Impact of Defense

[jay]

 

Even if we reduce each of those extremes by 20%, we're still at a difference closer to 110.

 

By the way, I'd like to see FG and other sites do that with their defensive metrics.  Weight them towards the mean by whatever an appropriate measurement error would be.

[Otto von Ballpark]

Another point: Fangraphs pitching value breakdown:

 

http://www.fangraphs.com/leaders.aspx?pos=all&stats=pit&lg=all&qual=0&type=6&season=2014&month=0&season1=2014&ind=0&team=0,ts&rost=0&age=0&filter=&players=0

 

FDP-Wins is how many wins they would add to their FIP-based pitcher WAR to account for defense/luck in the form of BABIP and LOB%.  The Royals had 3.7 FDP-Wins, dwarfed by their 18.8 pitcher WAR and 15.5 offense WAR.

[Otto von Ballpark]

 

By the way, I'd like to see FG and other sites do that with their defensive metrics.  Weight them towards the mean by whatever an appropriate measurement error would be.

I've heard that suggested before, but rather than regress every result toward the mean because it is imperfect, I'd rather just have the raw result with the understanding that it is imperfect.  Maybe sometimes it's high/low due to measurement error, but sometimes it isn't, and that could be useful information.

[diehardtwinsfan]

 

for all intents and purposes, home runs do not count on defense and they're the only way to get one run from a single plate appearance.

 

Unless your name is Jose Canseco and you happen to have a ball bounce off of you into the stands...

[jay]

 

I'd rather just have the raw result with the understanding that it is imperfect.  

 

I probably would too, but then we end up in these debates.  I think it would be more useful on a net whole to more people if they were weighted for an error range.

[Otto von Ballpark]

 

You have to remember where some of this started - a figure (120 runs) that's absurd on the face of it.  I think the arguments centering more around 60-70 runs seems much more reasonable.  I can get behind that.

The difference between 120 runs and 70 runs is like 5 wins over a season.  5 wins can swing on luck, bullpen management, etc.  The Twins were 5 wins short of their Pythag record last year.  Why is it so absurd that comparing the absolute best defense and absolute worst defense in the league stretch the top end of a range by that amount?

 

 

Like I said, 26 of 30 MLB teams were within 80 runs on defense last year, by popular calculations.  Now take the outliers on both sides of that, and yeah, you might see 120.  Regress it down to 100, 80, 70, or whatever if you want, but it's still a big difference, and it's not any consolation for the fans of Indians -- their defense still stunk and it cost them games and quite possibly a playoff appearance.  And no matter how confident a Fangraphs writer sounds, I don't think anyone will fight you to death over the exact 120 number.

 

[Willihammer]

Another point: Fangraphs pitching value breakdown:

 

http://www.fangraphs.com/leaders.aspx?pos=all&stats=pit&lg=all&qual=0&type=6&season=2014&month=0&season1=2014&ind=0&team=0,ts&rost=0&age=0&filter=&players=0

 

FDP-Wins is how many wins they would add to their FIP-based pitcher WAR to account for defense/luck in the form of BABIP and LOB%.  The Royals had 3.7 FDP-Wins, dwarfed by their 18.8 pitcher WAR and 15.5 offense WAR.

Actually I think you want to look just at the BIP-wins column, as that isolates out the LOB-wins. I think LOB-wins would be more heavily influenced by a pitchers ability to pitch out of the stretch (pitcher skill) and also sequencing.

[Otto von Ballpark]

 

I probably would too, but then we end up in these debates.  I think it would be more useful on a net whole to more people if they were weighted for an error range.

We don't demand an error range for our hitting and pitching outliers.  I think people just demand more from defense because it's harder to tabulate and compare.

[TheLeviathan]

 

 And no matter how confident a Fangraphs writer sounds, I don't think anyone will fight you to death over the exact 120 number.

 

I don't know, you should probably check that other thread.  I'm not sure you're right about that.

 

As for your conclusion, I think 5 wins is a fairly significant margin that shouldn't be so easily dismissed. 

[jimmer]

 

We don't demand an error range for our hitting and pitching outliers.  I think people just demand more from defense because it's harder to tabulate and compare.

There are long time stats found in box score for offense and pitching.  Unless you count errors as an accurate stats, and I'm sure some do.  So yeah, it's easy for people just to accept those numbers at face value.

[Otto von Ballpark]

 

Actually I think you want to look just at the BIP-wins column, as that isolates out the LOB-wins. I think LOB-wins would be more heavily influenced by a pitchers ability to pitch out of the stretch (pitcher skill) and also sequencing.

Thanks, I actually just discovered those stats today.

 

Any thoughts or experience with how BIP-Wins correlate with Fielding runs?  Converting the runs, I get 26 teams within 98 runs, or 0.6 per game.  Does that calculation bolster the argument that a standard 0.5 runs per game difference in fielding is not unreasonable?

[Otto von Ballpark]

 

As for your conclusion, I think 5 wins is a fairly significant margin that shouldn't be so easily dismissed. 

I didn't dismiss it.  I am saying that 5 wins can be swung over 162 games for a lot of things, so it's not "absurd on the face of it" to see outliers stretch a standard range by 5 wins over a season-long run approximation.

[jay]

 

I didn't dismiss it.  I am saying that 5 wins can be swung over 162 games for a lot of things, so it's not "absurd on the face of it" to see outliers stretch a standard range by 5 wins over a season-long run approximation.

 

And that stretch assumes a "real" base of only 80.

 

Following up on the math in the other thread, if we split up runs allowed into simply defense and pitching (ignoring luck and other factors), that's saying defense is only something like 27% to pitching's 73%. Feels too extreme.

 

Move that base to 110 and you're closer to a 1/3 to 2/3 split between defense and pitching. That feels about right to me. From there, some combination of measurement error and outliers stretching the gap to 140 seems reasonable. 

 

Even at a gap of 140 runs between the best and worst defenses, that's still only calling runs allowed a 40/60 split between defense and pitching.  The impact of luck, park factors, sequencing, etc. will then serve to skew the real-world results from whatever that real split is. 

[USAFChief]

FWIW, Just wanted to say I have very much enjoyed this thread.

 

I am going to read it several more times before even thinking about posting. There's a lot to chew on here.

[jay]

FWIW, Just wanted to say I have very much enjoyed this thread.

 

I am going to read it several more times before even thinking about posting. There's a lot to chew on here.

I can appreciate that you're at least willing to hear and consider the debate despite being quite skeptical (which, btw, accurately describes every USAF/USA Chief I've ever known).

 

A big trip up seems to be how many runs are being assigned to defense vs other factors such as pitching. To be sure it's clear, some of the math in the thread was too fuzzy and inaccurately used. If runs allowed were simply due to either defense or pitching, I'd be curious to hear what you think that split looks like... (we can make it far more complicated from there)

[Brock Beauchamp]

 

If runs allowed were simply due to either defense or pitching, I'd be curious to hear what you think that split looks like... (we can make it far more complicated from there)

Well, runs allowed are due to either defense or pitching (within a margin of ballpark error). Whether it's pitch framing, base-stealing, home runs, hits, or walks, either the pitcher or defense (or both) had control over the outcome of the play. I can't think of a situation where one of those two things didn't control the outcome of the at-bat (mostly the pitcher).

 

If you forced me to put a number on it, I'd probably go with 34/66 for defense/pitching. There are simply so many plays that defense doesn't impact (SO, HR, HBP, BB) while a large portion of defensive plays are routine plays made by every defender in MLB. We're talking about a handful of plays a game that differentiate a great defense from a terrible one whereas a great pitching staff will take the defense out of the game entirely 3-4 more times than a bad pitching staff (going from a rough estimation of 5-6 K/9 compared to 8-9 K/9).

 

For example, the Cleveland Indians recorded just shy of 1/3rd of their 2014 outs from the strikeout. That only leaves ~67% of all recorded outs up to the defense and a large portion of those outs are routine outs made by pretty much any defender in the league (particularly flyballs). My beef with defensive metrics is how they're recorded and how few outlier chances there are to glean meaningful data.

 

Every outfielder will reach the same balls or won't reach the same balls... The margin of defense is a few feet (for example, Josh Willingham might be able to run 25 feet on a flyball to left-center while a better left fielder might make it 30 feet). With infielders, the margin is even smaller (often as little as a foot). The difference between a great defender and a bad defender are narrow strips of territory that circle the fringes of a position player's range. I can't buy into the fact that balls are so often hit to those narrow "outlier strips" of the field (as opposed to the rest of the field) that defense can have more influence on the game than pitching.

 

It's not impossible that defense has more impact than pitching over the course of the season but it's contrary to what I see happening on the field on a daily basis.

 

[Otto von Ballpark]

 

It's not impossible that defense has more impact than pitching over the course of the season but it's contrary to what I see happening on the field on a daily basis.

Again, I don't think anybody is saying that.

 

By Fangraphs, the absolute worst pitching staff in the league was worth 113 runs above replacement last year.

 

The absolute BEST fielding team, as measured by UZR?  Only 67 runs.

[Brock Beauchamp]

 

Again, I don't think anybody is saying that.

 

By Fangraphs, the absolute worst pitching staff in the league was worth 113 runs above replacement last year.

 

The absolute BEST fielding team, as measured by UZR?  Only 67 runs.

I phrased that sentence really poorly. I'll just leave it at that.

 

But that's only half the swing (and different metrics have different outputs). There's the "below average" side of that coin as well.

 

For example, DRS:

 

Best: Reds: 67

Worst: Indians: -75

 

A swing of 142 runs. Again, I struggle to buy into that number.

[TheLeviathan]

 

Again, I don't think anybody is saying that.

 

By Fangraphs, the absolute worst pitching staff in the league was worth 113 runs above replacement last year.

 

The absolute BEST fielding team, as measured by UZR?  Only 67 runs.

 

If jay is saying that outlier defenses can make a 140 run difference a year.....isn't that what some are saying?  It feels like we're all settled in around 30/70, but disagree about outliers being 50/50.  Maybe I'm misunderstanding that earlier point.

 

But I think what Brock is saying is important - we seem to lack the equivalent of a "replacement" defense to see what the true baseline is for how much better than average these teams are and instead we seem to be assuming great defenses are constantly making great plays in run saving situations.  That's counter-intuitive to my baseball watching experience.

[Brock Beauchamp]

 

But I think what Brock is saying is important - we seem to lack the equivalent of a "replacement" defense to see what the true baseline is for how much better than average these teams are and instead we seem to be assuming great defenses are constantly making great plays in run saving situations.  That's counter-intuitive to my baseball watching experience.

Bingo. As I've mentioned a few times, I have less issues with the parsing of the data than I do the collection of the data, which is why I'm so excited to see Field f/x implemented across baseball.

 

Once we can track ball speed, height, and air time while also tracking the exact number of feet travelled by a defender to reach that ball's point of impact, I think we're going to see huge breakthroughs in defensive metrics.

 

Because right now, we're just kinda guessing and there's the potential for a ton of human error in gauging whether Willingham ran 20 or 24 feet on that line drive (which is a huge discrepancy in the data, 20% margin there) and how badly he missed the ball (was it one foot or three feet, a monster discrepancy of up to 200%). How many opportunities does Willingham have like that during a game? One? That's ~150 outlier chances a season, which we would dismiss out of hand were it a hitting metric (RISP, for example).

 

Once we start accumulating that data, we're going to have a much more concrete baseline (with most of human error removed from the data) of what equals an average, bad, or good defender.

[TheLeviathan]

Not just removing the human error, but incorporating positioning of players, distance they are covering, speed getting to plays, and just a host of other things that are vital components of defensive success that are only vaguely (if at all) in the equation.

 

Sport-Trac is one of the most interesting things to watch the next 5 years.

[Brock Beauchamp]

 

Not just removing the human error, but incorporating positioning of players, distance they are covering, speed getting to plays, and just a host of other things that are vital components of defensive success that are only vaguely (if at all) in the equation.

 

Sport-Trac is one of the most interesting things to watch the next 5 years.

Absolutely. The use of the shift can be exactly quantified once we can establish the exact positioning of the player and how far he had to travel to reach (or not reach) the ball.

 

The more quickly we can eliminate the Franklin Guitierrez defense metrics of the world, the better. More than either pitching or hitting, defense should be consistent from season to season (barring injury), yet we get weird outliers like Franklin posting a 32 DRS one season and a 0 DRS the next. Did somebody steal his cleats? Were his legs tied together? How do we explain away a guy losing 30 defensive runs in a season?

[Otto von Ballpark]

 

Because right now, we're just kinda guessing and there's the potential for a ton of human error in gauging whether Willingham ran 20 or 24 feet on that line drive (which is a huge discrepancy in the data, 20% margin there) and how badly he missed the ball (was it one foot or three feet, a monster discrepancy of up to 200%). How many opportunities does Willingham have like that during a game? One? That's ~150 outlier chances a season, which we would dismiss out of hand were it a hitting metric (RISP, for example).

UZR doesn't care how much he ran or how badly he missed.  If it is in a zone that an average LF catches 25% of the time, and Willingham doesn't catch it, he gets a demerit below average.  If he does catch it, he gets credit for the play (minus the 25% average).

 

Obviously the new Statcast "Field/FX" data will be awesome, but I don't think that means the current method is as bad as you paint it.

[jimmer]

 

UZR doesn't care how much he ran or how badly he missed.  If it is in a zone that an average LF catches 25% of the time, and Willingham doesn't catch it, he gets a demerit below average.  If he does catch it, he gets credit for the play (minus the 25% average).

 

Obviously the new Statcast "Field/FX" data will be awesome, but I don't think that means the current method is as bad as you paint it.

And even then, while taking shift into account MAY effect a players's individual defensive metrics, it shouldn't effect the overall team defensive stats since the TEAM deployed the shift to begin with. certainly  the team would get credit for that, one would imagine.

[Otto von Ballpark]

 

I phrased that sentence really poorly. I'll just leave it at that.

 

But that's only half the swing (and different metrics have different outputs). There's the "below average" side of that coin as well.

 

For example, DRS:

 

Best: Reds: 67

Worst: Indians: -75

 

A swing of 142 runs. Again, I struggle to buy into that number.

By WAR, the gap between the best and worst fielding teams is comparable to the gap between the best and worst hitting or pitching teams, true.  Which seems wrong -- fielding isn't nearly as difficult as hitting or pitching, right?

 

But you have to consider there is immense competition between teams for hitting and pitching talent.  Basically, if you live in North America and are capable of hitting or pitching in MLB, you are almost certainly in MLB.  Fielding?  Not so much.  While fielders are scouted and graded too, very very few players make MLB for their fielding alone.

 

So you're not going to see a godawful hitter in MLB, or a godawful pitcher (at least not for too long).  But you could see a godawful fielder.  So the fielders in MLB just happen to be the ones who are varying degrees above replacement as hitters, so you're going to see a mix of them, not some narrow range of player cultivated for their fielding skills.

 

Extend it beyond individual fielding skills to team positioning and that's another range.  Managers and coaches haven't necessarily been selected based on their abilities to optimally position their team defensively either.

 

Plus like any data set, you're going to have noise which pushes the range to the outliers.  It's understandably much easier to see in discrete hitting/pitching events than defensive run estimators, but I don't think its presence in those run estimators invalidates the stat.

[Otto von Ballpark]

 

Not just removing the human error, but incorporating positioning of players, distance they are covering, speed getting to plays, and just a host of other things that are vital components of defensive success that are only vaguely (if at all) in the equation.

They are baked into the UZR equation, in the sense that all of those things either help or hinder the player from converting the play into an out.

 

Agreed it would be useful and fascinating to get that level of detail, but the lack of that detail doesn't mean the current attempts at defensive stats are that bad.

[Brock Beauchamp]

 

UZR doesn't care how much he ran or how badly he missed.  If it is in a zone that an average LF catches 25% of the time, and Willingham doesn't catch it, he gets a demerit below average.  If he does catch it, he gets credit for the play (minus the 25% average).

 

Obviously the new Statcast "Field/FX" data will be awesome, but I don't think that means the current method is as bad as you paint it.

Except that "zones" are also decided by human beings and are prone to inaccuracy.

 

Was the shift on? Did the defender run 20 or 26 feet to the ball? Was it a line drive or fly ball? How long did it hang in the air? How easily did he catch the ball or by what distance did he miss it?

 

Couple all of that with the few chances on outlier plays to begin with and you have the recipe for a lot of bad data.

[Otto von Ballpark]

 

But I think what Brock is saying is important - we seem to lack the equivalent of a "replacement" defense to see what the true baseline is for how much better than average these teams are and instead we seem to be assuming great defenses are constantly making great plays in run saving situations.  That's counter-intuitive to my baseball watching experience.

We're not assuming anything like that.

 

UZR looks at every play from the past 5 years or whatever, classifying it into zones based on direction/landing spot and also batted ball speed.  For each of those variables, it determines how often that play was converted into an out by each fielder.  That's the average or replacement level rate.

 

Then, when one of those plays happens, in that zone and with speed, if the particular fielder converts the play into an out, he gets credit for the 1 play minus the average rate.  If he fails to convert it into an out, he gets docked the average rate.

 

Obviously further granularity would be cool, but this isn't a bad method for looking at the result and grading how often teams/players achieve that result or fail to do so.

[Otto von Ballpark]

 

Except that "zones" are also decided by human beings and are prone to inaccuracy.

 

Was the shift on? Did the defender run 20 or 26 feet to the ball? Was it a line drive or fly ball? How long did it hang in the air? How easily did he catch the ball or by what distance did he miss it?

 

Couple all of that with the few chances on outlier plays to begin with and you have the recipe for a lot of bad data.

All of those things are interesting facets of defense, but in an overall grade of team defense, they are baked into the result.  Maybe KC had a great shift, maybe Baltimore had great team speed.  Either way they converted plays into outs at that rate.

 

Also the zones are pretty small nowadays, and based on video as well as observation.  It's not just some guy holding up his index finder and thumb and guesstimating.  And the exact speed of the ball off the bat is already tracked, and I wouldn't be surprised if hang time is tracked too (pretty easy on video).

[TheLeviathan]

 

They are baked into the UZR equation, in the sense that all of those things either help or hinder the player from converting the play into an out.

 

Agreed it would be useful and fascinating to get that level of detail, but the lack of that detail doesn't mean the current attempts at defensive stats are that bad.

 

"Bad" is subjective.  The point is that the current metrics are missing a lot of important components.  Positioning alone is basically unaccounted for and that matters for more than just shifts.   You couple those weaknesses with all of the interpretation going on and you get something less than perfect.  It's why it's always wise to take these ratings in larger sample sizes than even season.

 

I'm not sure why there is anything other than appreciation for things being less subjective.