THE BOOK--Playing The Percentages In Baseball

Ugh. With a couple of exceptions, Huckabay’s work has been utter crap (especially that sensationalist ‘sabermetrics is dead’ post from awhile back.) It’s hard to take anything he says seriously now days.
My favorite part is the condition that the numbers line up with his visual perception. If you disregard any metric that conflicts with your own preconceived notions, then there’s no point in using stats at all.

ZR (zone rating) is outs made per “opportunity”.

RF (range factor) is outs made per “opportunity”.

In ZR, “opportunity” is the number of balls that are hit in your “vicinity”.  It’s a not bad way to measure opportunity.  Obviously, balls hit closer to your vicinity is better than those on the edges.  Even better are balls hit outside the edges that don’t count in the denominator.  It’s a decent metric.  Anyone who tells you otherwise should stop telling you otherwise.

In RF, “opportunity” is the number of innings played.  Now, if you have ZR, why in the world would you want RF?

The comparison is exactly the same as times on base divided by PA, and times on base divided by games played.  Now, if you don’t have PA, games played is good.  But, once you have PA, games played is useless.

So, to use RF, once you have ZR, is silly.

Now, you can argue that you like to apply adjustments to RF (BP’s FRAA, for example).  Fine.  Just make those adjustments to ZR instead (Dan Fox’s Simple Field Runs, or Rally’s TotalZone or my WOWY).

Finally, the author’s sole observation?  Isn’t the article screaming for a Wisdom of Crowd approach, instead of one person’s observation?  Indeed, the Fans’ Scouting Report was ready-made for this. 

I argued last year, without ever seeing Ryan Braun field a single ball, that Braun should be moved to the corner OF immediately, on the basis of UZR and the Fans’ Scouting.  Nate Silver, kind enough to reference the Fans’ Scouting Report, said similarly soon after.  And the Brewers move him there in the off-season.

If only one Brewer fan had told me of Braun, what am I supposed to do?  I have to ignore him, because how they heck do I know if this guy knows anything.  And even if he did, how do I know he knows much.  But, you put 20 or 30 fans together, all hardcore fans, then we’re talking.  The relationship between Fans’ Scouting and UZR is STRONGER than between BP’s FRAA and UZR (according to reseach by jinaz).  That’s f-cking impressive, isn’t it?

BP is fond of saying that they like to talk about baseball, and not argue with analysts.  Maybe they don’t need to argue with analysts, as much as simply listening to them.
http://www.baseballprospectus.com/article.php?articleid=7963

Sometimes, we’re criticized at Baseball Prospectus for not responding to outside critiques. It’s a conscious decision, made at the management level—we’d rather talk about baseball than about ourselves or about our colleagues in the world of baseball analysis. While we don’t respond to every broadside in every blog, we also by and large don’t spend much time critiquing others’ work…

It’s a b.s. position that allows for no accountability.  Except in their comments section, and I hope their readers alert them as much.

Btw, MGL, can you point a link to the article in question?  I can’t find it.  Was it pulled off the site?

It’s from a really old article:

(Click name)

Last paragraph.

The one reason at least in theory that one might prefer RF to ZR (or variations on ZR) is that RF is purely objective.  The numerator in RF is the number of times a fielder participates in an out, and the denominator is the number of innings the fielder is on the field.  ZR and its ilk, in contrast, import a level of subjectivity that arrives from outside the confines of the balls/strikes/bases/runs/outs of the game—an onlooker has to decide whether a ball fell within a somewhat subjectively chosen zone of opportunity that is not itself defined with the baseball rules. 

That doesn’t mean RF is better than ZR for evalauting a fielder’s contributions to team defense—to the contrary, ZR approaches are certainly more reliable than unadjusted RF.  But to the extent ZR is better that’s because it works better in practice, not because as a logical matter it does the same thing RF does only more inclusively.  In fact ZR is conceptually different than RF.

As to fielding evalaution and the wisdom of crowds, check, for example, the book “Nudge” by Sunstein and Thaler, on the mass irrationalism of crowds.  Among other things, most people will often see what they expect to see rather than what is actually in front of them, which means relying on the wisdom of crowds risks merely reinforcing popular prejudice.  Ryan Braun was a disaster in the infield by every possible measure—even RF as I remember.  To rely on that example as a confirmation of the wisdom of crowds is setting the lowest bar possible.  And as to the correlation of UZR and the Fans Scounting Report, is it possible that many of the same people responding to your Scouting Report are also checking the same stats you find persuasive and then seeing what they expect to see on the field?

Ay-yay-yay.

Perhaps people are not aware, but when BP rebuilds its home page every day, it does through some crazy sh-t, such that you’ll see some really old articles on the front page.  Almost like a time machine.  When I went to BP earlier, I saw a Zumsteg article on the front page.

My guess is that MGL saw the Huckster’s article, presumed it was today’s, since he saw it on the front page, and commented appropriately.

I’m definitely going to give Huck a pass on this, as it was written eight years ago.  Other than murderers, all other criminals have the luxury of the statute of limitations and baseball analysis is no different.  As long as he doesn’t stand by his article and repents.

My Fans Scouting Report point now has no bearing at all on Huck’s article, since I started the thing six years ago.

All in all, I won’t say it’s phony outrage, but it is… Mistimed Outrage.

Don’t worry MGL, BP has fooled better men than we with their front page time machine trick.

birt: I wasn’t using Braun as the example to prove anything.  n=1 proves nothing.

And no, the voters don’t look at UZR.  Most of the voters come from outside of here, probably never heard of UZR, or if they did, will barely remember anything of it.  And MGL doesn’t even publish it regularly (and not at all in the last 1.5 years).  Furthermore, people evaluate players on 7 categories, not one overall category.

Ichiro lead the league in Fans’ every year, and UZR no likely.  I know, I know, n=1.  The point is fans will vote on their perceptions, and not on UZR or RF or ZR or whatever.

The worst Huckaby article was that silly “Value of Jack Cust” one, where he argued that Jack Cust represented replacement level--since, you know, anyone could have had him--and figured out a lot of people’s “VOJC” using his performance in 2007 as the baseline.  Of course, using this idea you manage to conclude that, say, Mike Lowell was barely above replacement last year.  I mean, come on.

Wow, that definitely fooled me!  Sorry about that.

The last few days, every time I tried to click on a Neyer “NL closers not that good” (or something like that) article, linked from the baseball home page, it took me to another blogger’s article and there is, to this day, no sign of Neyer’s “closer” article.

Anyone have the same problem?

Tango, I agree with birtelcom on that issue.  It’s my gut feeling, I can only back it up with my own experience [I know, n=1 right ] but it’s true.

When I read that Lowell doesn’t have the range he used to, when I read that Lugo isn’t an above-average SS, that knowledge goes with me when I watch games.  And it makes me pick out instances where it’s true (confirmation bias).  I know not all of the ballots come from readers of this blog, but USS Mariner and Hardball Times readers do, I think, have some sense of these types of fielding metrics, and who falls where.

Not to mention, the people who see value in the data of your fans scouting report are more likely to fill out the ballots than the readers who say, “How can you say Jeter is a bad SS?!?!  Do you even watch the games?!?!”

This is easy enough to prove: rookies.  If I can show the relationship between rookies Fans’ and UZR, will this satisfy everyone?

I’ll add that to my list.

Birtlecom, there are some things that “crowds” are good at and others that they are not (and everything in between).  While I agree that in evaluating players without the luxury of using stats, Fans will fall prey to a lot of bias (thus making their judgment poor in some respects), as it turns out, Fans appear to be quite good at evaluating defense, given a decent set-up (finding enough serious fans, etc.).  That does not surprise me.  To some extent, asking serious fans who the best fielders are is like asking them who the fastest players are or who have the strongest arms, etc., especially when Tango is not asking them who is good or bad overall - rather he is asking them to rate individual components of defense, which, like speed or arm strength, is not all that hard to do.

I don’t really get your point about ZR and RF.  There is no comparison between the two, and I agree with Tango that once you have ZR, RF adds virtually nothing, although I suppose you could, if pressed, come up with some way to combine the two to make the combination a tad better than ZR alone (maybe 99% ZR and 1% RF). Just because someone is deciding what zone a ball falls in does not mean that there is a subjective element to it, except perhaps in a literal sense.  A subjective element would be when someone is deciding something that has to do with the skill you are trying to measure.  As long as there is no bias (with respect to the player) among the stringers as far as whether a ball is in zone A or B, it doesn’t really make that much difference whether the stringers get it exactly right (which they won’t of course).  The fact that RF is “clean” and ZR is not makes no difference whatsoever in terms of comparing them.  The bottom line is that one is 100 (or 10, or 8, or whatever) times is better than the other and with one, you don’t need the other (it adds nothing to the discussion).

Does anyone know if Bill James still uses RF in his Win Shares? I certainly hope not, but Mr. James is also one that doesn’t follow much current/newer research or developments.

RF, on its own, is useless.  But, you can have a version of an adjusted RF that would match or exceed unadjusted ZR.

If you account for the number of BIP, instead of innings, that improves RF.  If you know the split in number of batters faced LHH and RHH that improves RF substantially.  If you know the GB/FB tendency of the pitchers, the improves RF.  If you know how fast the batters are, how strong they are, this improves RF.  If you know how many runners are on base, this improves RF.  And if you know the park, this improves RF alot.

If you know all that for RF, but don’t know any of that for ZR, then adjusted RF might exceed unadjusted ZR.

But, if you know all that for RF, you MUST know all that for ZR too.  So, again, ZR trumps RF, if both use the same adjustments, simply because ZR knows one thing more than RF: the general area where the ball was hit.  RF will infer it, while ZR knows it.

Any fielding system that has subjective measures in it such as how hard the ball is it, and where exactly it is hit, has the potential for flaws. There will be a variance between observers as to what zone something falls in, and how hard a ball is hit. The potential for variance based on the observers is huge. There is an old saying of garbage in garbage out. So an excellent subjective system with bad data can be worse than a poor objective system.

Let’s stick with basic ZR and basic RF, which I’ll define as: outs made per opp.

They both have identical numerators, in my definition.  The denominator is “balls in vicinity” for ZR and “all balls in play” for RF (this definition for RF is better than innings played, which is based only on outs made, including strikeouts, which is pretty silly).

Let’s say that there are around 30 balls in play per game, of which some 8 balls are hit between the thirdbaseman and secondbaseman.

For some SS, there may be 29 BIP and another there may be 31 BIP per game.  Maybe 28 to 32 at the extreme.  But, for those 30 BIP, we have no idea if you have a RHP staff or a FB staff or whathaveyou.

But, what if I tell you that, based on where the 3B and 2B were standing, that there were 8 balls hit between them when Jeter is at SS and 6 balls hit between them when John McDonald is at SS.  Now, this may be “subjective”, but my 6yr old could have told you that information.  Isn’t it better to know the 8 and 6 numbers, than the 28 and 32 numbers, respectively?

I disagree that the potential for variance can be huge, insofar as location of ball hit.  (The speed of batted ball is irrelevant to ZR, and should not be considered in this discussion.) Only if you make the zone very tiny will the impact of the error variance be large.

For example, we can both watch the same play, and how much far off can we possibly be in judging the play?  Peter Jensen did a great article on this where he actually looked at it.  I forgot what he said, and I’ll post a link later for it.  But I will guess he said 1 SD = 15ft in the OF.  I’ll guess 1 SD = 5ft in the infield.

If I make the “balls in vicinity” as the area between the thirdbaseman and secondbaseman, that’s some, what, 90 feet of distance?  Furthermore, by using actual fixed landmarks (the actual position of the secondbaseman and thirdbaseman), I think it would be pretty darn rare for two observers to say that a play did or did not go to the left of the thirdbaseman.  Like I said, because of the wide range here, this version of vicinity is far better than “balls in play”, which includes balls in the OF or at 1B.

Now, the actual ZR uses a smaller range, which is probably some 30 feet or so (of some fixed part of the infield between 3B and 2B).  So, there is a potential here for missing balls on the edges between two observers.  But again, what is better to know: the number of balls in the vicinity of a range of 30 feet estimated, or the actual number of balls hit anywhere in the park?  The observer variance error simply can’t make the 30 feet estimate worse.

There is a point where the observer variance error will make ZR worse than RF (as I’ve defined RF, which is based on BIP, and not based on the silly business of total outs, including strikeouts).

And it is an interesting exercise to figure that out.  But, I don’t think that the current form of ZR is anywhere close to that.

So, there is subjective and there is subjective.  If my 6yr old can figure it out, that subjective is really objective.

Here’s the Peter link:

http://www.hardballtimes.com/main/article/is-seeing-believing/

Tracking only FB, 1 SD = 10 or 12 feet, or 1 SD = 2 or 3 degrees.

Presuming 1 SD = 3.6 degrees for GB, and presuming balls hit about 100 feet away, then that would imply:
circumference of circle = 2*PI*100 = 628 feet
And 3.6 degrees of a 360 degree circle, means 1 SD = 6.3 feet in the infield.  And I think I was being pretty generous.

Tango: Unfortunately, one source of semi-systematic error is exactly what you have stated: treating the thirdbaseman and shortstop as fixed points.  While the bases are, the players are not. 

However, people have a tendency to percieve them as more-or-less fixed, especially if their positioning isn’t obviously changed, while in fact they’re likely moving by ten feet or so (especially the SS).

However, I wholeheartedly agree with the statement that given ZR RF is useless.  It’d be nice to get ‘vector’ information out of the MLB camera setup: I assume that’s part of what’s planned for hit f/x?

The main practical reason to try to adjust RF to make it as useful as ZR is for historical context—Zone Rating type evaluations go back only a few years, RF numbers go back as far as there are box scores. ZR is nice for comparing 21st century players to other 21st century players, but for anything relating to the 20th century (including comparing current players to 20th century players) were pretty much stuck with RF and box score data adjustments to it.

To some extent, I think tango’s discussion exaggerates the scope of the difference between RF and ZR.  Yes, in theory RF includes all BIP.  Butwhen you compare the RFs of two shortstops a common base minimum of BIP that are, say, fly balls to the OF will wash out of the comparison.  The source of error is in the scope of the difference, team to team, of BIP that are fly balls to the OF, rather than the entire universe of fly balls to the OF. 

I certainly do not question that well-executed ZR systems are superior to RF systems in judging individual fielder contributions to run prevention.  I agree that the scope for static in the numbers in ZR systems is much lower than in RF systems.  My original point was simply that ZR systems do introduce a form of potential static that does not exist in RF systems (the judgments that go into both assigning zones and then assinging BIP that are close to zone boundaries), a new form of static that is different in kind (while lesser in scope)than the types of static that make RF such a flawed (and thus adjustment-demanding) stat.

"While the bases are, the players are not.  “

Well, at the point of contact, all matter in the universe is static.  In any case, you can simply count the number of balls hit between 3B and 2B instead if you like.  What we care about, for ZR, is the distribution of balls.

Let me point out a couple of things about RF.  First off, “games” is what is used.  But, if we have inning played, that is what has been used.  No one, however, does what I say, and use balls in play.  That’s what you SHOULD use.  So, I’m trying to make RF better by giving it a better denominator.

For ZR, you should use the area that a SS would normally field a ball (say some 30 or 40 feet wide).  Instead, I’m proposing simply counting the number of balls that are 90 feet apart (be it base-to-base, or fielder-to-fielder).  So, I’m trying to make ZR less relevant by giving it more noise in the denominator, balls that are “for sure” the thirdbaseman’s or secondbaseman’s.

Even under that scenario, I’d still prefer ZR to RF.  Indeed, if you just used “groundballs” anywhere in the infield, I would prefer THAT in the denominator to all balls in play.  That is, agreeing that we will keep the numerator the same, the best denominator will always be the one that removes flyballs for the SS.  If you can remove some groundballs, all the better.

It is only when you shrink the zone of responsibility so short that the judgement of the scorer because an issue can we say that maybe RF is better than ZR.

That said, that’s why I like WOWY.  I don’t use estimates.  What I do introduce is the identity of the batter, pitcher, and park.  With those parameters, it gives me an inference to the distribution of balls hit that works better than RF, and possibly (probably?) better than ZR.

If the variance for ground balls is 6 feet, and the zone is 30 feet, is that a 20% error?

Isnt 6 feet in the infield a lot of ground to cover? That’s 2 steps.

In the OF, 10 feet, 3 steps - the difference between a 5 yard and 8 yard run.

The number of balls hit between the 2b and 3b is not subjective.

If 8 balls are hit, and the KC thirdbasemen gets to 2 of them and the NY thirdbasemen doesn’t get to them, then 6 balls are hit between the KC 2b and 3b and 8 are hit between the NY 2b and 3b.

Or, if the NY 3b guards the line more to cut off possible extra base hits, then there is a larger space.

Or, if the 2b is closer to 2b due to possible base stealer.

Fine, then just use the number of balls hit between the thirdbase and the secondbase.  The point is to figure out a general distribution of balls hit.  You don’t even have to get exact numbers of fieldable balls by the shortstop.  RF gives you ALL balls in play.  I’m saying to remove all balls hit in the air, and remove all balls to the left of either second base or the secondbasemen.

Just as it doesn’t matter if the SS can field a ball hit to rightfield (as Range Factor would count in the opps), I’m saying it doesn’t matter if the SS could actually field the ball hit to the thirdbaseman.  But, knowing the number of balls hit between 2B and 3B (the bases or the players) will give you a better inference as to the number of fieldable balls for the SS, than the inference based on all balls in play.

I hope that’s clear.  It’s not important to know how many truly fieldable balls the SS actually had.  All we care about is to remove as much of the noise as possible.  We don’t have to get 100% signal here.  Range Factor has ALOT of noise.  All the flyballs are noise.  And they don’t even out, especially if you have a FB-heavy or FB-light staff.

Zone Rating attempts to remove noise.  The key is not to remove so much noise to the point that you are also removing some of the signals.  That’s when you get into the danger point.

----"Zone Rating attempts to remove noise. The key is not to remove so much noise to the point that you are also removing some of the signals.”

How much of a chance is there that this is what is happening in UZR?

David: there’s a good chance that UZR is missing the boat on some players.  This is why what UZR should do is present two versions: the one where it is the most objective-like, and one where all the adjustments are applied.

We cannot tell if there are any systematic biases in UZR.  We are presuming they are random biases, but we cannot accept that on its face.

The same can be said for park adjusting offensive stats.  As I’ve said in the past, Barry Bonds in SF’s home park has hit as many HR as he did on the road, over the same time period.  But, if you look at all other LHH, the SF park severely depresses HR.  I think it is far more likely to presume that Bonds hits monster shots everywhere, and so, the “almost HR” in SF that are “just over it HR” elsewhere affects other hitters far more than Bonds.  Park adjustments won’t necessarily capture this: it is biased with respect to LHH superduper power hitters.

I’m definitely not cutting UZR any slack in this respect.