks players down by how good they are relative to the league (NHL).  All of your data treats all NHL players the same.  You have completely missed the point of my long and tedious explanations.  Furthermore, while historical data is certainly interesting and important and gives a larger sample size, the critical question is whether or not we are being impacted today.  Thus “running the numbers back to 1980”, while admirable, is not necessarily all that relevant if we have enough of a sample with fewer years of data.

I didn’t miss the point at all.  I’m just saying that we need to look at points-per-game for more than 30 games of 2008-09.  Your assertion is interesting but the sample size is small enough that we could be seeing an erroneous result. 

Perhaps this effect is new?  Perhaps not?

I looked at 2000-2001 because it was the first season on the sortable stats at espn.com (or I think it was anyway).

2000-2001 Canadian-born points leaders (92 out of 200 total players born anywhere).
Jan-Mar: 19, Apr-Jun: 25, Jul-Sep: 29, Oct-Dec: 19.

2000-2001 Canadian-born “average time on ice” leaders (90 out of 200 total).
Jan-Mar: 20, Apr-Jun: 22, Jul-Sep: 26, Oct-Dec: 22.

2000-2001 Canadian-born “average time on ice” bottom tier (136 out of 200 total)
Jan-Mar: 35, Apr-Jun: 52, Jul-Sep: 27, Oct-Dec: 22.

I looked at the top 25% of NHL players who played Canadian Junior in 1) Points; 2) Games Played; 3) PIM; by year from 1984-2007.  The percentage of first half birthdays as a percentage of the total number of first half birthdays is:

Top 25% of Pts: 94.6%
Top 25% of GP: 93.4%
Top 25% of PIM: 98.1%

So if 70% of overall birthdates were in the first half of the year, then 66% of top point scorers were born int he first half.  The individual yearly data points are very noisy due to the small number of players in the 25% sample - there doesn’t appear to be a trend in the data. 

For points, the percentage of first half birthdays has been between 55% and 60% for the last decade.  If we look just at Q1 vs Q4 birthdates among top scorers, it’s a bit broader - 55 to 65% Q1.

If the two halves of the year were equal, we’d be looking at about 10 extra Canadians in the top quarter of the league, or a 7% boost.  If the age bias exists in all of the hockey player-producing nations, that would be enough to offset the last expansion. 

This of course assumes that without the bias, we’d have a 50/50 split between first and second half birthdates.  But given two junior players with identical stats, the second half player has a higher ceiling than the first half one...So perhaps the equilibrium would favor the second-half players?  My guess it aht Gladwell’s idea, implemented worldwide, would generate enough players for 4 additional NHL franchises.  That’s not an insignificant total, but he probably thinks it would generate 12 or 13.

Interesting numbers.  I would guess, from looking at the current data (and your graphs) that this bias has been slowly disappearing over time.  We know the current first half to second half ratio for the whole league is about 58% to 42%.  By your numbers that means 55% first-half babies in the top players.  My guess is that it is even less now as scouting and player evaluation gets more and more sophisticated.  At any rate, you say enough players for 4 new franchises, I say enough for maybe 1 or 2 more.  I think maybe the answer is somewhere in there.  But by Gladwell’s quote (that I cited above), he very explicitly says “twice as many adult hockey stars”.  That means enough for 30 more franchises.  He is off by a factor of 10.

Perhaps we are going through a special time during which there are a glut of late birthdays just by chance, but so far, for 2000-2001, 2005-2006, and the 30 games of 2008, I see no bias at all towards the first half of the year for Canadian-born points leaders.

Of the 200 points leaders in 2005-2006, i found 94 born in Canada.

Jan-Mar: 16, Apr-Jun: 31, Jul-Sep: 30, Oct-Dec: 17.

I don’t have good access to a database that lets me know who played Canadian Junior Hockey, or even who exactly is Canadian, only where the players were born.  So for that reason my numbers could be slightly different.

It doesn’t seem like the proportions are changing over time.  There were more late birthdays in 1986 and 1994 than there are now, but with 400-600 Canadians in the league (and 100-150 in the top 25%) a dozen players having good seasons can swing the numbers.  Also, I only looked at players who played juniors 1980 and later, so in the early years, I’m not getting every NHL player who played junior.