On the reliability of defensive abilities, part 2

And as often happens here at StatSpeak, we return to the curious case of Denny Hocking.  Hocking, the former Minnesota Twins “10th man” is to this day one of my most favorite Sabermetric players.  Not because he was any good, mind you.  In fact, he never… really… did much.  He simply amuses me for the fact that a man with a career .251/.310/.344 line with at-best-average speed could hang around for thirteen seasons.  What was his secret?  He was “versatile.”  Hocking played all 7 non-battery positions in his career, and played all of them… correctly.  But, he’s a symbol of a guy who can extend his career simply by willing to be a jack of all trades (and a master of none) and a symptom of the fact that it’s hard to find someone who can play more than one position in the field.
Really though, how hard is it to play more than one position?  Hocking did it.  If a player is a gifted enough athlete, shouldn’t he be able to perform, no matter where he’s placed on the ball diamond?  Let’s take a look to answer this one.
First off, let’s look to see if there is a correlation in skills within a position.  I ran the OPA! system on all players from 2004-2007 and took all those who had logged at least 450 innings (50 games) at a position.  Then, I looked, position-by-position whether there were inter-correlations among the various skills that a player might show.  That is, I looked to see among first baseman, was “range” OPA! on ground balls correlated with “hands” OPA! on ground balls or “arm” OPA! on ground balls and so on.  (I normed each skill by dividing by the number of chances each player had.)  This serves as a test of whether “defensive ability” is one skill or several.
I found almost nothing.  Just about everything was uncorrelated with everything else.  (For the really statistically super-savvy, I even went so far as to run some factor analytical models to see if I could salvage anything.  If you ever want to see some ugly factor loading plots, call me.  This could be the result of the fact that I have some unstable parameters year-to-year, or it could be that the that fielding skills are unrelated to one another.  If you have no idea what I just said, don’t worry.)
The few significant correlations that I did find were modest at best (< .30), although they did put my mind a little bit at ease on a few things.  At both second and third base, a strong arm was correlated with a few unexpected things.  In both cases, a strong arm meant a good amount of skill at fielding the ball cleanly.  A strong arm also correlated (weakly) with, of all things, range on popups (third base) and range on line drives (second base).  Those might be type I errors, due to running so many tests.  But there’s also a correlation at shortstop between arm OPA! per ground ball and range OPA! per ground ball.  It was modest (r = .209) but it was positive.  I had actually worried that these would be negatively correlated in that a player who had excellent range would get to a lot of balls that others would simply let into left field, but which he would have no chance of actually making a throw on.  This probably does happen, but it looks like a shortstop with good range is more likely to have a good arm rating.  Looks like those gross-motor skills develop together.
Turning the double play at second (r = .689) and short (r = .633) was correlated strongly with a player’s arm ratings on ground balls generally.  That makes sense since both are essentially throws anyway.  In the outfield, almost nothing was correlated with anything else.
This means that fielding skills (things like soft hands, a good arm, etc.) do not all come as one package, but are all component pieces that a player must have.  A player can easily have soft hands but a noodle arm… or soft hands and a good arm.  It’s hard to find a player who is particularly good at all of them, but what we can do is look at which ones are most important at which position.  I calculated a player’s OPA! per inning at the position, and looked at which of the player’s skills were most closely associated with the amount of out probability he was adding.
On the infield, not shockingly, ground ball abilities were most closely associated with OPA!  After all, that’s most of what an infielder does.  But, the parts of the ground ball that were most associated with making outs differed between the positions.  Below, I present the correlations at each position between the skill per GB rating for each position and OPA! per inning.
pos — range — field — arm — catch
1B  — .604   — .670 – .477 — .265
2B  — .581   — .625 – .740 — .005
SS  — .651   — .386 — .741  — (.037)
3B  — .623  — .615  — .724  — (.045)
Line drive range is also significantly correlated at all four positions, but not equally so.  Line drive range is much more important for “basemen” than the shortstop.  Liner range correlated with OPA! per inning at .433 at first base, .490 at second base, .298 at shortstop, and .415 at third.
A first baseman actually benefits from being able to cleanly field a grounder most of all the infielders.  He’s also the only one who really benefits from being able to catch the ball, although he’s pretty much the only one who needs to use that skill on a consistent basis.  The big surprise is that a shortstop booting a few ground balls isn’t the best determinant of his defensive prowess.  A shortstop doesn’t benefit most from soft hands, but from good range and a good arm.  A player who has soft hands, but a lousy arm at another infield position might benefit from a move to first.  A shortstop who has lost some range but still has soft hands might make a good second baseman, and a second baseman who has good range, but some trouble fielding the ball on a grounder might actually benefit from a move to short.  A second baseman and a third baseman seem to share about the same profile for success.
In the outfield, range on flyballs and liners was uniformly important (no shock there), although a center fielder and a right fielder’s arm rating in gunning runners down was also moderately correlated with OPA! per inning (CF, r = .371, RF, r = .334).  Again, not much of a surprise that a right fielder’s arm would come into play, but I don’t know that a center fielder’s arm quite gets the credit it’s due for how important it is.
Are skills portable from one position to another?  For example, if I have a strong arm at second, will it translate into a strong arm at third if I play over there.  To study this further, let’s take a look at the Denny Hocking’s of the world.  I looked for all players in a year who had logged more than 90 innings (10 games) in a season at two different positions between 1995 and 2007.  I ran correlations to see if, for example, range at second base correlated with range at short.
Ground ball range was very portable across the infield.  In fact, other than first base and shortstop, every other combination of positions on the infield had a correlation of better than .74.  Range on line drives and to a lesser extent, popups told a similar story.  If you’re fleet of foot (or leaden of foot) at one position on the infield, you’re going to be similarly fleet (or leaden) elsewhere on the infield.  If a team has a slow infielder, it’s just a matter of figuring out where to hide him so that he will do the least damage.
Hands (not making a fielding error on a ground ball) was moderately consistent (correlations around .35) between first basemen and second baseman, and first basemen and shortstops (although oddly, not between second and short).  Arm ratings on ground balls were somewhat correlated (r = .224) for second and third basemen and catchin throws was consistent across first and second base (r = .448).  So, some skills can be taken with you from position to position, but in general, it’s interesting to see how many correlations were not significant.  In particular, arm ratings did not translate from position to position.  Since we’re measuring players against themselves, arm strength is the same.  Maybe it’s a matter of different types of throws that need to be made.  The throw from third to first is long, but it’s a pretty straight shot.  The throw from second is shorter, but may require the fielder to throw from a much more contorted body position.
However, once you put everything together, total OPA! per ground ball is moderately consistent across most infield positions.  First base and third base are correlated at .360.  Shortstop and second base are correlated at .606.  Performance at third base also correlates fairly well with performance at short (r = .504) and second (r = .568).  The same pattern emerged for line drive OPA!  Because range is so consistent, and it’s the first step in doing anything, it’s going to have a profound effect.
One other finding that might mean something.  When it comes to catching line drives, SS and 2B have a correlation of .315.  However, 1B and 3B have a correlation of -.285.  The angle that a line drive comes off the bat might make a difference, and fielders might be better at some angles than others.  In the middle infield, where the angles are likely more similar, and the fielders play deeper anyway, it’s probably a little easier.
So, considering all of the above, if a player is a good fielder at one infield position, would he thrive at another?  After all, it doesn’t matter how it’s done, you just have to produce outs.  Correlations of total infield OPA! per inning at the various positions are presented below.
pos — 1b  –  2b  –  3b  –  ss
1b  — 1.0 — .278 — .347 — .101
2b –  xx  — 1.0  — .504 — .528
3b –  xx  — xx   — 1.0  –  .434
ss  –  xx  — xx  –  xx  –  1.0
Correlations are pretty good between second, short, and third, and if a first baseman is going to play anywhere else, it’s probably going to be third base.  That’s basically the pattern that plays out in actual bench construction.
In the outfield, things are much more murky.  I couldn’t find a single correlation worth reporting in which a skill at one outfield position carried over to another.  Nothing.  In part one, I found that most of the outfield abilities that I measured were not very stable from year to year.  It seems that outfield defense is a little bit like batting average on balls in play.  Sure, there’s some skill to the craft, but the outfield is a big place, and over the course of a season, there’s not enough to measure to really get a good idea on how good a fielder is.  At least in my system.
I have to leave open the thought that while my system seems to work well for infielders, the lack of location data is crippling me.  I can imagine that if I had more concrete location data to use (rather than I know that the ball was hit into the general vicinity of the left fielder), I’d be able to sharpen things a bit better.  As of right now, I have to say that I’m not thrilled with the results that OPA! is giving me in the outfield.  The other possibility is that outfield defense is much more luck than we’d like to give it credit for.  At this point, I’d have to say it’s an open question.
So, was Denny Hocking superman?  His range rankings on the infield show up as usually pretty high, so he had the most basic skill, and the one that would help him the most.  Hocking got pretty high marks at 2b, ss, and 3b in the seasons that he played there on most of the other skills that OPA! measures.  So, he managed to master several different and seemingly un-related skills.  Not a bad thing to be able to say.  And he can say he was a major leaguer for 13 more years than I was.  So, reader, you now know a little bit more about OPA! and a lot more about what talent it takes to be a utility infielder.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: