THE BOOK--Playing The Percentages In Baseball

The leveraging of speed to park is very interesting.  Good stuff.

***

You had very few CF games in the first two charts.  How about doing speed scores by position?  For example, if the mean speed score at CF was +0.7 SD, then make the fast runners those at +1.7 SD or higher and the slow ones at -0.3 SD or lower (or whatever process you used).

***

Mike Cameron is one of your 5 CF in large parks.  I didn’t look at the other 4 parks, but it’s certainly likely that the CF are not randomly distributed among the parks.

***

Here are the Fans’ overall position-specific evaluation of the top 10 / bottom 10 (by speed score) for each position (there’s about 51 players at each OF position):
LF: 72 points / 25 points (gap of 47 points = 25 runs)
CF: 79 / 51 (gap of 28 pts = 20 runs)
RF: 68 / 28 (gap of 40 pts = 21 runs)

Note that unlike MGL, I have an equal number of players in each top/bottom group, which I think, in the position-centric comps, makes a bit more sense.

Since the two Speed components makes up 55% of the evaluation, I’m not surprised at all by the results.  So, if you have two guys who have the same sample UZR (say +10), and one is super fast and the other is super slow, their regression points are far different.  The +10 fast guy gets regressed to the mean (+10), so he’s a (likely) true +10.  The +10 slow guy gets regressed toward the mean of -10, so he’s likely a true +0 or something.

***

Btw, to convert to position-neutral UZR:
(cornerOF - 9) * 4/3 = CF

The 4/3 is required, because CF gets 4/3 more opps.  The “-9” is solved by looking at players who play dual positions.  MGL, you can figure this out with your current dataset if you like.  I’m quite sure you’ll get something very close to -9.

Is the -9 an average between RF and LF? I am pretty sure that the average LF and RF are of difference neutral defensive quality, if no other reason than you need a good arm in RF (so that a decent fielding corner OF with no arm gets forced to play LF).

My article was admitedly based on fairly quick and sloppy research.

For one thing, I should use more years of data.  For another, as you say, I should break up the speed categories better.

I should also break up the OF into 3 sections and use small/large sections rather than the whole OF, as I mention in the article.  If a park has a small LF but average RF and CF and is small on the average (in total square feet), then what is the point of including all OF’ers in that park?  Only the LF’er should be studied of course.

Anyway, if I have the time, I’ll work some more on it.  I think the ramifications are interesting and useful.  As Tango gives in his example, and as I mentioned, it allows us to regress the sample UZR’s to very different numbers depending on a player’s speed score.  And since speed scores are quite accurate, it enables us to get a much better handle on a player’s defense even with limited samples of data.  I have ALWAYS proclaimed, that at least for the OF, that it is unlikely (not impossible of course) that a slow player is a good OF’er no matter what his UZR says, and vice versa.  I think this confirms that notion, more or less.

MGL:  does UZR take account of DP proficiency for 2B and SSs?  And if not, do you think including DPs would increase the importance of speed for those positions?

Right, it would be an average.

The gap between LF / RF (range plus arm) is fairly close to zero.  The RF doesn’t need to be ahead simply because he has a better arm, since the guys with the better arms usually have worse legs and you find those guys in LF. 

Obviously, year-to-year, that’s not necessarily true.  According to the Fans, 2007 has a huge advantage for the RF.  Using the position-specific Fans’ Runs, the average RF is +3.5 runs ahead of the average LF (range + arm).

MGL, can you correlate your speed numbers to the “first step” and “speed” numbers found here:
http://spreadsheets.google.com/pub?key=pkimQBCeCjbjEqvnXANBEPQ

There are some good reasons to go with observation over performance numbers:
1. sample size of performance numbers… we don’t need a large sample of viewings to know how fast Willy Taveras is

2. the applicability of speed to fielding… a guy can be fast, but if he doesn’t have a good first step, that doesn’t help him much

The good reason to not go with the Fans:
a. observational bias (which is the same as #1 above)

Clearly, we’d prefer both, to some extent.

***

I agree about the “shape” of the OF, like Yankee Stadium.

***

I agree with you, insofar as the OF goes, that speed is critical, as I’ve got it counting for 55% of their score.  The breadth of skills required is smaller in the OF than the IF, and of those skills, they are the most readily evaluated with visual observation.

Waht I’d like to see are the inter- and intra- position speed differences, relative to fielding ability.

IOW, the avg CF and SS are both fast. But maybe there is a very noticeable difference between the (relatively) fast and slow CF in UZR, suggesting that speed is the primary ingredient in good fielding at that position.

But maybe there is little difference in the speed of good SS (in UZR) vs bad SS, suggesting that, among the players who are fast enough to be put at SS, it is the other skills which primarily determine the quality of glove.

That’s what I would expect, anyway.

UZR does not include DP proficiency.  When I calculate DP runs, they come out very small.  I give equal weight to the thrower and the pivot man, which may or may not be right.  We are talking about 1 or 2 runs a year at most.  So I don’t think it is going to make a difference as far as the correlations go.

Because I am not controlling for other variables in the regressions, it could be that speed correlates more than I found for SS (and perhaps other positions).  Imagine that managers put fast players at SS who are otherwise not so good at SS.  It would screw up the correlations, right?

And, as I said in the article, there is some evidence that that syndrome is true - that managers put fast players in defensive positions that require speed, but that some of those fast players lack the skills necessary to play those positions.  Remember in the article I found that the best SS (and 3B and LF) are the average speed ones, suggesting that average speed players need to be really good at other skills required to play SS, whereas slow players are not going to be too good no matter what other skills they possess and fast players are lacking some of those other skills but are put at SS nonetheless.  IOW, scouts are inefficiently putting players at various positions, perhaps putting too may slow ones at SS (or some other positions) and too many fast ones and not enough average ones, for the reasons speculated above.

Tango, the correlation coefficients for 390 players who matched up from your database and mine are:

This time I computed the speed scores to the nearest 10th of a run.  Before, they were just 1-5.  The average speed score was 2.85 for 1106 players, the SD was .85.

first step/speed score = .268
speed/speed score = .615

Also, here are the average speed scores (1-5) for each position:

3 2.5
4 3.4
5 3.1
6 3.4
7 3.2
8 3.6
9 3.2