THE BOOK--Playing The Percentages In Baseball

So, every now and then, UZR misses one.  We all see Teixeira.  He’s played on several teams already (Angels, Braves, Rangers, Yankees), and the Fans of each think he’s a well above-average fielder.  Unless he’s happening to hoodwink everyone, I’m fairly comfortable calling him an above-average fielder, and that UZR is missing something on him.

Could someone explain how the Teiexeira-defense example is evidence of stat (UZR) being incorrect? Or, in reverse, how the example shows the Fans to be more accurate than a stat?

Here’s one thing I’ve recently discovered, and it completely and totally baffles me.

A lot of people don’t seem to understand that defense, like offense, is prone to flukes and slumps. Many can’t wrap their heads around the idea that a bad defensive year from a great defensive player is akin to a great offensive player slumping for a couple months. When you see how someone like Adrian Beltre had a bad 2007 (Fangraphs), you can either assume the system is broken or, maybe, the guy had a few balls bounce the wrong way on him.

JD

Great question about whether defense can slump, but I am not sure that it can. There are bouts of unlike with defense (sun, bad bounce), lack of focus (mental fart), and physical limitations (age leads to range, injury, etc.)

Can defense really slump? If you believe defense is effort and focus, which I do, then you should perform up to your capabilities. Maybe this is more of an NBA debate (defense never slumps in the NBA if you want to defend), but I think you can make the argument for baseball as well.

You could make an argument, but it would be a very weak one with no proof and little logical barrings. 

What is defense at it’s core?  It’s speed, reactions, arm and surehandedness.  Speed probably doesn’t go into a slump, but you don’t think that reactions do?  You’ve never seen a player take a good route to a ball one day, and take a poorer route to that same ball later in the year?  I’ve seen Pujols, one of the best defenders in the game, miss plenty of balls that he usually gets to.  That’s just reactions.  A player’s throws aren’t always on the money, and players will sometimes drop balls they’ve gotten to as well. 

In short, there is no reason to expect that defense isn’t effected by random variation (slumps and hot streaks).  And when you have a stat like UZR that backs that up, then you have to conclude that defense does in fact slump.

I’ll agree with Mike Silva on this one. I’m still fairly skeptical about the bad defensive year explanation that is commonly tossed out. I still think that defensive play is a lot more consistent than offensive play just as he does.

I find proponents of UZR to reach for explanations too often instead of just admitting that UZR might have been off base for that particular year. Instead, we have people trying to justify that Texeira just looked like he was playing good defense because he might have been falling down a lot while making plays that made him look good.

I don’t know how many times I saw that this year and it is just a stupid argument.

Texeira might not have been as good in the field last year as he was in the past for any number of reasons. But it is just as likely that the UZR number just was not very good that particular year.

It seems to me that Tango and MGL are perfectly willing to admit that sort of thing while others try to justify the number too often.

The numbers cannot be taken literally but there are lots of people that want to take them that way and try to justify something that doesn’t need to be justified.

Question for E-6 and Mike.  What are you basing this on?  Why do you think that defense isn’t effected by random variation?

Primarily, because fielding is a lot easier than hitting. They make lots of plays and do it very consistently. I just don’t think there is as much variance. That’s not to say that they are going to be exactly equally effective each and every year.

Fielders are successful 90% of the time whereas hitters are successful 30% of the time. I think it is a little silly to think that fielding varies just like hitting.

Mike: Even if you truly believe that defense cannot slump, i.e. a player reacts nearly the same to all situations, there are other factors that could affect his getting to balls. Some should penalize fielder such as initial positioning, some maybe not, speed and curvature off the ball off the bat.

Do you have any proof to back that up?  Is there any reason we should take your opinion seriously?  Basically, the only thing you said to back up your opinion was the “fielding is easier than hitting”.  I’m not sure if A) that’s even true, or B) if that’s relevant to why their would be less variance in fielding.

Well, I think we know that there are tons of minor league OFs that can play major league defense. The primary problem they have is in hitting the ball as well as Bobby Abreu.

I don’t have any more proof to show that it doesn’t fluctuate than you do to show that it does.

I’m not someone that thinks that UZR is not a good statistic though. I just doubt the huge fluctuations.

And if you look at the different fielding metrics, they all report different numbers and many times it can be more than 10 runs for the same player in the same season. So obviously there is a lot of wiggle room in the metrics.

I’m not sure how valid this argument is since I’ve never played at the highest levels, but I think anybody who has played the game at even the college level will tell you that there are always balls they “should have had.” And there are balls caught that players have no idea how they got to them. I know for a fact I’ve gone through “defensive slumps” (only in the sense that, for a stretch of time and for whatever reason, I was not as good defensively as I usually am). I guess I just don’t see how anybody can think a player is always on his game in the field, but no at the plate. Sure, hitting is harder, but once you’re in the majors the level of difficulty doesn’t really change (outside of pitcher to pitcher and park differences, neither of which are usually blamed for bad stretches).

The difference between good and bad fielders in the majors might be smaller than the difference between good and bad hitters (is this true?), but that doesn’t prove a guy just can’t have a bad year with the glove.

The large variations between seasons are more comparable to “off years” rather than “slumps.” If a fielder is slightly worse over a full season, we probably won’t notice, just as you couldn’t tell the difference between a .300 hitter and a .260 hitter without statistics.

Also fans base many of their estimates of defensive value on a players’ athletic ability.  But couldn’t this cause problems with 1B, because so much of 1B does NOT depend on athletic ability?  What if Teixeira positions himself badly before each play?  He’s still very athletic and looks like a good fielder, but he will miss some balls that other 1B’s get to because of their better positioning.

Or what if UZR just screwed it up. TZ has him saving 7 runs last year. Teixeira plays a pretty mean 1B.

I don’t think there is any doubt that players play better in some years than others in the field. The question is just how much variance is there? I can see 5 runs or maybe even 10. But 20? Not too sure about that. I’m open-minded about it but still skeptical.

And just how accurate are fielding metrics and how much do the results vary? Can runs saved be over/understated by 10 runs in a year. From what I’ve seen that is highly likely in some cases.

If so, then how much variance can simply be attributed to fielding metrics? And why are you so hesitant to acknowledge it? Is plus or minus 5 runs even meaningful from year to year?

When you can have a 10 to 20 run difference in UZR and Total Zone for the same player in the same season, that’s a heck of a margin for metrics that are considered by many to be so accurate. And that is not uncommon.

So even if you had consistent performance from the fielder, it seems to me that you would have inconsistent results from the metrics. If not, you would not have these kind of differences between the fielding metrics for the same players in the same season.

If the metrics are highly accurate, they should pretty much agree across the board from player to player.

That doesn’t mean that they aren’t worthwhile. They are the best we have right now. But they can’t be taken literally as the actual number of runs saved or given up. They lead you in the right direction in determining the quality of a fielder.

It seems to me that large variances are at least open to question at this point. You can probably flip a coin as to which is the cause from year to year. And in some cases, it’s the metric and not the fielder.

At the top of the thread Tango says that Andruw Jones’ UZR differs by 100 runs STATS vs. BIS over 7 years, so in extreme cases you can attribute 15 runs to stringer bias.  In most cases it’s far less; for instance I think IFs have less variation than OF’s, and part of the reason that Jones had such a spread was that he always called off the SS and 2B on easy popups (I think there was a thread about this a while ago).

its pretty obvious to me that defense slumps, in baseball and in basketball.  it just doesnt slump as much as it does with hitting.  there is SO much chance involved in hitting that the amount of ‘slump’ would be atypical of most other athletic events.  a sprinter can have a slump.  the defense of a baseball player or basketball player can likewise slump.  to what degree and in comparison to what is what stats may hopefully help explain.

Traditionalist baseball writers don’t roll there eyes at that ‘batting average’ metric if a career .300 hitter has a batting average of .250 during the first two months of a season.

But if a corner infielder’s UZR moves from +4 to -2 in subsequent seasons, then the metric is clearly too subjective and “up and down” to be trusted.

There is a lot of noise affecting both metrics in these examples. But scouts and middle aged sportswriters saw batting average on the back of a baseball cards long ago, so it lives by a different standard.

The question is just how much variance is there? I can see 5 runs or maybe even 10. But 20? Not too sure about that. I’m open-minded about it but still skeptical.

wRAA, 2006 to 2009, various hitters:

Joe Mauer: +33, +9, +26, +55
Alex Rodriguez: +34, +70, +42, +34
Derek Jeter: +42, +22, +8, +37
Michael Young: +13, +9, +1, +27
Aubrey Huff: +6, +3, +32, -16

Are you skeptical of wRAA? These are established major league veterans with long track records, and the variance in their offensive performance is +30 to +40 runs from one season to another.  These guys weren’t cherry picked - I just grabbed five names off the top of my head. 

This idea that offensive variance is totally normal and acceptable while defensive variance is a flaw of the metric is bizarre.

I suspect the perceived difference in variation between offense and defense is entirely linked to the rate of converting an offensive opportunity to success (~40% in the very best of players) to the same conversion rate of defensive opportunities (~90% in the very worst of players).

Actually measuring the differences in players is both relative and “on the margins” for both offensive and defensive stats.  Both should be susceptible to random variation, but it will be more obvious to a casual observer when the success rate is low as compared to when it is very high.

Paul, where do you get 90%?  70% of balls in play are turned into outs.

33% of PA make the batter reach base.

Looks like the same thing to me…

And if you take out the “gimmes” on the fielding side, then it’s more like 30% to 50% are balls in play are turned into outs.

I don’t buy this “90%” business… it’s a thought process derived from errors and fielding percentages.

Side question: since you are fond of the word, Tango, have you read “On Bullshit” by Harry Frankfurt?  It’s very good.

I never read that.  My admiration for the word came from the card game that we played in college ALL.THE.TIME.  Plus my barely-english speaking mother-in-law likes to use the word too.  It’s a good word.  Direct, no ambiguity, without it being too crass.

With batting, we have a clearly defined hsitorical record of hits, at bats, etc. Of course we have cheap hits and caught line drives, but fater a few hundred plate appearances things start to get reliable.

In fielding, first off we have an incomplete record. Maybe Hit f/x will finally give us the enough of a historical record. It’s just that the current record keeping doesn’t even tell us which fielder had the best chance to make a play on each batted ball. Therefor, we are forced into probabilistic models, basically taking the observed performance and comparing it with what’s expected, based on a theoretic model. mgl’s model that he uses in UZR isn’t quite the same one I use. Thus, we have more variability in the final numbers.

I did a column at FanGraphs last winter (The Great Derek Jeter Conspiracy) that pointed out that the fielders with the best ‘hands’ (pct of converting infield grounders into outs) had the most Gold Glove awards. And this is a number we can all pretty much agree on. Errors are based on the judgement of the scorer of what’s ordinary effort, so I lump the ROE’s in with infield hits, my reasoning being that if the infielder got to the ball, whether it’s an infield hit or an error, he didn’t make the out. Range is more difficult to quantify, as we have to make assumptions and guess which infielder had the best chance to field the ball (and sometimes more than one could have). And, as my article also discusses, this is an aspect of fielding that we probably don’t appreciate with our eyes - do we really remember if Jeter allowed one or two extra base hits this week? We can certainly remember the highlight play he did make. Fielding metrics will then knock down sure handed but no range guys like Jeter and Helton to the bewilderment of the general public.

Are the zones represented by the other fielders oout there?

In other words, does it matter who plays 2B next to Texiera. Cano vs whowever was there for the Angels/Braves?

Would this change his zone, and could this affect his numbers, particularly since it was new team and a new stadium?

Also, because I’ve never seen this (or can’t remember it), are the zones equal in size? In other words, does each position get 11% of the field. Or more for the outfield and infield, and less for the catcher?

For exmaple, could Texiera’s zone be 15% of the field one year, and 10% next year, dependent on who plays 2b and covers the mound? And how would that effect his numbers, if so?

Also, how do outs in foul terroritory come into play?

"I don’t buy this “90%” business… it’s a thought process derived from errors and fielding percentages.”

I completely agree.  I should have been more clear.  It was was what I meant by “casual observer”.

I don’t think you will find many people (outside of sabremetrics) who will understand that a ball put in play that could reasonably never be anything but a hit (absent insanely bad, and therefor lucky, positioning by a defender) would none-the-less be credited against a defender.  So they are seeing a success rate on defense at (and generally well above) 90%+.

Defensively, I think the casual observer makes a mental distinction between fault and no-fault on balls in play.  This is a distinction not considered by PBP defensive metrics.  Thus the disconnect between UZR and what fans “see” - including the belief that luck could have little to do with defense.

This, btw, is also why I think there *could* be an apparent incongruity by fans and UZR vis-a-vis Teixeira.  (It also may not explain it.  As you suggest, UZR might simply be wrong on Teixeira.).

20 runs of difference in performance is the difference between Adam Dunn and an average fielder right? So, you are saying that for the average fielder I’m not able to tell that type of difference in performance from one year to the next. It’s too subtle to notice.

Yet, the 20-run difference that Adam Dunn has in left field is apparent to anyone that follows the game on a regular basis.

Here’s something that I’m interested in. How accurate do you think that UZR, Total Zone or Dewan is in accurately assessing runs saved in a particular season? Is it within a couple of runs in general?

And what do you make of the 10 to 20 run difference that can apply to the same player in the same season with different metrics?

In watching the Rangers, I definitely noticed the 15 run difference between Young and Andrus according to UZR. I never much noticed the 20 run improvement in Young from one year to the next. And I didn’t notice at all the 5-run dropoff from Young to Andrus in TZ.

Why shouldn’t I think that the variance in fielding numbers could just as much be the metric as the player?

This whole “Teixeira invalidates UZR” nonsense is absurd.  If you can’t look at his performance in 2009 and see that he had a down year, then you’re just trying to see what you want to see.

Compare his 2008 and 2009, for instance.  Despite playing nearly the same number of innings at first base, he had less than half the assists he usually has—49 this year vs 99 last year.  In fact 49 assists is by far the fewest in the majors—Prince had 66 while Pujols led with 185.  .038 assists per inning is way, way below Teixeira’s previous career norm of .072.

Don’t misinterpret this to mean that I think assists is a fair measure of someone’s defense.  However it’s impossible to look at these numbers and surmise that he had a normal Teixeira-like year in the field.  UZR tells us the same story.  The argument that UZR is wrong because Teixeira is always good and defense is a constant does NOT fly.

As someone who’s created my own defensive metric (OPA!) and who is obsessed with statistical reliability, I hear your frustration on this one Mike.  We all want a measure that achieves acceptable reliability (r = .70, right Tom?  in a short time frame.  The problem with measuring fielding is that a fielder has zero control over the batted balls that are hit toward him.  It does seem that it takes a couple of years before the sample size is big enough that it all washes out.  The nice thing is that we can do studies that tell us how much faith to put or not put into a measure (reliability analyses).

The problem with relying solely on scout info is that scouts are human beings, and human beings don’t have a check on how reliable their opinions are, other than their own *sense* of how sure they are, and humans have been shown time and again to be very over-confident in their estimates of how sure they should be.

As to whether reflexes/reaction times should be more consistent, the psychologist in me knows that they are surprisingly not.  Even taking out the effects of diet (how good are your reflexes after Thanksgiving dinner?) baseball players are at a point in their lives (mid-20’s to early 30’s) when the area of the brain responsible for quickly evaluating information, sorting it out, and making a plan (the pre-frontal cortex) is going through a period of rapid development.  Development of this nature is more directed chaos than orderly progression, which, not surprisingly, might lead to some uneven results.

More battling windmills! The point of a statistic is to tell us what happened. Offensive stats always tell us is the player was successful at what is being measured. By your own admission, defense metrics don’t always do that. If Ted Williams hit .225 one year, everyone would agree he didn’t hit well (in terms of batting average), yet now you argue that UZR may be missing something with Teixiera. Well, I think that’s right...because defense metrics are still highly subjective.

Consider the following:

1B #1 can only catch balls hit right at him, but he is expertly positioned so he nearly makes every play.

1B #2 has incredible range, but his manager insists that he play with his back to homeplate. As a result, he doesn’t make any plays.

Whom you select as the better defended?

It’s an extreme and silly example, but it illustrates a point. Offensive statistics measure direct confrontations, while defensive metrics measure indirect events. As long as you trust that the relative ability of the talent pool is similar and can adjust for park effects, offensive statistics can mitigate the variables. Current defensive statistics, however, can not.

When i see UZR so low on a guy like Teixeira...it makes me question the system no doubt.

if indeed “UZR is missing something on him”...then i want to know what and why. and i don’t think thats unfair at all.

If not UZR, what is the alternative? 

Your compilation of what you read about 500 google searches on each player?  How do you reconcile everything you read?  How was a Seattle Mariner fan supposed to know about the awesomeness of Franklin Gutierrez?

defense doesnt slump

it is about effort and concentration

It’s an extreme and silly example, but it illustrates a point. Offensive statistics measure direct confrontations, while defensive metrics measure indirect events.

Consider the following!

Batter #1 can only hit fastballs down the middle, but he has a miraculous run of facing stupid pitchers, so he has an amazing season beating up on hittable pitches, batting 1.000 and hitting 150 home runs.

Batter #2 can hit anything in the strike zone, so pitchers fear him tremendous and walk him every time up.  He bats .000 and hits 0 HR. 

It’s an extreme and silly example, but it illustrates a point.  People who don’t want to believe that defense may just be vary from year to year make up some really stupid examples to defend their preconceived biases.

There’s a lot that baffles me when I read about people questioning the efficacy of defensive stats, especially those who claim that defense cannot be quantified in any possible way (because you “have to see a player every day to evaluate him!").

Granted, the accuracy of defensive stats isn’t up to par with offensive ones, although Field F/X will get us a lot closer. Having clearly admitted this, I don’t understand some of the objections.

1) I cannot believe that someone would invalidate a whole statistical framework based on his own perception of a single player (or a cherry-picked small sample of players). UZR in fact agrees with the assessment of the overwhelming majority of fielders, quantifying it. The scope of a stat is to describe what happened, not to confirm somebody’s personal bias, or even groupthink. Maybe it IS wrong on Teixeira. But that fact that its opinion differs from what observers believe doesn’t necessarily make it wrong. Maybe it does, maybe it doesn’t. Surely, however, it doesn’t invalidate the whole stat. Teixeira might be an exception that isn’t properly evaluated. Or maybe he is. After all he comes out as above average for his career, over a decent sample.

2) I’m not sure people understand the sample size that fielding deals with. Fielders have fewer chances than plate appearances and the data of a single year might be flawed just like small sample size in hitting terms. Teixeira hit .200 in April and it meant absolutely nothing. Nobody claimed that we should change how we calculate BA because Tex is better than that.

3) -3.7, that’s Teixeira’s 2009 Fangraphs UZR. That’s really just average. I mean, he makes an extra play per month and he comes out as above average. Honestly, I don’t believe anybody has eyes so fine (even by watching every day) to even notice the difference. It goes for everybody when the difference is so small. That’s where understanding the stat comes into play. If you’re +3 and somebody else is +3.2… there’s basically no difference. I’m not even sure it is the case to be quoting decimals at all.

4) Why shouldn’t defense slump? Because somebody believes so? Doesn’t make it true or right. It’s about reflexes, physical, technical skills and reaction. So is offense. You start your swing 0.02 seconds later, despite your usual excellent swing path and bat speed, and it turns a line drive into a popup. You react to a batted ball 0.02 seconds later, despite your usual athleticism and range, and it’s into right field. There isn’t much of a difference. Most of the times offensively in a slump you have the same type of swing, the same bat speed and you’re the same player, but your timing is off and something isn’t clicking. The same can happen on defense when you get a poorer read on the ball or a slower jump. In fact, it happens on the base paths as well. I remember people saying that speed doesn’t slump. Possibly true (although players play every day, and sometimes they have knocks that slow them down), but instincts get clouded by tiredness, and while the physical tool may not slump, baserunning might. Take Jacoby Ellsbury in 2008 as an example (50 SB, 11 CS). He started the season with 33 SB and 3 CS over his first 63 games. Over his next 39 he had 2 SB and 4 CS. Looks like a slump to me. He hit poorly, he ran poorly. It’s not like he was slower, but his instincts were failing and maybe he was having trouble at getting jumps on the pitchers. If that can happen on offense and it can happen on the bases (although many believe that speed doesn’t slump), why shouldn’t it happen in the field? Just because it’s counterintuitive for some?

And it is important that in no way does in fact UZR try to evaluate “true talent level” of a fielder. For Will at #28: yes, what if a guy can only catch balls at him, but positions himself well every time? Who cares? UZR isn’t trying to tell us that he is a stud defensively. It tells us that his defensive play has yielded optimal results. Same for your other example: the guy with his back to the plate isn’t yielding any result. UZR isn’t knocking his true talent level or his actual skills, but is evaluating his actual performance. Which, in fact, is what we want to know about.

So, even taking this into consideration, maybe Teixeira IS a much better fielder than stats give him credit for: that is, he’s got the tools and the style to be one. But for whatever reason, his results don’t match up with that. It happens. I mean, Corey Patterson was a top prospect, with tools. He was on the cover of BA’s first prospect handbook, and obviously the results never matched his electric tools. He’s a player that looked better than his performance, and arguably still does. We accept that, so why can’t we accept the same line of thinking in other domains? Remember that talent and performance are different things, and so are performance and results. UZR doesn’t claim to evaluate talent or tools, just the end results, which, for whatever reasons, may not match up with raw talent itself. This isn’t to say that UZR will always be exactly right on every player, of course.

@32

Either my example was too difficult for you to understand, or you just feel threatened by common sense. Aside from the fact that I never stated that defense never varies, I think you can see the differences in our extreme examples. Mine was meant to show how dependent defense is on other variables (not only the pitchers and batters, but also the team’s defensive strategy). Your example, however, illustrates the direct batter-pitcher confrontation. What’s more, defense is beholden to a more confined set of variables, while hitting is much more broad (i.e., fielders play the bulk of their innings behind 6-7 different pitchers, while hitters can face as many as 100 different pitchers).

Sorry, try again.

@33

UZR is measuring actual performance, but the measurement is based on someone’s observation. Also, it is trying to do more than just measure performance (fielding percentage does that). It is also attempting to identify those fielders who have greater “range”. So, when UZR gives a 1B credit for making a play because that fielder’s pitcher was able to execute his location, well, that becomes misleading.

Because most major league defenders should be successful almost 100% of the time on easy plays, how one defines easy play (the zones) is essential, and, like it or not, that is subjective because every AB has too many variables to simply pre-define it.

As for defense going into a slump, while I think that is possible (especially for throws), it is far less likely because playing defense is an easier skill than hitting a baseball. That’s why fielders are expected to succeed at a much higher rate than hitters. Usually, a mechanical flaw leads to being “off” at the plate, but defense (aside from throwing) doesn’t really have mechanics. Dropping ones hands, for example, can lead to being slow, but there really isn’t anything similar to point to on defense.

"Because most major league defenders should be successful almost 100% of the time on easy plays, how one defines easy play (the zones) is essential, and, like it or not, that is subjective because every AB has too many variables to simply pre-define it. “

I make this point ALL the time.  Basically, UZR knows enough about the easy play to call it “96% chance of making an out”.  And so, a player that makes the easy play gets +.04 outs per easy play.

The problem is that the difference between a great fielder and an average one is… .04 outs per play.

So, you can see how we get into a danger zone here.  That’s why we need sample size, so that the potential non-randomness of such plays will get washed away if we have a few thousand such plays.  At the same time, once you have a few thousand such plays, the need to determine zones and trajectories start to go away, as relying on the identity of the batters and pitchers might be just about all that is needed, since everything else should pretty much wash away.

The spectrum goes: less than half a season?  Rely on scouting.  Less than 2-3 seasons?  Rely on UZR.  More than 5-7 seasons?  Rely on WOWY (or something similar).

There’s a place for everything, each one weighted more than the other under different sample sizes.

Another issue with defensive metrics is that they are not park adjusted. If a ballpark can impact offense, why not defense? Fenway has a notoriously hard and uneven infield. Is this true? If so, what impact does it have on an infielder’s range? The Yankee Stadium LF is larger than many, so outfielders are often forced to play deeper. Does this depress the UZR of Yankees who play that position?

Furthermore, the position of the observer/camera in each ballpark can also play a role, creating a second way in which each ballpark can impact defensive results.

"Another issue with defensive metrics is that they are not park adjusted. “

They ARE park adjusted.

Where do people like Will come from? Do they think we’re all idiots, that we sit around thinking of new metrics all day but never think of something as obvious as park factors?

re #26:

Counting assists per inning played is not necessarily a good way to measure a first baseman’s defense if something else has important has changed.  In 2009, Teixeira moved to a ballpark with the closest RF fence in the league, and the analysis I did over the whole season looking at long fly balls and home runs suggests that the Yankees pitching staff made a concerted effort, particularly after June 1st, to limit the number of balls hit to right field, which would of course impact Teixeira’s assist numbers.

Perhaps UZR accounts somehow for this (I forget if it would correct for an unusual split of balls hit to one side vs. the other), but certainly a basic count of assists would not…

Will #35,

yes, there is some subjectivity by the stringers. UZR isn’t gospel, but it’s very good. There is subjectivity in other aspects of the game too, think about hits/errors as classified by the official scorer. There are fewer of those with respect to plate appearances, so the impact is smaller. Then again, the issue boils once again down to sample size: over a small sample size nothing has any meaning. Over a bigger sample size, it progressively gains meaning and the impact of randomness and subjective scoring becomes smaller. One season of UZR is relatively small sample size, quite simply.

Also, I keep not understanding why defense shouldn’t slump. First, I don’t think that slumps are only mechanical (I showed you Ellsbury’s SB numbers in 2008… in September he also had another cold streak of back-to-back-to-back games with 0 SB and 1 CS, and running isn’t exactly mechanical). They can also be the byproduct of poorer reflexes or many other factors. On the other hand I have absolutely no idea how you can claim that defense isn’t mechanical. Footwork, ball transfer, approaching the ball with the correct angle. Those are fundamentals that get worked upon every day while players take infield and outfield, in order to get better and get everything right with their mechanics.

What you are doing here is allowing your feelings and intuition (that defense doesn’t or shouldn’t slump, at least not “much") to get in the way of logic, without offering an actual proof. You’re making a claim, and you may even be right, but you’re basing it absolutely on your subjective feelings, not hard proof. It is not a fact that defense isn’t mechanical. It is not a fact that slumps are only caused by mechanical glitches. And it is not a fact that fielding percentage measures performance. I measure performance also by range.

Finally, you claim that fielding is an easier skill than hitting. That may be true (although try asking Adam Dunn), however that is also reflected by stats. Look at the leaderboards on Fangraphs for UZR. Let’s take shortstops. Jack Wilson led the season with +14. Rafael Furcal came in fifth at +8. Cristian Guzman was fifth-to-last at -2.3. Yuniesky Betancourt was dead last at -20.5. The difference at the extremes is 34 runs, which is a lot, but which absolutely pales in comparison to the offensive difference between Hanley Ramirez and Yuniesky Betancourt (best and worst offensive SS of the year), which comes in at 68.1, exactly twice as much. The difference between fifth and fifth-to-last in fielding is 10.3 runs, but in hitting it’s 25.1 (Yunel Escobar and Jimmy Rollins).

Chances are you wouldn’t be able to quantify with your own eyes (even by watching every respective game) the defensive difference between Furcal and Guzman, which comes out to about an extra play made every 10-15 games. So when you hear about “above average” or “below average” you also need to understand the context in which we are speaking, because defensive scales aren’t the same as offensive ones. Maybe it IS because fielding is much simpler, but the difference is much finer. That is also a byproduct of smaller sample size as opposed to offense.

Does this invalidate UZR? In fact, I think it strengthens the need for quantifying defense, since the difference is small and hard to notice with the naked eye (just like it is hard to understand who is the better hitter between two players given only 2 months of playing time, which are subject to any kind of random variation). But a small difference doesn’t mean that there is no difference. And somebody has to be the best, the worst and average.

Finally, please don’t take it as an offense, but I also don’t understand how you can come here and offer criticism of any kind upon a stat (or in fact anything at all) when you have no idea of how it works. I mean, you didn’t even know or imagine that it was park adjusted and didn’t even care to ask, while attacking its credibility. I mean that if anybody is allowed to criticize something, he should first at least take the time to know what he is talking about and how that thing works. But here you are mentioning that there is no park adjustment, no slumps on defense, no mechanics involved, etc. Some of it is plain wrong, some of it is pure subjective and baseless opinion. Come on! I guess the very root of sabermetrics is taking a little extra care in getting facts right. It is a small effort we can all make.

The main problem in evaluating defense is that we have the numerator (plays made) but we do not have the denominator (total opportunities to make a play). Try figuring out who is a better batter when you know how many hits everyone has, but do not know how many at bats.

I can tell you, each year, in eahc ballpark, for left and right handed batters, for bunts and non-bunts, how many ground ball outs Teixeira made, how many batters reached on an infield hit he fielded, how many reached on his fielding error, throwing error, or dropped catch error. Those are all balls that Teixeira got his glove on, so it’s mainly a process of deciding how much better or worse than average he performed.

I can also tell you how many ground ball base hits went to right field while Teixeira was playing first base. I don’t yet have the hit xy coordinates linked in (mgl does for UZR). This would tell me what angle the ball was hit on, and give me a clue as to how close the ball was to 1b as opposed to 2b. Still, I have to do some statistical modeling to guess how many of those ground ball hits to the outfield to assign to 1b and how many to 2b.

Many years ago I was the statistician and head scorekeeper for a college summer league, and I taught my scorers how to record the responsible fielder for each play. Every batted ball, out, hit, or error, was assigned to a fielder. I’d like to do that with MLB, but other than watching video of each play it’s not currently available. There are still positioning issues (does Utley shade towards 1b to cover Howard? Reducing hits through the hole helps Howards rating, but presumably hurts Utley and the SS up the middle. Still, If I know which fielder was closer to the ball I am closer to the truth).

There’s roughly a 10% difference between the best and worst fielders at each position. The worst MLB fielders still convert 70% of their chances into outs, while the best convert about 80%. So 70% of the batted balls are always outs, while 20% always reach base. There’s 10% left over for the fielders to be responsible for. Still, for a SS this can be 40 or more plays a year between best and worst, but that’s only about 1.5 plays a week.

If you identify those probabilities by zones, then a player gets virtually no credit for making a play in a 95% zone, or hardly loses anything by not making a catch in a 5% zone.

Compare the observed to what’s expected, being as detailed in the description of the play as the scoring data will allow. These include ballpark, fielding position, batted ball vector, bathand, bunt flag, bases occupied and outs.

The strange part about Will’s declaration that there are no park factors in fielding stats is that at the top of this thread I listed the considerations for UZR and it included this line:

“- what park the player is in, “

So, Will somehow missed reading the line, is unaware of how UZR works, and then came to his conclusion.  It’s hard to have a healthy debate with someone like that.

Greg/40 was marked for moderation and is now open.

Greg: UZR does account for it, since it considers spray and distance.

I have just finished updating my BZM fielding metric (PBP with MLB Gameday hit locations) for 2009.  BZM has Teixeira as a very good fielder in 2009, more than 11 runs above average, which ranks him 11th among first baseman.  First base is a particularly difficult position to analyze because of the first baseman having to hold the runner on first.  This requires that he position himself in a drastically different location than his normal location.  If more than an average number of hits go through while he is holding a runner on first, it is going to give a first baseman a low ranking.  Basically, it increases the problem of small sample size.

WOWY for his career has Teix as +15 runs per season… but average last year.

Isn’t WOWY just how many runs the Yankees allowed when Texeira wasn’t on the field compared to when he was on the field?  Texeira was only off the field for less than 200 innings last year - I don’t see how WOWY would be reliable at all.

Re: #40

I certainly agree—like I said, counting assists doesn’t give an accurate measure of Teixeira’s defense at all, but it does give us an indicator that *something has changed*.  Which is why it makes no sense for anyone to say “but Teixeira is always good and UZR had him as bad this year, and since defense is a constant then UZR must be wrong”.  Because clearly there was a huge change over the last year and UZR was not the only stat that noticed it.

It might be that the Yankees’ pitchers don’t cover first and so he scoops up grounders but doesn’t get outs.  It might be that his coaches have him positioned poorly.  Perhaps the new Yankee stadium lights make it harder for him to catch popups.  It might be that he’s holding on runners differently.  It could be one, or all, of a million things.  But that doesn’t matter—the key is that UZR is measured his defensive value in the real world last year, and just because it doesn’t match with a preconceived notion of his talent level doesn’t mean the algorithm is wrong or useless.  It just means his output last year was not what you would expect out of someone with his talent, and that for whatever reason the Yankees were below average last year at generating outs on balls hit near Teixeira.

Nick/47: no.

http://www.tangotiger.net/catchers.html

It looks at how CC Sabathia, IN HIS CAREER, did with Teix at first and not.  It goes through every pitcher like that.

It looks at how Carlos Beltran did with Teix at first and not, in his career.  It goes through every batter like that.

It looks at how all other 1B plays at each park, and compared that to how Teix plays in each park.

The “without” is not limited to just 2009.

This is purely factual information, on the number of balls in play when Teix is on the field, and the number of outs he initiated (I exclude pitchers-as-hitter and bunts):

YEAR BIP OUTS OUT_RATE
2003 2,883 217 0.075
2004 3,791 326 0.086
2005 4,321 337 0.078
2006 4,368 307 0.070
2007 3,277 219 0.067
2008 3,825 303 0.079
2009 3,666 247 0.067

So, in terms of outs Teix actually made per actual BIP, his 2009 was tied for his career low.

That is purely FACTUAL information, no subjectiveness at all.

We don’t know WHY it was low.  It could simply be just bad luck that he didn’t get as many balls hit his way.  The out rate of one standard deviation given 4000 BIP is .004 outs per play, or 16 outs.  Just by bad luck, he could make at least 16 fewer outs 16% of the time. This is why you can’t look at any one year in UZR and say he had a “down” year. 

It could be down because he actually had a down year.

There’s lots of things that can explain it.

Now, at the career level, we’ve got a bigger problem.  He has 26,131 balls in play with 1956 outs, or .075 outs per ball in play.  The average 1B is .070 (given the handedness of the batters/pitchers Teix had).  That puts him at +.005 outs per play, or +127 outs for him, or about +100 runs.  UZR has him at +11 runs.

Now, why would that be?  UZR is the one that has to explain itself here.  It has to show the case that the BIP distribution for Teix was anomolous enough that he got way more opps to make an out than an average player would, which is why he was able to get .075 outs per BIP while everyone else got .070 outs per BIP.  Over a 7-yr span!

The problem is when I look at the hitters and pitchers Teix has had, they were pretty typically average (as you would expect from someone who played on 4 different teams).

This means that when those pitchers and hitters had Teix on the field, they allowed and hit balls more in Teix direction than they did for other 1B.

Again, a mountain for UZR to climb.  That’s not to say it didn’t happen, but the burden is on UZR to show that the BIP distribution Teix had was as skewed as it was.

So, that’s the factual argument.

The scouting argument is that Teix looks so good, and he does so of fans of 4 different teams, and does so every year.  Not to mention that he has 99 innings as a 3B.  That gives more credence to the fans that he’s been at least a better than average 1B.

In this particular case, the onus is on UZR to show itself.

And I think it’s fair enough to say that UZR is right 80% of the time, and wrong 20% of the time, and Teix is part of the 20%.

No offense Tango but you draw some pretty weird conclusions after you’re done with your facts.

You pulled the 80%/20% right/wrong split out of nowhere, for starters, so I don’t think it’s fair to say that.

As for saying the onus is on UZR to justify why it doesn’t credit Teix with as many runs saved as you’d expect given the number of outs per batted balls… that’s like saying the onus is on WPA to explain why Teix’s wRC were nearly equal in 08/09 but his WPA was 39% lower.  The whole point of the WPA stat is to examine the context of each PA & its result, relative to global average result.  Same goes for UZR—it looks at each batted ball rather than averaging over the “typical” 2000 outs of a career, so by saying +11, the algorithm IS showing itself.  Peek under the hood and you’ll get your detailed explanation—perhaps AL West stadiums allow lots of easy foul popups, for instance—but the reason for even having an algorithm like UZR is that it sums up all of those tiny analyses so you don’t have to.

This is not true.  Over a career, I’m arguing the ball in play distribution of a batter and pitcher will average out to their personal averages, regardless of who the 1B is.  So, the expectation is that all the little things that UZR is picking up will, in a career, cancel out.  That’s why the value of UZR exists if you have 2-3 years of data.

Once you get into the 6+ years, I would suspect that MGL can remove most of the parameters he has in his arsenal, because they will mostly cancel out.  For example, will he need the adjustment he has for the FB tendency of the pitcher?  I’d say no.  If MGL gives us the UZR for players with 6+ years of data with and without the FB tendency parameter, you won’t see much, if any, difference.

Same for the base/out configuration.  We don’t expect Teix to have more runners on 1B than someone else, per PA, with 6+ years of data.

I’d say the #1 item is the handedness of the batter.  And #2 might be park. 

That’s my line of thinking.  Do you think I’m wrong?

Some more facts on Teixeira

YEAR   GB oIFH oROE eIFH eROE Plays
2006  217   10    5   11    6  +2.1
2007  164    4    5    9    5  +4.3
2008  224   12    3   11    7  +3.0
2009  182   11    1    8    5  +1.8

The observed infield hits and reached on errors are facts, the expected is from my modeling and are rounded to zero decimals.

Although he’s consistently a few plays positive each year, it’s not a big difference.

One of the things to take note of is how small the number of plays not made is. We frequently look at the number of chances (denominator) as the indicator of sample size and thus reliability, but metrics that have very small numerators also need larger samples to become significant. In park factors, it takes three years to get a very good reading on a HR factor, but still longer for triples because they occur so infrequently. When the number is small, a random variation makes a a larger percentage of the observed total. This is measured by variance divided by mean, with larger samples reducing the variance of the population until you find a sufficiently large sample which brings the ratio below an acceptable limit.

Ground balls to the outfield

YEAR  Obs Rate Exp Rate Plays  
2006   33 .134  30 .121  -3.1   
2007   30 .161  22 .121  -7.4
2008   17 .073  29 .125 +12.0
2009   22 .108  24 .117  +1.8

Again, we are only looking at a couple dozen plays not made, 32 total in 2008, 34 in 2009. Outfielders can ‘allow’ more than a hundred hits a season. Only the catcher has fewer chances than 1B.

After adding up infield grounders, ground balls to the outfield and double play starts, Marcel’ed for the past three seasons and assigned run values, my top rated MLB 1B for 2009 are Pujols +6.8 runs, Kotchman +6.1, Pearce +4.2 and Teixeira +3.2. The worst are Garko -6.8, Fielder -6.0 and Jacobs -5.3.

Tango/52

Bathand is probably the largest factor to be accounted for, as pulled grounders are more likely to be hits and have a different vector distribution, and LHB get to first faster.

For 1B, is the runner being held? (BASE_CD=1 OR 4).

There are large differences between ballparks on groundball hits to the outfield, which I believe are related to the hardness of the turf and probably height of grass. Very dry locations seem to have much higher hit rates (Lancaster CA is tops, plus Arizona Rookie League & Fall League, Cactus League, Norfolk VA is lowest).

Tom, in #50 you said it could be that he didn’t get his share of balls hit to him.  I thought earlier you said UZR accounted for spray and so forth?  I’m just trying to figure out if my theory that the Yankees pitchers strategically depressed the number of balls hit to the right side could still be a factor here…

Brian - It is possible that differences in those locations are due to the quality of the infield in some way, but Colorado’s experience with the humidor suggests that it is more likely that it is the effect that dry locations have on the COR of the ball that is causing the differences.

Greg, if Yankee pitchers allowed zero balls hit to the right side, Teix’s UZR would be zero.  If they allowed all balls hit to the right side, his UZR would be between -200 and +200, depending on how many outs he got.

UZR accounts for the spray pattern to determine how many outs he should haev gotten (that’s his “expected outs").  You compare that to his actual outs.

Since in 2009 he got to as many outs per BIP as the average 1B, and since UZR has him as around 0, then that means the spray pattern he saw was roughly average (in 2009).

Peter/56 - I had been thinking of the ball mainly for flyballs, but yes a dry ball should jump off the bat better for groundballs as well.

For grounders, it’s probably a combination of humidity on both the turf (if natural) and the ball, and groundskeeping on the hardness or softness of the turf and the height of the grass.

I have to say that I have read everything Dave Cameron has written for a long time and enjoyed it immensely until I read this thread..... What an arrogant self important blowhard.... I agree with will. While defensive metrics do add value to the discussion they are hardly complete. I can say with absolute confidence That Jeter and Cano’s defensive shortcomings and the yankees pitchers strikeout tendency (a 10 % jump over 2008) had more to do with Teixeira’s falloff in assists Than any supposed defensive slump.

Defensive metrics are in their infancy and are by no means entirely accurate. While I readily agree that the Data adds to our knowledge as any increase in data adds to the pool of understanding, arrogance and a refusal to recognize that further refinement may be necessary is the height of ignorance.

one thing that Bill James, the father of Sabermetrics always acknowledged was that While statistical analysis would help us to understand the game better, is was by no means going to define the game itself.  So Mr Cameron get off your high horse and recognize theres still work to be done and that even then the results wont be definitive.

You have “absolute confidence” in your opinion, and I’m an arrogant blowhard.  Got it.

I can say with absolute confidence That Jeter and Cano’s defensive shortcomings and the yankees pitchers strikeout tendency (a 10 % jump over 2008) had more to do with Teixeira’s falloff in assists Than any supposed defensive slump.

What does Jeter and Cano have to do with Teixeira’s assist totals?  I mean, if you want to say that Yankee pitchers are old, slow or fat, and so, Teix has no one to throw there, that’s one thing.  But, what does Jeter at SS have to do with Teix getting assists?  Are you talking about with a runner on 1B and Teix tries to turn 2, and Jeter/Cano don’t cover 2B?

As for the strikeout tendencies, again, what does that have to do with it, if the denominator is balls in play (not inning played)?

It seems more like you have a conclusion ready-made and looking for some semblance of evidence to cover that opinion.

I have to say that I have read everything Dave Cameron has written for a long time and enjoyed it immensely until I read this thread..... What an arrogant self important blowhard

I don’t believe it’s possible to immensely enjoy someone (and never tell him), and then to only say it as a prelude to insulting someone. 

Did this really have to be your first post on this blog?  We tolerate family members who act like fools every now and then because they built up currency throughout their lives.  You need currency to act like a fool and still be accepted.

tangotiger fair enough your criticism are valid to some extent. this was my first visit to this site after being referred to it from a friend at the Yankee scout.com board.

yes I was referring to assists from 3-6-6 dps and ewer opportunities for assists due to the almost 10 percent increase in staff strikeouts. I do think those factors plus the switch of leagues may have played a part in the number of assists Teixeira accumulated.

As regards Mr Cameron, I have read him religiously at fangraphs and USS Mariner and always found his work to be very informative and thought provoking. Imagine my shock at finding someone who i felt a measure of kinship and greatly appreciated on this site? I was qoite happy to say the least. them imagine my horror to read these posts.....

“It’s an extreme and silly example, but it illustrates a point.  People who don’t want to believe that defense may just be vary from year to year make up some really stupid examples to defend their preconceived biases”

“Where do people like Will come from? Do they think we’re all idiots, that we sit around thinking of new metrics all day but never think of something as obvious as park factors? “

While it is true I do not know Dave and perhaps I caught him on a bad day but it struck me as incredibly sad that someone I admired wasn’t all I had thought him to be. Certainly part of the fault is my own for putting his talented writing on a pedestal, but those comments reek of dismissal and disregard and led to my post. If i’m wrong about him i appologize, but based upon what was written i felt it was warrented

yes I was referring to assists from 3-6-6 dps and ewer opportunities for assists due to the almost 10 percent increase in staff strikeouts.

Whether Jeter/Cano complete the 3-6-3 or not, is irrelevant to Teix: he still gets the assist in the “3-6” portion of the play. 

In any case, in 2008 Jason Giambi has 36 DPs in 898 innings.  Teix had 49 in 1303.  Per inning, Giambi had more.  Even if you want to suggest that Giambi had more opps, he still had Jeter/Cano.

Basically, I don’t buy the argument, and you’ll have to make your case if you want to sell your argument.

***

As for your backstory, that’s of course your right to feel that way.  It would simply be better to be more even-handed in terms of what Dave has given to you, and what you feel about him.  If he’s been a 99 plus and 1 minus for you, your comments should reflect that.  If you are suggesting that his 1 minus (in your view) is so large as to negate the 99 plus, and that all of your posts with regard to Dave (one single post I take it?) comprises that overall feeling, then that’s fine.

It would just seem outlandish if that’s the case.

I’m just saying that your posts should provide a reasonable window to your view of Dave.

as regards dave, yes you are correct, it was more of a reactionary situation because I was shocked by his responses. your suggestion is a much more balanced approach.

with regards to Tex all I can say is that the yankees infield defense has been mediocre for years. anecdotally Jeter had his best defensive year in his last 5 and cano was visibly better with Tex. Though I cannot point to any specific statistical basis for it (hence the need for better analysis which UZR attempt to provide to its credit) I have severe doubts in the efficacy of a rating system that says Tex was below average defensively last year at least anecdotally. Anyone who watched the yankees play game by game raved about the difference in the infield defense of which Tex was the only change. The 18 game errorless streak is further anecdotal evidence.

Saqbermetrics is a constantly evolving science, all I’m saying is that its rating of Tex points to a problem, Will in his post painted a scenario where the metric will ultimately fail. pointing that out should not be ridiculed it should be used to help refine and ultimately improve the method

There are two currents running through this thread with what looks like 70% overlap and 70% disagreement despite it.

The first is about the accuracy of defensive metrics, and everyone here (including tango) agrees that defensive metrics aren’t perfect.

The second is about the usefulness of defensive metrics, and the disconnect seems to be that many people who agree with the first see it as a reason to disagree about the usefulness.

Nobody should be looking at UZR as the be-all and end-all to defensive evaluation. That’s why Tango is talking about the fan report, and WOWY. Just like how we don’t look at wOBA and determine offensive worth without looking at playing time.

But UZR is undeniably a tool that can be used in player evaluation so long as you can see where it does well, and where it does poorly.

This is a problem not of math, but of communication. I think that many people put too much emphasis on uncertain numbers in the name of sabermetrics, when the people who know the most (like tango) would never say the same.

Please try to separate your biases from what’s being discussed, and we’ll all end up happier for it.

I think everyone agrees that UZR or +/- are the best defensive stats we’ve got. 

The whole problem is we don’t know just how accurate stats like UZR are, and how they compare to a good scouting report.  Tango always says that the more data you have, the less you should rely on scouting, and that makes sense in theory, but when applying it to projections it requires a lot of guesswork.

"Anyone who watched the yankees play game by game raved about the difference in the infield defense of which Tex was the only change. The 18 game errorless streak is further anecdotal evidence.”

Are you saying that Tex’s ability to receive difficult throws makes Jeter/Cano better fielders?  This may be true, I don’t know, but at least you have an argument.

They don’t have much to do with his assist total though. Jeter could have a little to do with it, being the one to receive the 3-6 throws, but it’s hard to believe.  I’ve never seen a situation where a 1B fields a ball, wants to throw to 2B, and the shortstop is not there.  It just really isn’t difficult to cover the bag in that situation.  Cano has less to do with it.  1B don’t make 3-4 assists except on bunt plays.

Baseball reference has a breakdown of his assists.  He made 13 to 2nd, which is lower than his usual.  What really stands out is only 20 assists to pitchers covering 1st.  He had over 50 every year before it.  Maybe Joe Girardi tells his 1B to take the play himself unless he absolutely can’t?

Tex had 144 unassisted putouts, this number is typical for him.  The 2008 Yankees had 12 players see time at 1B, and had 66 assists to 147 3-u putouts.  The 2006 Marlins had 84 1B assists.  So it’s not conclusive that this is a Girardi trend.

TotalZone, btw, does not care if a 1B makes an assist or takes the play himself.  Either way it’s a ground ball, fielded, and turned into an out.  I’m sure it’s the same for UZR.