THE BOOK--Playing The Percentages In Baseball

Good stuff from Chris.  In last year’s THT he had an article looking at similar questions showing that a 22 year old who was a college draft pick will develop more power than a 22 year old with similar stats who’s drafted out of high school and thus has 3 years of experience facing professional pitching.

I think this is more important for projecting players.  I see an MLE as a value metric, an OBA of .375 at a given minor league park means the exact same thing whether you are 19 playing your first minor league season or 33 in your 14th try at that league.  But for projections age and experience makes a big deal.

I have not read Chris’ article yet, and I have not done any research on how age (other than using typical age adjustments) and experience affect development curves, but there can certainly be other reasons why a college draft pick might develop more power than a high school pick of the same age.  In fact, experience aside, I would EXPECT that to be the case.

When I was in undergrad, one of my professors used an old (from the 40’s or 50’s) published study to illustrate how even published studies by colege researchers can be significantly and obviously flawed.  The study he was referring to, looked at student athletes and non-athletes in junior high school.  They looked at their height increase from one age to another.  The athletes had significantly less height increase than the non-athletes. The researchers concluded that engaging in school sports stunts growth.

Of course this was a ridiculous study which failed to control for the developmental stage of the two groups of students and it suffered from an extreme case of selective sampling.  Even though we, as baseball researchers, are acutely aware of selective sampling problems and we constantly discuss it, apparently academics, at least in the 40’s or 50’s, were not so aware.  Apparently this study was not only done, but slipped through the vetting process (peer review, etc.) with impunity, I guess.

Anyway, the student athletes were already well into puberty (bigger and stronger) as compared to the non-athletes for obvious reasons (who is more likely to be on the football team - the 13 year old who is 5’8” tall and 150 lbs. and 2 years into puberty, or the 4’10” pre-pubescent kid?).  Therefore more of the athletes had already had their “growth spurts” and gained less height even for the same age (they did control for age of course).

What is the point of that story?  Most of the research I have done and seen indicates that the offensivive development curve is much more a function of age rather than experience.  I would assume that high school players who are drafted are much more advanced physically than the college draftees, given the same age, so of course they would develop less power than the college draftees, given the same age. A high schol 18 or 19 year old hitter is probably equivalent, physically, to a 21 or 22 year old college player.  So at 22, that high school draftee is probably physically equivalent to a 25 year old college player and possibly a 25 year old male in general (the high school draftees may be early maturers in general).  Everyone has a chronological age and a physiological age.  Those things need to be controlled for, if even possible, when doing studies like the one mentioned by Rally (the 22 yo college and high school draftees).

I just read the article and his results are interesting.  I would issue a warnning though.  One has to be very careful with selectice sampling issues in studies where players change roles due to decisions by managment.  I don’t think there are any large selective sampling issues with Chris’ study, but I am not 100% sure.  Also, although it is likely that there are no league or park biases in the two groups that Chris creates, I would like to see him at least look at that to be sure (or use park adjusted numbers).  The samples are small enough for there to be non-biased differences in park or league and there may be biases by organization as well.

Here’s an example (as far as I know the only one, his career path is not recommended) that suggests age is a much bigger factor than experience:  Josh Hamilton.

At age 19 his OPS is .823 in low A.  Missing most of the next year, its .871 in the CAL league.  He doesn’t play at all the next 3 years, and gets only 55 PA the year after that.

Then in the majors at age 26 he hits 295/371/559.  What Hamilton has done is about what most would have expected from his decent performance 5 years ago in the low minors, combined with (more significantly) his scouting reports and his being the #1 overall selection.

That he’s doing it after missing 4 years and never competing against AA and AAA is amazing, and makes you wonder how important experience is when you have talent + health + maturity.

Some interesting insight by mgl.

Here’s what I did.  Take all American born players born since 1960, and flag them as to whether they went to college or not.  I excluded Barry Bonds.  Min 200 PA in any season.

Between the ages of 25 and 31, the non-college players hit 18.2 HR per 600 PA in MLB, and the college players hit 16.5 HR per 600 PA.  The walk rates went the other way (53.9 for non-college and 56.6 for college).  The K rates were higher for non-college: 97.0 v 93.6. 

So, we can say that, of those players that made MLB, the average non-college player is slightly better players than college player in their late 20s.

60% of the players went to college.

From age 32 onwards, the college players drastically outhomered the non-college players (19.5 v 17.4, which is quite a reversal).  They also outwalked them (63.5 v 58.9) and matched the K (91.8 to 91.5).

72% of the players went to college.

It seems that mgl’s thesis would hold.  That the guys who didn’t go to college probably are “early bloomers”, whose peak age in MLB will be a bit earlier than college players. 

Fascinating.

I should note that this is a quick study.  I should at some point do “aging hitting patterns” by “college or not”, position (catcher,2b, pitcher), birth country, bb/k ratio in the preceding two years, etc, etc.

MGL’s hypothesis about why players drafted out of high school develop power less than players drafted out of college in the early 20’s is reasonable.

I’m not exactly sure why the effect is there, but it’s there.
My best guess was that year-round training (when you consider instructional leagues, winter ball, etc) for 19-22 year olds drafted out of high school leads to more power developments at this age than the training of a typical college hitter so the HS reach their peak sooner. But the HS group could also contain more early bloomers than the college players to start with… that certainly makes sense. One way to investigate this might involve creating a comparison group of hitters who were drafted early in high school but chose to go on to college.

Does the distribution of birthplace vary for players who played college vs high school?  Year-round players (south, texas, california) are way more advanced than players from the north at age 18.

You know, I’ve looked at regional effects before and haven’t found any significant differences. That doesn’t mean they’re not there, but it’s a tough thing to study.

the HS group could also contain more early bloomers than the college players to start with [chris #7]

This is certainly worth exploring as a contributing factor. Can’t remember where I read this, but (along the lines of MGL’s example in #2 about the 13 year olds), high school athletes tend to be in the higher end of the age range in their grade cohort. It makes sense intuitively that between an equally talented 18.0 chronological year old senior and a 17.5 year old, that the 18 year old will perform better. The aging curve is not linear, and at those ages a small difference in age will be worth more than it is later.

The older players are slightly more “projectable”; could that make them more likely to be drafted in a high round or drafted at all, slightly out of proportion to their innate talent? Are there subtle age differences between the players who go directly to pro ball and those who go to college first? The causation could go both ways. Just as younger players are riskier for pro teams, so also a choice of pro ball over college might be riskier for a younger player who isn’t getting an overwhelming contract offer.

But don’t forget what I showed: from age 25-31, the HS players either outperformed, or was at least the equal to, college players.  But, from the age of 32 onwards, the college players were better than the HS players (even though at that point there were even more college players, meaning that they were being compared to the really best that HS had to offer).  I don’t think it has to do with projectability, since I chose an age range that would really not care how those players performed at the minor, college, or HS levels.

I think the most likely scenario is that the top HS picks are early bloomers, and therefore, have their peaks a bit earlier. 

As someone else mentioned, a great way to test is to look at top HS picks who chose to go to college instead.  Is it the late blooming aspect, or is it the competiton experience of college that helps the college players maintain their edge over the HS players, in their 30s?

I would think that adding body weight as a variable, if that is available at draft time (probably not) or first year in pro ball, would help in the analysis.

There are two types of “early blooming” with respect to developing power (and other skills) in baseball.  One is the player who has a physiological age greater than his chronological age, which is what we typically mean by an early bloomer (starts puberty at 10-12 rather than, say 12-14).  Most of the kids in the LL WS are these kinds of early bloomers.  They literally have the bodies of a typical 14 or 15 year old even though they are only 12 or 13.

The other kind of early bloomer we probably see among high school draft picks are those kids who put on strength and weight at an early age regardless of whether their physiological age is greater than their chronological age.  For example, when I was 12 or 13, I was a typical bloomer, neither early or late, from what I can recall.  But I reamined a scrawny “kid” (no power in basebal) until my late 20’s or early 30’s.  Many other kids I knew, early or late bloomers, deveoped size, weight, strength, power, etc., in their mid to late teens or their early 20’s.

I think it is mostly the latter we are talking about as far as high school draftees are concerned.  By the time you are 18, whether you were a late or early bloomer does not make that much of a difference.  But, for whatever reasons, you see big differences among 18 yo’s in terms of developing their peak weight and strength (and height).

Tango: 
You may want to control for year/environment in your study, just to be sure the college/non-col difference is real.  If the proportion of college players has been growing—which I’m guessing it has—then their age 32+ years will disproportionately fall after the 1993 HR explosion.  That would also account for the odd pattern that the college guys lost their advantage in Ks after age 32 (since K rates have also risen over last 10-15 years).

Following up:  if the proportion of college players has been increasing, it could also be that your college sample has more active players, while non-college has more retired players.  This could mean the non-college sample has played more seasons in their later declining years, while your college 32+ sample is more clustered in the 32-36 range.

Limiting to USA guys born since 1968 (and I also now removed IBB), I have virtually exactly 50% college and 50% non-college, for the age 25-31 performances.  Non-college, college numbers:
HR: 19.0, 18.0
BB: 49.8, 53.3
SO: 102.2, 99.3

And for age 32-35 (so as to avoid the really down years):
HR: 16.9, 19.8
BB: 52.3, 58.6
SO: 89.8, 99.2

I’m also down to 44% of players being non-college.

So, I’m looking at a survivorship bias that should go against college players (only looking at the better HS players), and still the college players trounce.

In order to match the performance levels of college players aged 32-35, I had to get to non-college players aged 29-30:
19.4 HR, 50.8 BB, 98.7 SO. 

It looks to me that there’s a 4-yr peak gap between non-college and college players.  What is significant is that you have 4-years of college.  It’s almost as if guys who go to college “bloom” 4 years later, so they are 4 years behind the non-college players.  And their bodies will simply peak 4 years later anyway.  There’s no extra time savings.

Though, I think I’d want to look at the speed scores (SB, 3B) to see if that’s consistent.

Whether it has to do with experience or more likely, college players being late bloomers, I’m convinced this is something that should be considered in a projection model.

Can you compare the numbers of non-US players to these groups?  A lot of them signed at age 16, so they may be even more of early bloomers than the US high school players.

Here’s the full data for guys born since 1968:

 wOBA      hr600      bb600      so600     country    age    College    n    G


 0.338      23.9      43.7      96.2     CAN    24under    1    3    336


 0.371      24.1      66.5      115.8     CAN    25-31    0    10    1305


 0.386      28.5      59.9      125.7     CAN    25-31    1    6    767


 0.352      25.7      69.0      122.0     CAN    32-35    0    6    672


 0.353      19.6      66.4      107.1     CAN    36over    0    3    330


 0.329      15.6      40.0      99.0     other    24under    0    156    18117


 0.341      17.9      45.4      106.1     other    24under    1    41    4623


 0.345      19.8      44.8      91.5     other    25-31    0    486    58728


 0.336      13.2      54.2      96.6     other    25-31    1    82    10235


 0.359      22.3      48.1      94.6     other    32-35    0    70    8679


 0.346      20.1      66.5      127.8     other    32-35    1    14    1630


 0.327      19.5      50.4      101.3     other    36over    0    9    1097


 0.312      8.9      44.7      99.8     other    36over    1    4    438


 0.337      16.9      48.7      105.7     USA    24under    0    155    18188


 0.337      15.6      48.8      102.7     USA    24under    1    117    13277


 0.344      19.0      49.8      102.2     USA    25-31    0    786    94350


 0.348      18.0      53.4      99.3     USA    25-31    1    773    92444


 0.341      16.9      52.3      89.8     USA    32-35    0    140    16108


 0.350      19.8      58.6      99.2     USA    32-35    1    179    21556


 0.345      21.8      56.0      91.3     USA    36over    0    9    1123


 0.349      19.9      59.4      96.5     USA    36over    1    35    3889

One note of caution: it’s not the same players in each group.  For example, you have 786 player-seasons in the USA/25-31/noncollege group (wOBA of .344) and then 140 in the age 32-35 group (wOBA of .341).

It’s not that the aging was so little, but that only the best are in the latter group, whil the former includes plenty of guys who would have dragged down the 32-35 group if they were allowed to play.

So, be careful!

Also, what is the average age in each age interval?  That might be different for the college and non-college players.

What do the two “others” represent and why are there two groups for Canadians 25-31?