Impact of Defense

[jimmer]

 

Or, a guy can go out and throw 10 shutout innings in the 7th game of the World Series, and get the W.

 

Extreme examples aside, the single biggest factor in winning and losing a big league game is the performance of the starting pitcher.  

 

Crediting him, or debiting him, for that performance, despite all the noise, seems entirely reasonable to me.  

A pitcher can't win a game if his offense doesn't score.  A starting pitcher can't get outs if his fielders don't make plays,  A pitcher and a catcher and a coach in staff come up with a game plan. If a pitcher only goes five innings, he was only on the mound for a little more than half the game.

 

If I had to choose one stat to get rid of when it's used to try to judge a player's actual performance, I'd say the win/loss stat of pitchers. IMO, of all the traditional stats, it tells us the least about the actions of the player who was assigned the stat.  It is very likely THE stat most dependent on other player's actions.  Even moreso in today's game.

 

[Otto von Ballpark]

 

It's going to take a long time to balance it out theoretically because of all the routine catches that are made routinely before you get a chance at a balance out making play. 

 

If Ben doesn't make that catch... He will be UZR punished. He makes the catch and gets very little for it. To me that's a bottom line. 

But his UZR punishment will be, what, 0.8 runs?  That's really not that insurmountable.  It's less than 9% of 1 WAR.  An isolated play like that wouldn't be responsible for wild swings in UZR.

 

And if that play isn't isolated, if Revere's positioning is causing him to miss multiple plays in that zone, and if it's not helping him convert opportunities elsewhere (deep left center?) into outs, then he and by extension his team should be punished in an estimate of defensive performance.  That's a feature of UZR, not a bug.

[Brock Beauchamp]

 

Is it that bad?  It's a great way to scale a larger sample to a season size for comparison purposes.  Again, some folks will use it to scale a too small sample but just call them out or ignore them if they do that.  I'd rather have more of that information easily available and just tune out the misusers, personally.

It really is that bad. It's taking noisy data and multiplying it. Nothing useful will be gleaned from that.

 

What would be far more useful is UZR/4000 or UZR/3 (shorter name). The UZR of a player over his past 4000 innings, or roughly three seasons. That would churn out a number that might actually be of use, something UZR/150 can't do without dumb luck and/or coincidence.

[Otto von Ballpark]

 

It really is that bad. It's taking noisy data and multiplying it. Nothing useful will be gleaned from that.

 

What would be far more useful is UZR/4000 or UZR/3 (shorter name). The UZR of a player over his past 4000 innings, or roughly three seasons. That would churn out a number that might actually be of use, something UZR/150 can't do without dumb luck and/or coincidence.

It is not multiplying it if you look at career UZR/150. It's dividing it.  It's like Rdrs/yr at B-Ref -- those are great career numbers, because then I don't have to get out my calculator to do a rough comparison of a guy with a 5 year career vs a guy with a 9 year career.  Puts them both on a rough "seasonal" scale.

 

If someone is relying on a quarter-season's UZR sample for anything too great, hiding UZR/150 isn't going to deter that person from misusing statistics.

[Brock Beauchamp]

 

It is not multiplying it if you look at career UZR/150. It's dividing it.  It's like Rdrs/yr at B-Ref -- those are great career numbers, because then I don't have to get out my calculator to do a rough comparison of a guy with a 5 year career vs a guy with a 9 year career.  Puts them both on a rough "seasonal" scale.

 

If someone is relying on a quarter-season's UZR sample for anything too great, hiding UZR/150 isn't going to deter that person from misusing statistics.

No, people will still misuse numbers but it's a mostly useless statistic to begin with... Why not replace something kinda crappy with something useful, like UZR/3?

 

UZR/150 is still going to be a problem in career numbers because it's multiplying shoddy data. If something is inaccurate enough to award a 3 when a 1 is deserved, multiplying that data is going to cause discrepancies, even if you divide it and average it out later.

 

At least with UZR, the number is never multiplied so inaccurate data isn't compounded.

[jimmer]

And as far as the RBI stat, let's look at this very common example.  Guy gets on base (by walk, HBP, single). Then next guy gets a single and this allows the guy on 1B to go to 3B.  The next guy flies out to the OF and the guy from 3B scores.

 

Whoever made this RBI stat has decided that the only guy who made an out deserves the credit for the RBI.  He's a run producer.  The guy who scored gets the much less hyped stat, run scored.  The guy who put the guy on 3B with his non-out-making-single gets ZERO credit for any of it in the traditional box score.

[Otto von Ballpark]

 

No, people will still misuse numbers but it's a mostly useless statistic to begin with... Why not replace something kinda crappy with something useful, like UZR/3?

 

UZR/150 is still going to be a problem in career numbers because it's multiplying shoddy data. If something is inaccurate enough to award a 3 when a 1 is deserved, multiplying that data is going to cause discrepancies, even if you divide it and average it out later.

 

At least with UZR, the number is never multiplied so inaccurate data isn't compounded.

Career UZR/150 isn't multiplying anything, I don't think.  It's career UZR scaled to 150 games.

[Riverbrian]

 

Career UZR/150 isn't multiplying anything, I don't think.  It's career UZR scaled to 150 games.

 

You know what we need.

 

A Daily 2015 Torii Hunter Tracker thread. Just to watch the swings as they happen.   

 

It could be informative for those inclined to watch such things.

[Brock Beauchamp]

 

Career UZR/150 isn't multiplying anything, I don't think.  It's career UZR scaled to 150 games.

I couldn't find any info on this and a check of Torii Hunter's yearly UZR/150 divided by seasons played output -1.27 (his career UZR/150 is -1.3), so my gut is telling me they're multiplying and averaging (used Hunter because he has a few partial seasons mixed in there, which should return a different number).

 

Hunter has 18188.2 career innings. Divide those by 1350 (150 games times 9) and divide that by his career -14.5 UZR and the return is -0.9, which should be his career UZR/150 if the number was calculated by a career whole.

 

So yeah, UZR/150 blows.

[jimmer]

'You may also see UZR/150 around the site, which is simply UZR scaled to an average number of chances for a season.'

 

I really don't look at UZR/150 that much, but it seems that it's just used on a season to season basis. Over the course of a career, UZR/150 wouldn't be used. I've never seen/heard anyone do it.

 

As far as using it for an individual season, I wouldn't put much faith in it if the guy actually only played like 30 games or so.  The assumption that however he performed over 30 games means that's what he'd do over 150 is too much for me.  I wouldn't make that assumption with ANY stat. If a guy played 145 games, I may have less issues with it.

 

That's just me though.

 

 

[USAFChief]

And as far as the RBI stat, let's look at this very common example.  Guy gets on base (by walk, HBP, single). Then next guy gets a single and this allows the guy on 1B to go to 3B.  The next guy flies out to the OF and the guy from 3B scores.

 

Whoever made this RBI stat has decided that the only guy who made an out deserves the credit for the RBI.  He's a run producer.  The guy who scored gets the much less hyped stat, run scored.  The guy who put the guy on 3B with his non-out-making-single gets ZERO credit for any of it in the traditional box score.

I can make up examples, too. Guy gets on first. Next guy gets a single, first and third. Next guy can't get the ball out of the infield. No run scores.

 

Next guy strikes out.

 

Next guy lines a two run double into the gap, driving in two runs. It would be my opinion his contribution was the most valuable of the inning.

 

The object isn't to get to first base,it's to safely reach home.

 

BTW, the first two guys DO get credit in the traditional box score, whether they score or not. First guy gets credit for a walk, second guy for a single.

[Otto von Ballpark]

Don't have a chance to look at it too closely at the moment, but I don't think UZR exists for Hunter's entire career. That could be skewing your calculations here.

[Brock Beauchamp]

 

Don't have a chance to look at it too closely at the moment, but I don't think UZR exists for Hunter's entire career. That could be skewing your calculations here.

I'd have to double-check but I'm almost certain I only counted years where UZR data was available.

[jimmer]

 

BTW, the first two guys DO get credit in the traditional box score, whether they score or not. First guy gets credit for a walk, second guy for a single.

second guy in my example gets ZERO credit in regards to the RBI, the stat we are talking about, even though he had a huge effect on the guy being able to score.

 

And my example isn't an outlier either. Not one of these rare unicorn occasions. Mauer is a good example of a guy effected by this.

 

Like the win/loss stats for pitchers, there are too many players involved in scoring a run most of the time to give the most important stat to one guy. By itself, it tells us a run scored but it doesn't tell us the guy with a lot of them was the most important factor in that run scoring or that  he is necessarily a good run producer (or the guy with not so many RBI is a bad one.)

[ashbury]

 

His arm ain't good... I'll never argue that. However... He can still throw the ball faster then a base runner can run... Even Billy Hamilton.

I've never seen Billy Hamilton reach second base on two bounces.

 

Revere misses his cutoff man all the time - but the under, not the over.

 

As for the main topic, defensive stats' problem is they are too sparse, period. Waiting for 3 years' data is not a practical solution. Whereas a season contains only a few dozen interesting plays to consider for the average outfielder, his batting numbers can be contemplated usefully because of the rich variety - it's not simply 700 or so PA, it's that each PA has several pitches, each of which required skill (or lack) to decide on, leading to literally thousands of decisions per season on which the stats get based. Defense is thus approximately 1% as rich in information to work with. And yet defense is important to the team's success. An unsolved dilemma, at present.

[jay]

I agree with you 100 percent... UZR is more a measure of outcomes then actual skill.

That's exactly right. I don't understand why there's a common expectation or perception otherwise. Even the most traditional stats are that way. That a guy hit 20 HRs in year doesn't tell us everything, but it tells us something. Same for defense stats...

[jay]

Brian Dozier, 5 WAR player, yes!

 

Alex Gorden, 6 WAR player, eeeesh. Hey, he might be a 6 WAR player, I just have no faith in the metric at that point.

Except that's another example of how not to use WAR. Just like defense stats, home runs, RBIs or most other stats... accumulating a 6-WAR season, a +20 DRS, 50 HRs or 120 RBIs doesn't mean you're a "[fill-in any of the above] player".

 

Why does that same characteristic make you distrust UZR or WAR?

 

How many years of home run data would you want to see before saying a guy is a 20 HR player? More than one...

[Brock Beauchamp]

In no way did I imply that 2015 Brian Dozier is a 5 WAR player. It was implied that he was a 5 WAR player in 2014, as in "5 Wins Above Replacement".

 

I can't say the same about Gordon, who was a decent bat that derived around half his WAR from defense. It's possible he was a 6 WAR player but because of the inconsistencies with defensive metrics, it's also possible he's closer to a 4 WAR player.

[jay]

The forest is still there...

 

Why is it expected that advanced metrics like UZR and WAR will tell you what the player's actual skill level is when other stats don't do that either?

 

The whole "3 years of data" thing gets misused as a way to dismiss defense stats. It's 3 years of data to give a good picture of actual skill level. It still tells us what happened (within the measurement accuracy, which is better than it gets credit for).

[Brock Beauchamp]

But it doesn't actually tell what happened. It said he made some hard catches. It's not a given that those catches were terribly valuable to the team. They were valuable, like all catches... But a home run is more valuable than a double. Every. Single. Time.

[jay]

But it doesn't actually tell what happened. It said he made some hard catches. It's not a given that those catches were terribly valuable to the team. They were valuable, like all catches... But a home run is more valuable than a double. Every. Single. Time.

A home run has inherent run value so that's a tough comparison, but change that to a double and a single.

 

A double is harder to hit than a single, but it's not a given that it was terribly valuable to the team... same issue.

[Brock Beauchamp]

A home run has inherent run value so that's a tough comparison, but change that to a double and a single.

 

A double is harder to hit than a single, but it's not a given that it was terribly valuable to the team... same issue.

Nope, still different. A catch is a catch. It counts for one out no matter whether the fielder stood there and made the catch or whether he ran 200 ft to make the most spectacular catch in baseball history but UZR grades them differently. Defensive metrics assume that making a hard catch is inherently more valuable than an easy catch because it's a repeatable skill based on a single event. That's not necessarily the case.

 

On the other hand, a walk is nice but it only advances everyone a single base... But a triple advances everyone three bases. When counted in the aggregate, a triple is far more valuable and the metrics count it as such. There's no defensive equivalent to a triple because after a triple, the bases are cleared and a guy is standing on third base. That actually happened. There are no instances where a single is more valuable than a triple but there are plenty of cases where making a hard catch is far less critical than making an easy one. The offensive metric weights those two events correctly but the defensive metric might not.

[jay]

Nope, still different. A catch is a catch. It counts for one out no matter whether the fielder stood there and made the catch or whether he ran 200 ft to make the most spectacular catch in baseball history but UZR grades them differently.

That's like saying a hit is a hit...

 

But we know they aren't all the same just as we know not all catches are the same.

 

I'm not quite sure what your point is here... UZR is wrong for not valuing all catches the same?

[Brock Beauchamp]

That's like saying a hit is a hit...

 

But we know they aren't all the same just as we know not all catches are the same.

 

I'm not quite sure what your point is here... UZR is wrong for not valuing all catches the same?

But at their core, all catches are the same. The end result is a single out.

 

A triple is not the same as a single. One advances the runner three bases every time. The other advances the runner a single base every time. There is inherent value in advancing more bases and a single is never more valuable than a triple.

 

And that's why UZR needs a lot of data to be accurate. It's rewarding difficulty. It's assuming that the "run across the field, making the spectacular diving catch" is inherently more valuable than the "stand there, don't screw up catch". That's not necessarily the case. They're both outs... But one nets the defender a 1-2 RAR and the other barely moves the needle.

 

To counteract that problem, UZR needs a lot of data because in the aggregate, those spectacular diving catches indicate ability so the metric assumes the defender is routinely making spectacular diving catches and that some of those catches are super-valuable.

 

The best way to put it is this way:

 

Defensive metrics rely on a Boolean result every play.

 

Offensive metrics can have a multitude of results in an AB and weight them on value. Home run is best, triple next, etc. In no situation is a walk a better result than a single or a double better than a home run.

 

Over the course of a season, the weighted metric with multiple results is going to be more accurate when valuing a player.

[Richie the Rally Goat]

But at their core, all catches are the same. The end result is a single out.

A triple is not the same as a single. One advances the runner three bases every time. The other advances the runner a single base every time. There is inherent value in advancing more bases and a single is never more valuable than a triple.

And that's why UZR needs a lot of data to be accurate. It's rewarding difficulty. It's assuming that the "run across the field, making the spectacular diving catch" is inherently more valuable than the "stand there, don't screw up catch". That's not necessarily the case. They're both outs... But one nets the defender a 1-2 RAR and the other barely moves the needle.

To counteract that problem, UZR needs a lot of data because in the aggregate, those spectacular diving catches indicate ability so the metric assumes the defender is routinely making spectacular diving catches and that some of those catches are super-valuable.

The best way to put it is this way:

Defensive metrics rely on a Boolean result every play.

Offensive metrics can have a multitude of results in an AB and weight them on value. Home run is best, triple next, etc. In no situation is a walk a better result than a single or a double better than a home run.

Over the course of a season, the weighted metric with multiple results is going to be more accurate.

so defensive metrics should be weighted based on runs avoided? Say a catch made with the bases loaded and two outs should be weighted heavier than a catch made with bases clear and no outs? Or robbing a home run?

[Brock Beauchamp]

Depends on what you want from the metrics. My post is in response to this statement about defensive metrics:

 

"It still tells us what happened (within the measurement accuracy, which is better than it gets credit for)"

 

For the reasons listed above, defensive metrics don't actually tell what happened because they're weighting difficulty with the expectation that the result is repeatable. Offensive metrics weight result like home run, walk, single, etc.

 

To draw a comparison, it'd be like OPS giving extra points to the 500 ft home run over its 400 ft counterpart with the expectation that a hitter who can rip a 500 ft home run is stronger and will hit more of them. Probably true but not result-based and not a reliable way to weight an event over a few data points.

[Otto von Ballpark]

 

I couldn't find any info on this and a check of Torii Hunter's yearly UZR/150 divided by seasons played output -1.27 (his career UZR/150 is -1.3), so my gut is telling me they're multiplying and averaging (used Hunter because he has a few partial seasons mixed in there, which should return a different number).

 

Hunter has 18188.2 career innings. Divide those by 1350 (150 games times 9) and divide that by his career -14.5 UZR and the return is -0.9, which should be his career UZR/150 if the number was calculated by a career whole.

 

So yeah, UZR/150 blows.

Just checked -- the issue is that UZR isn't available until 2002.  And you used his total career innings from 1998.  (It's confusing, because the career total innings are still listed under the Fangraphs "Advanced Fielding" table, but that table only begins in 2002.)

 

If you do the math with just the innings starting in 2002 (15054.2 innings), it works out exactly as represented (-1.3 UZR per 1350 innings or 150 games).

 

I'm all for being critical (sorry for the on this thread ), but let's give the Fangraphs guys some credit here -- they're not complete morons to equally weigh a 10 game season and a 150 game season when presenting a career rate stat.

[Otto von Ballpark]

And B-Ref does the same thing with their Rdrs/yr stat.

 

And if someone assumes those are terribly meaningful over a small sample, it's really no different than someone doing the same for any rate stat over a small sample (remember Chris Parmelee's 2011 OPS?).  I really don't think that every online stats resource is doing some great disservice to the community by continuing to publish rate stats for small sample seasons.  Focus your rage on the stat abusers, not the providers.

[ashbury]

 

so defensive metrics should be weighted based on runs avoided?

Maybe or maybe not, but I think it's fair to say that a fielder takes the baserunner situation into account when faced with a non-routine fielding chance.

[Otto von Ballpark]

Maybe or maybe not, but I think it's fair to say that a fielder takes the baserunner situation into account when faced with a non-routine fielding chance.

During the play, it is probably a very small effect, and mostly something we want to include in UZR (i.e. credit a player for not screwing up the catch while anticipating the throw after).

 

The bigger effect would be positioning. The Braves OF shouldn't lose much for letting Gene Larkin's hit drop safely deep at the end of the 1991 series. Although those plays are probably sufficiently rare, and if UZR ignores the results of plays with extreme or rare shifts, it might not hurt them anyway.

 

Another example would be infielders guarding the line, or a deep "no doubles" outfield. It would be interesting to know how often those impact defensive opportunities and if/how metrics deal with them.