Baseball Primer Newsblog— The Best News Links from the Baseball Newsstand
Friday, July 06, 2018
When you find a metric which surprises you, it’s best to dig into the data a little before you say something silly. Anway, and I know this is an outlier, the entire ?thread? is worth reading.

Bookmarks
You must be logged in to view your Bookmarks.
Hot Topics
Newsblog: OTP 2018 September 24: Baseball and the presidency (942  12:26am, Sep 26)Last: Gonfalon BubbleNewsblog: Taking Back the Ballparks  Dodgers voting thread (15  12:10am, Sep 26)Last: QLENewsblog: Detroit Tigers food services employee in custody after video showed him spitting on pizza (31  12:03am, Sep 26)Last: Howie MenckelNewsblog: OT  2018 NBA Thread (PreSeason Edition) (597  11:52pm, Sep 25)Last: there isn't anything to do in buffalo but 57i66135Gonfalon Cubs: The Final Push (195  11:46pm, Sep 25)Last: Man o' SchwarNewsblog: No more Jello Pudding Pops for OMNICHATTER, for Sept. 25, 2018 (72  11:46pm, Sep 25)Last: Walks Clog Up the BasesNewsblog: Marlins Finish With Lowest Home Draw Since 2004 Expos (14  11:40pm, Sep 25)Last: Steve Parris, Je t'aime Newsblog: Carlos Santana at third base in 2019? He’d be open to it  NBC Sports Philadelphia (15  10:43pm, Sep 25)Last: Howie MenckelNewsblog: Five Tool Players  Articles  Bill James Online (45  10:34pm, Sep 25)Last: Jose Bautista Bobblehead DayNewsblog: Mariners’ Felix Hernandez ‘so proud to be an American’ after becoming a U.S. citizen (20  9:50pm, Sep 25)Last: phredbirdNewsblog: OT  CatchAll Pop Culture Extravaganza (September 2018) (427  9:20pm, Sep 25)Last: McCoy Newsblog: Fowler, still owed almost $50 million, eager to be part of Cardinals' future  St. Louis Cardinals  stltoday.com (17  9:00pm, Sep 25)Last: cardsfanboyNewsblog: Bobby Evans’ days as the Giants’ GM appear to be numbered (9  8:22pm, Sep 25)Last: SoSH U at workNewsblog: 10 Degrees: It's the real 'Golden Era' as baseball today has the greatest group of young, prime players in the sport's history (12  8:00pm, Sep 25)Last: phredbirdNewsblog: OT: Soccer Thread (201819 season begins!) (839  7:13pm, Sep 25)Last: J. Sosa

Reader Comments and Retorts
Go to end of page
Statements posted here are those of our readers and do not represent the BaseballThinkFactory. Names are provided by the poster and are not verified. We ask that posters follow our submission policy. Please report any inappropriate comments.
1. Infinite Yost (Voxter) Posted: July 06, 2018 at 05:24 PM (#5706306)Just like Matt Kemp is putting up a wRC+ of 149, somehow. Nobody thinks Kemp is going to keep that up not only next year but this year.
Schwarber has 8 assists already  leading all LFers  including 3 at the plate. Maybe this is purely elective stupidity on the baserunners part and it could correct itself as the season progresses but these assists have occurred and every metric (fangraphs & bbref use different algorithms) will incorporate them.
Passan has been around a while, he should know how to do this.
I mean it's like looking at Avi Garcia holding a .320 BA and then looking at his BABIP, seeing how it's .380 and realizing it's a fluke and the 4 fWAR season in '17 he just put up is highly unlikely to be repeated.
fWAR, at least for position players, only describes what has happened, as you noted (here we have the flukey assist totals) it's not a measure of "true talent" and Passan should realize that.
To quote his final two lines: "Don't believe everything your eyes tell you. And with a sample size this small, don't believe everything the stats tell you, either".
The assists are high because for some reason, runners keep thinking "Kyle Schwarber, shitty defender"  not realizing that if he doesn't get turned around and fall down, he's fine. In fact, I'd go so far as saying that I think his arm is near plus.
I think I've seen more than half of 'em and they're not all illadvised. The one just a couple weeks ago is a case in point  Harrison Bader is a fast guy. On a single down the line to LF, so long as he doesn't slip coming out of the box, that's probably a hit a guy like Bader should be trying to stretch. It took a perfect throw and Kyle came up with one. He did the same to Lindor back in May  it was another long single into the corner that a guy as fast as Lindor probably should be trying to stretch. Some of the others  I remember him throwing the ChiSox's Delmonico out at the plate  were unwise, but he still made a good throw.
Sure, it's a good idea to watch the play lest he do a little keystone cop tracking the ball  but you can usually tell those immediately. If he breaks right on the ball, runners would be better off playing him straight up.
Let's start with basic defensive stats. Last year in 821 innings, Schwarber had 7 assists and 5 errors. This year in 542 innings, he already had 8 assists and just 1 error. Last year he had a RF/9 of 1.62 in a league of 1.83  .2 fewer plays per game in about 91 full games is 18 fewer plays made. This year his RF9 is 1.84 vs a league average 1.86.
Now looking at some fancier stuff ...
TZ at br gives him +10
DRS at br gives him +7
TZ says he's made 6 runs worth of extra fielding plays. Oddly all 6 have come at home which could be real, a weirdness in Wrigley data or familiarity. Still it puts him at dead average on the road. It then gives him 4 runs for his arm (assists and holding runners).
DRS gives him just 1 run aboveaverage on plays made and +5 on his arm (and some rounding to get him to +7 overall). I find that more plausible (or better reflecting his current true talent) than TZ.
DRS at fg reports slightly different numbers, putting him at +8 (5 arm, 1 on "good fielding plays" and 2 on total plays made).
BIZ at fg suggests he's getting a lot more opportunities for some reason. In those 821 innings last year, he saw 112 BIZ, with a conversion rate of 0.893; in 2016, prorated out to about the same number of innings, he's on pace for 135+ BIZ with a conversion rate of 0.900  the difference in rate is trivial but 23 more opportunities is 23 more opportunities  like having more PA, we don't pretend the player didn't add (or subtract) value relative to a player with fewer PA. His OOZ rate was actually a bit higher last year.
UZR at fg  now here we have a quandary as UZR/150 put him above average by a good bit last year (a +9 pace). It likes him even more this year, due to improved throwing and the reduced error rate (+20). I agree that both of those numbers (prorated out to 150 games) look pretty silly but the gap between them doesn't look too silly to me.
Inside Edge at fg ... 2017 vs 2018
Remote (<10%) 0 for 6 ... 0 for 5
Unlikely (1040%) 2 for 9 ... 0 for 2
Even (4060) 0 for 5 ... 0 for 1
Likely (6090) 6 for 11 ... 2 for 2
Routine (90+) 98.5% of 135 ... 99% of 102
So no real change there. He's seeing a jump in opportunities and a much smaller %age of those are nonroutine in 2018. That seems odd but again it's very small numbers. Anyway, there can't be more than maybe a run or two of difference there. Note, Inside Edge is just range/plays made, not throwing.
Statcast at baseball savant ... puts him at 1 compared to an average OF which is probably aboveaverage compared to an average LF (which is what the others are doing). In 2017, they put him at 8. Statcast catch probability is also just about range, nothing about arm. In more detail, again 2017 vs 2018:
5* (025) 0/11 0/8
4* (2650) 1/8 0/2
3* (5175) 1/3 1/6
2* (7690) 6/12 9/9
1* (9195) 13/15 8/8
Again, nothing dramatic (so it's not entirely clear why he's gone from 8 to 1 unless overall OF defense has gone down substantially) but some improvement in range in terms of getting to the not hard but not routine balls which matches my eyeballs. Note statcast and IE seem to disagree on the number of nonroutine opportunities he's had with statcast seeing the two years as reasonably similar.
So pretty much everything except Inside Edge seems to credit him for improved range. That includes the more traditional stat of RF9. This hasn't been mindblowing, it's more that he's come up to being an average LF in terms of range. It's not clear why statcast sees a 7 out difference between the two years; it's not clear why UZR seems to think he has excellent range. The measures which include a throwing component all seem agreed he's added excellent value there. And I suppose one can review video and quibble with the scorekeeper but there's no denying his official error rate is down.
We're not unbiased of course but over in Gonfalon Cubs and in game chatters, us Cub fans have been noting his improved defense all year. Seems real enough to us. He's trimmed down, he's a bit faster, he's making good throws, he's had a couple of plays cutting off hits from going to the wall that we'd never seen from him before.**
And folks don't seem to notice but LF has been a bit of a cesspool for a decade or so now and pretty much no team has a fulltime starting LF. Last year only 5 LFs played at least 130 games there  Gordon, Duvall, Upton, Rosario, Ozuna. That compares with 12 in RF (or 12 at 2B). It's a bit of a dog's breakfast out there  including for the Cubs as Schwarber is only on pace for about 1000 innings and probably just barely 130 games (125 starts). Who knows what rotating lots of guys through LF means for "average LF defense?" It will mean that, for most individual LFs, we'll have even less data in a season than for "starters" at most other positions.
Anyway, I have little problem believing that Schwarber 2018 (to date) has displayed roughly average range for LF and the assists and reduced error rate and increase in the number of opportunities are indisputable.
And to be honest, looking at the table Passan posted, I wasn't surprised by Schwarber, I was surprised by Zobrist. Maybe I haven't seen the right Zobrist games but I'd have marked him average, maybe a bit light on range. I can't recall any really good or really bad plays he's made.
** Which doesn't mean they were necessarily aboveaverage plays, just that, off the bat and hit towards Schwarber, you assumed it would be a double but he held them to a single.
Yup.
Kid's a hard worker and it shows.
He's never gonna be an elite flycatcher  a few guys are Willie Mays, but most are not  but cutting off balls down the line has been a particularly noticeable improvement.
No one thought the home run stat was broken when Brady Anderson hit 50 in a single season.
But pretty much everyone with a pulse realized that something was up with him (and Luis Gonzales for that matter).
Yeah, that's right  they took steroids, it dramatically increased their performance, putting them in position to potentially earn hundreds of millions of dollars ... and then said to themselves "I won't do that again".
I guess that's the origin of the phrase "dumb jocks".
I think there are a lot of folks who think a player's defensive ability is constant over their career  except maybe at the tail ends  from the same mindset that says, "Speed never slumps," although if a guy is dealing with leg injuries, it damn sure does slump. Also there's no widely available metric of defense  other than fielding percentage  but even then people don't really care about that except at the extremes. The guys who have had the long runs of consecutive Gold Glove awards certainly haven't led the league in Fielding Percentage every year of their streaks.
I don't think it's constant over a career, but I do believe that fielding should be more consistent game to game and season to season, because there are simply fewer variables involved in fielding than there are in the other aspects of the game (as well as the other two main prongs involve the direct efforts of the opponent).
It's similar to how Usain Bolt can win virtually all of his 100meter races, whereas the topranked golfer does not.
Sure, the smaller sample size can offset some of the narrower performance band in the numbers. But Pat referenced ability, not stats.
Elvis Andrus, amously, made 3 errors in like 5 minutes
Defense isn't constant
Anyone who's played or watched the game should know that
Your performance is not relevant.
Yup, he did.
I didn't say it was. I said that I believe defensive ability (particularly outfield defense, the primary topic of the thread) is more consistent than offense, because it involves fewer parts. And I've both watched and played a lot of baseball.
Defensively, the (nearly) indisputable stuff are putouts, assists, errors. Unfortunately, we don't have a useful and indisputable measure of opportunity which means we don't have a universallyaccepted rate stat. At that point, we're already a bit stuck but then the fancy stats try to go one better and include a measure of difficulty.
We could do some similar things on offense and we'd probably see similar level of incredulity. We saw some of this back in TroutCabrera when Miggy won the Triple Crown but us nerds were arguing that, once you add in the more nebulous stuff, Trout was way more valuable. More broadly, it's not at all clear why we use AB instead of PA in the denominators for BA and SLG (i.e. how we define opportunity can be debated). But we could also finally bite the bullet, realize that hitters face different sets of pitchers, different mixes of pitches, different defenses, different defensive alignments, etc. and try to incorporate "difficulty" into our offensive measures too. Then we get a situation where we add a Rdiff to WAR.
The closest we have to a difficulty measure on offense is park factor which does occasionally generate disagreement (see TroutCabrera again). If writers bothered to understand the WAR measures, I suspect fWAR's use of FIP would generate a good bit of controversy. As I noted the other day, in Ed Whitson's "disastrous" season for the Yanks, he's credited with 2 fWAR. You think that would have gone down well in real time? I can certainly say that argument didn't go down too well when Glendon Rusch was getting pounded.
All that said, there is legit concern about the defensive measures. It's sometimes quite difficult to reconcile the different measures. Inside Edge and statcast (which should be our most accurate) suggest that OFs simply don't have enough opportunities to affect the outcome. If those measures are accurate at all, the sample size issues are way worse than we let on. If 85% (or whatever) opportunities an OF faces are so routine that everybody this side of Todd Hundley catches 98100% of them, then we're really only talking about a sample of 50 balls a year when an OF can actually make a difference and about half of those are of the "nice play but nothing special" variety. Under statcast or inside edge, it just seems impossible that an OF could add 2530 runs.
Then there's the issue that statcast hasn't done anything for IF yet (that I'm aware of).
When looking at oneoff comparisons  questions like "when did Schwarber get decent?" or "how can they say that Kiermaier is better than Pillar?"  I like to start with those basic measures like RF9. Start at "well, by the measures we all generally agree on, Schwarber is making more plays every 9 innings, making fewer errors per 9 innings and has thrown out a lot of guys." By the accepted measures of OF "events", he has clearly done better so far this year than he did last year. In this particular case, that puts those who think he's still awful in the position of having to argue about whether he's made more plays just because he's had more opportunities, the difficulty of those opportunities, etc.
But when we look at the fancy stats that do try to adjust for the number and difficulty of opportunities, they seem to all agree that he's been better than last year and at least average overall. So now if Passan wants to say those stats are wrong and Schwarber actually stinks, he's got to propose a "better" method of determining opportunities and difficulty. And we all know that's not going to happen, we know Passan is never going to say anything more than "but didn't you see him butcher those flyballs in the playoffs back in 2015?"
** Except by the Trump administration of course.
I'd like to see an adjustment on what constitutes an AB, but I can't think of what the argument would be for using PA instead of AB in those cases. It doesn't seem remotely useful in telling us what both those measurements are designed to do.
why do people keep saying this? My best guess is that in extreme cases an OF could save 35 runs vs an average OF. Broken down as:
20 runs on fielding range, forget what I was using as a value for a missed fly ball, but most misses should be coming at the exteme edges of range, so maybe say half are singles and half are doubles? and half the time men are on base? one can see someone getting to 50 balls more a year than average OF and if they are worth 0.4 runs, there's 0 right there. For example Kiermaier in 2016 is 0.32 PO/asst more than the league average CF; that's what about 48 catches that year? (he had 7 assts so I dont think the difference is there). That's vs the average CF though..
10 on "kills" (assists). Kills on the basepaths are worth about a run last I checked.
5 on holding runners. Usually the better arms keep runners from taking the extra base 10% more than say an avg guy. Its not really a lot, and it varies by era. There's a little over 200 base runner opp. over a full season, and if say he holds 20 runners more, and multiply by 0.25 runs (a crude estimate of moving up one base) there's 5 runs there.
NOTE: I am not sure any OF in history has ever done this because he'd have to be performing very well in at least 2 or maybe three physical aspects of fielding. I mean there's range, there's also arm accuracy, there's also arm strength, there's also instincts. Maybe Willie Mays in a good year did it...
ALSO NOTE: When comparing to an average OF'er, most/all teams are going to play their best OF in CF, so this measure of difference is probably not extreme when comparing say Willie Mays to an average CF, any more that it would be comparing Willie May's hitting to a guy batting in the middle of the order. Its more like comparing Willie Mays having a great year vs some average/good RF.
ALSO: Not sure exactly what is meant by "under statcast..." If you are referring to whatever numbers they are generating? My argument is just based on where I think runs are being saved.
Just to underscore your pt. IT's RIDICULOUS how the two most prevalent defense measures are quite often at odds with one another.
Take Kiermaier; the discrepancy between BIS and TotalZone (Is that the name or Baseball Projection?) from 20152017
23, 15, 12 runs (BIS giving him more runs each year).
And its not like Kiermaier is some obscure players. He's one of the best OF'ers out there. One would think someone that position would get more scrutiny than anyone. Like imagine if we had an argument about Barry Bonds and ARod and we couldnt agree w/in 20 of what their offensive values were.
Looking at Keon Broxton the other day I think he's 7 runs apart on those systems this season.
Its a bad effin joke...
I mean, I think its obvious to say that; buy my hunch is that its not the sample size thing that makes defensive metrics hard to agree upon. We can count innings played, we can count fly balls to the OF. That's not hard. And I have to assume managers position players to maximize their usefulness, if for some reason Ty Cobb was catching less fly balls then Tris Speaker sure as hell his manager would move him to where he could catch the more fly balls. That's reasonable right? and we can measure park effects, and we can measure GB/FB ratios which really dont seem to be a significant factor in any case, although sure as hell someone will bring them up to make some silly pt.
No, I dont think its sample size, its more the other things you mention, degree of difficulty and perhaps how to weight certain plays.
Well, OK at a certain point sample size does become an issue and from what I can tell, its usually when we get to less than half a season that we start to see anomalies that dont make sense. Just going from memory.
It's absolutely sample size though. Of the typical 300 or so balls hit to the outfielder in a given season, about maybe 50 of them are actually difficult plays, with infielders it's more chances but the percentage of routine plays is higher, so the defensive stats are basically only looking at a small number of plays in a given season, and combining them with the mistakes that were made(the errors on the routine plays) and you are going to get stats that is based upon small size. And some systems are effectively punishing players who are following positioning orders of their manager, etc.
The issue shouldn't be the discrepancy in runs between the two, but in whether or not they rank players relatively consistently between the two. Systems are using different run models, just like many of the runs created systems out there for hitters. If one system is saying someone is a good fielder, while another is saying he's a bad fielder, then you need to look into the numbers to see what is going on...but worrying about the whether Kiermaier is a +15 or +20 fielder isn't really important in looking at the systems.
As I have said many times, I prefer to just smooth the players out and lump them in categories (say 15, 10, 5, 0, +5, +10, +15 and +20 with players being lumped into the categories by how they fare relative to their peers over the past two seasons of datait's not a fully accurate measure in determining current season value, but it feels like it's a bit more accurate in determining roughly where the player's actual ability is)
Well that depends on your purpose right? If you want to know the whether we should play Greg Luzinski or Curt Flood in LF then we absolutely have to know the value of his defensive contribution in terms that compare offense and defense. Runs saved, or runs created, but basically "runs" is the standard coin of the realm here.
what makes that a small sample?
are you just making this up, or do you have some reference for this? Cause even without looking up anything or googling anything I'm pretty sure this is wrong.
In what way...the part about infielders having more chances or the part about routine plays? The more chances is easy enough to prove, the routine play is an opinion thing. But just watch any game and count the number of routine plays that any player gets and outside of third base, it seems to me that it would be easy to see that the number of routine play percentage is higher for infielders than outfielders. Simply because even on routine play for outfielders they usually have to cover two to three times the distance than an infielder does.
The 2008 Mets were a decade ahead of their time!
There's no way. for one thing the error rate for infielders on GB is higher. I think its 3.6% what is it for OFer's? Its less not sure but I think less than 1% for fly balls right?
A better way to do it might be to see what the difference is in terms of fielding range between an elite defender, an average defender and a bad defender. I am pretty sure the difference is greater for middle infielders than outfielders. Not sure about corner infielders, you might be right about corner infielders.
THen you have another issue about discretionary chances (in this case it probably more to do with 2b taking the throw from catcher on SB) so you would probably be better off going with assist totals as a better measure of real chances.
Of course if you are going to count every throw and catch in the "routine" play count, you might be right. So lets stop there... Are you counting like every single catch and throw or just getting to batted balls?
In terms of batted balls, Im pretty sure there's a larger difference among middle infielders. in terms of catching the ball and throwing it, you maybe right.
I'm counting every actual play.. meaning every ground ball out, fly ball out. But your 3.6% comment does illustrate something, routine plays for infielders are a bit more difficult than routine plays for outfielders. But I don't think it changes my viewpoint. Players are still being judged based upon their ability to do the relative rare play, and their ability to perform well on the routine, but they are only really being dinged on the routine when they make mistakes, and not getting any real credit for being average.
A guy like Jeter might never make a mistake on the routine play, but he also rarely makes plays outside of the range of the routine... That is actually a nice skill, but his defensive value is not any good overall, and these systems do eventually catch it. But at the same time, they might overrate or underrate a season value of a player because there just isn't enough data... and yes 50... 100, 200 plays is absolutely the definition of small sample size.
MGL/Tango in the book came up with numbers on offensive splits that determine what is an acceptable sample size for offensive numbers and sometimes those numbers required 3000 pa to have any statistical significance.... and with fielding it's even larger.
Scorekeepers don't have a knowledge of the big picture and average, which is what advance defensive stats are designed to compensate for, but they are still left to some of the limits of official scorekeepers.
now you're just taking stuff out of context. You dont need 3000 AB to tell someone is a good hitter and you dont need 3000 Put Outs. Otherwise there might be 5 or 6 guys in the entire league that we're certain about. And even that wont be certain because after 5 years nobody is the same player.
Come on, that's ridiculous.
What do you think they were designed to do? Do you think that's how they've actually been interpreted over the years?
H/PA immediately tells us something useful  how likely is it (has it been) that this batter's PAs result in a hit.
H/AB tells us how likely is it that this batter's PA ends in a hit conditional on the fact it's not going to result in a walk, HBP, SF, SH or whatever else I've forgotten. This is a very odd construction and not immediately clear what we do with it. We have to know the outcome of the PA before it has any meaning. And what good does it do us to say that two players are both 280 hitters when one of them walks 5% of the time and the other walks 10% of the time  what is it about those hitters that BA is saying is the same?
It presumably was originally intended to measure how good of a "hitter" the player is. But what does it mean by "hitter" (i.e. strikeouts count even though they did not result in a hit ball). I also believe this original construction contributed to the importance of walks being diminished for the first 100 or so years of baseball. Certainly when I was a kid, we thought BA told you how good a hitter a guy was and walks were just the fault of the pitcher.
Then there's the more recent issue of OPS where adding OBP and SLG is mathematically ... well bizarre. They're two numbers, you can always add two numbers so you can't really call it incorrect but having the same denominator seems more "natural." There's certain no natural interpretation to OPS, all we've got is "it kinda works." (Not that times on base + total bases per PA has any natural interpretation either ... but times on base + extra bases per PA gives us a "bases produced rate.")
It's not a big deal but the AB is the artificial construction that serves no obvious purpose.
why do people keep saying this? My best guess is that in extreme cases an OF could save 35 runs vs an average OF. Broken down as:
I'm not sure if you skipped over my opening clause or you aren't familiar with Inside Edge and Statcast. Let's use Inside Edge (at fangraphs) since it gives us all the categories.
Let's take 2017 CF. We have the following breakdown of opportunity types:
Remote (110% caught): 574 opps, 6.8% caught (approx 39 caught)
Let's start there. That's all 30 teams, an entire season. That's only 19 such balls per team, that's only 1.3 plays made per team. Sure, each of those plays not made probably results in a double or triple and those are worth (give or take) .8 or 1.1 runs each. Therefore converting a double or triple into an out is worth 1.11.4 runs. Huge value. But there were only 39 of them in all of MLB.
Now all of them:
Remote (110%); 574 opps, 39 caught (6.8%)
Unlikely (1040%): 311 opps, 86 caught (27.7%)
Even (4060%): 225 opps, 124 caught (55.1%)
Likely (6090%): 499 opps, 419 caught (84%)
Routine (90100%): 10,817 opps, 10,763 caught (99.5%)
Setting aside the 1,825 "opportunities" that IE deemed "impossible" (and 0% caught), that's 12,426 flyballs to CF of which 87% are routine. Now when I first saw that number I figured "oh, they're categorizing too many FBs as routine" ... and maybe in the margins they are but given 99.5% of the routine flyballs are caught, they clearly are pretty much routine flyballs. And when you look at that by individual, the rates go from about 98% to 100%. The missed ones here are pretty clearly the "I can't believe I dropped that" or "I lost it in the sun" variety.
So in terms of adding value, your average OF has only 13% of their opportunities in which to make a difference, about one of out 8. Even if they played every team inning in CF, on average that guy would see only 5055 such chances. And on those 5055 such chances, an average CF would make the play about 22 times. Of the plays not made, 18 are in that remote category. Even the best CF is going to do no more than take away 23 extra of those. So at most, the nearperfect CF who catches everything except most of the remote ones can make maybe 35 of these nonroutine plays or about 13 more than the average CF. If they were all going to be triples, you can get that up to about 18 runs.
So, in 2017, Buxton saw 18 remote ones and, impressively, made the play on 5 of them. That puts him 3.8 plays ahead. He saw only 9 unlikely ones, made the play on 6  another 3.5 plays ahead. He saw 9 even ones, making the out on 8  another 3 plays ahead. He saw 23 likely ones and made the play on all of them, another 4 plays ahead. So there he is 14.3 plays ahead. If all were going to be triples that's basically 20 runs. All doubles  15 runs. If any were bloop singles that he robbed, those were just 0.8 runs. (On average blah, blah, blah.) But darned impressive and could be 20 runs.
Awesome  but in 2016 he caught 10% of his remote chances, 60% of his unlikely, 75% of his even and 70% of his likely (below average on that one). I don't know what that adds up to but lower enough that we wouldn't want to consider +20 to be his talent (not that anybody said it was).
Now IE doesn't add up those extra plays for you but statcast does. How close the two systems are in measuring "difficulty" I don't know. But for 2017 Statcast put Buxton at a whopping +29 outs so that could easily be 3040 runs. We have the problem that Statcast is comparing to an average OF, not an average CF but he was likely at least at +20 relative to an average CF by their measure.
So I guess I was wrong  the very best CF might manage 2025 runs from range (which is really all I meant, I should have been clearer about that). LF and RF will generally have fewer opportunities I believe so would have more trouble getting to +20 but Betts and old Heyward might get close. (Betts basically a CF in RF so a sizable advantage there comped to an average RF.)
Come on, that's ridiculous.
No, it's spot on. It's basic statistics. The problem for baseball is that trivial differences are huge differences.
Leaving aside technical stuff, the basic distribution to apply to baseball stats is the binomial distribution  the number of successes out of N trials. This requires knowing p (the probability of success) and the number of trials N. The expected number of successes is pN and the variance around that expected outcome across sets of N trials is is p(1p)N. The square root of that is then the standard error.
So a true 350 OBP (assuming we even know that) and 600 PA is expected to reach base 210 times. The variance on that though is 136.5 giving a square root of 11.7. Plus or minus two SEs is (give or take) a 95% confidence interval. So entering a season, we're pretty confident that guy will reach base between 187 and 233 times. Unfortunately, that's an OBP between 311 and 388  that's way too broad to be any use for us in projecting him forward.
Of course, we're in an even worse situation because we don't know the batter's true OBP, we only have past performance to give us a prediction but that prediction obviously comes with uncertainty. We turn that problem around and say the batter who had a 350 OBP last year probably has a true OBP between 311 and 388. Then, in projecting him forward, we not only have to incorporate that uncertainty, we also have the random uncertainty I covered above. So there's a reasonable chance that last year's 350 OBP player was in fact, say, more a true 330 OBP player then there's a pretty good chance that true 330 OBP player will put up a 310 OBP in the coming season.
Then you get the issue of comparing two players. If you have two statistical estimates (call them p1 and p2) and you want to test whether they are "significantly" different, you form the difference and the standard error of that difference is (assuming independence between the two batters) the square root of the sum of the two variances. If both had 600 PAs and had the sorts of OBPs we realistically see then those two variances will be about equal, meaning the standard error of the difference will be about 1.4 times the standard error of each. So that means, over 600 PAs, we'd need to see an OBP difference of about 50 points to declare with 95% confidence that one of these guys was better at OBP than the other.
The simpler way to remember this is to think about political polling. If you ever look at a poll result in the newspaper, especially a major national or statewide poll, you'll often see them state something like "Candidate X was supported by 52% of the respondents, plus or minus 3%. We interviewed 1,042 eligible voters..." For a proportion of 50% (and due to that variance formula, basically anything between about 30% and 70%), a 95% confidence interval of +/ 3% around a yes/no outcome requires approximately 1000 trials. After 1,000 PAs, we can only state (with high confidence) for an interval covering 60 points of OBP.
Unfortunately, 60 points is a massive interval in baseball. That's not the fault of statistics or sampling variation, it's just the way it is. For a quickie little opinion poll, generally folks are comfy about +/ 3% because nobody cares that much about being precise. When it comes to, say, the unemployment rate, you damn well better get that accuracy down to tenths of a percent. That's why labour force surveys generally have 30,000+ people in them.
And we haven't even talked about regression to the mean yet.
It's true that, when it comes to baseball decisionmaking, this sucks. You can't wait around for 3,000 PAs to know with great confidence that this guy is good ... and, as you note, over that time it's not like everything's constant so you need to add your uncertainty about that in. You have to decide on very little reliable information and wizened scouts' eyes whether to draft a player. You then generally need to decide over maybe 2,000 minorleague PA whether he's ready for the majors and ready for a starting spot. Then, maybe he does well in year 1 then struggles in year 2 and now you've got to decide which is the real one.
Now occasionally performance is so extreme that it's obvious in a very small sample  Ruth, maybe Trout. But it's hard to be that extreme in baseball  even for somebody like Trout, we had no good reason to think we'd see a steady string of 910 WAR seasons and, I'll be damned, he might now be even better. About the best we could say with confidence after his awesome debut (or Judge's) is that they would be good to very good. That is, sure, after that first nearfull season, it was pretty obvious Trout was better than #2 pick Dustin Ackley and even better than the Cubs' pick of Brett Jackson just 6 slots later. But the fact he went unpicked by 23 teams (or 23 slots) points out the massive uncertainty on draft day. And of course while debuting at 19 is impressive, he wasn't that good in the majors and even adding in excellent numbers at AA and briefly at AAA didn't lead anybody to predict he'd be the best player in baseball by May.
I suppose the most obvious example is pitcher hitting (and maybe position player pitching). We can tell quite quickly that your typical starting pitcher is a much worse hitter than even most poorhitting position players. But that's also true statistically. The average NL pitcher is hitting 111/143/142 with a Krate about twice the average position player. By OBP that's about 18% below the average player. You can get a 95% CI of 18% on a difference after a sample of 42  or just start with the obvious historical assumption that this new pitcher is probably a sucky hitter and then don't change your mind until he's hit well for a while.
As far as the OF goes, the high proportion of plays made greatly reduces the random variation in any sample. For 600 opportunities and making something like 8590% of the plays, the variance on that is just 54 to 75 and so the SD is about 7.5 to 8.5, a good bit lower than on the number of times reaching base. But it might take 1.52 seasons to get to 600 opportunities and, if each of those misses/successes is worth 1+ runs (while the value of reaching base vs. making an out is probably around .7), the runvalue of the uncertainty is probably basically the same.
H/AB tells us how likely a player is to get a hit when he has the intent and opportunity to get a hit. H/PA tells us how often a PA ends in a hit. I really don't see the value in the latter, given how walks/HBPs (the true heavy lifters) skew the AB/PA question for every player. SF/SH are virtually irrelevant, for all but the one position we don't care about.
At the extreme end, Barry Bonds led the NL in BA in 2004 with a .362 BA. His H/PA was .219. I think the former presents a much better picture of Barry Bonds rate of base hitting success than the latter (even if you adjust every hitters' WaltBA to this new H/PA formulation, that .219 is going to look pretty crappy).
AB may be an artificial construct, but it's a sensible one. It removes the PAs that happen when the player either wasn't trying to get a hit (the sacrifice*) or when the PA ended without him having the full opportunity (the HBP or BB, either of the intentional or unintentional variety).
* The flaw, of course, being that the sac fly should count as an AB for the same reason the runscoring grounder should. We don't have any way of knowing that the player was sacrificing himself when he did it.
And how does your proposal solve this? You would still have situations of two .280 hitters when one of them walks 5 percent of the time and the other walks 10 percent of the time. They would just be guys that are hitting .285 and .295 now.
The one you forgot (there may be others I'm forgetting) is C/I, which follows the same basic formula (no opportunity). And happens even less frequently than SF/SH, so it's even more insignificant.
i said at the outset that I was comparing the best OF to an average outfielder; not comparing CF to CF. Because obviously teams put their guys with the best range in CF. So by analogy if I am talking hitting, I would be comparing Barry Bonds to an average hitter not to someone hitting in the fourth spot.
OK so we agree. Great!
Walt: The gigantic flaw I see in your reasoning is that the outcomes produced by batters are not the same type of outcomes that were first used by Poisson or whomever when science behind probability was developed.
Those were outcome that were either yes/no or some sort of linear number e.g. 9.5.
BUt outcomes in hitting arent like that are they? Each outcome is not a yes/no, its producing at least four separate sets of data:
1. there's contact, did the batter make contact or strike out?
2. there's OBP did the batter get on base, this brings up his batting eye
3. bat.avg. did he hit the ball hard enuf to get on base?
4. there's power did he hit a double or some other extra base hit.
So I grant you if you're only 2 possible outcomes to an AB then you, and MGL and CFB and POisson might have a point. BUt that's the data we get from an AB its far more detailed than that.
One could also bring up BABIP, pitches faced, etc. but I think four is enuf for now.
So here's my question to you:
Take an imaginary batter, he bats say 30 times, he strikes out 25 times, and gets one hit a single. no walks.
1. Can you use the statistical methods you invoked in your recent post to create a level of confidence USING ALL FOUR variables that i referenced above?
2. Has anyone in baseball history managed to overcome such an ignominious start to a career? If so why not?
MATH ERROR.
Ok literally what you say is correct, 1.3 per team. But we are not talking teams we are talking about the chances for an individual to catch a particular fly ball.
If there's 19 remote fly balls per season/team; thats more like 30+ chances for a fielder to catch one of those remote FBs.
Cause nearly every ball that is difficult to catch is difficult for two fielders not one. The exception would be balls hit near the foul line and balls hit directly behind an OFer. I dont know how many that would be but Im guessing most of these would be within extreme range of two outfielders.
So all your comments about how few chances OFers get I think we have to probably add 50% to whatever you were suggesting back there (was it 55 balls per year? on the remote fringes?)
Because just because Billy Hamilton dives and fails to come up with a catch doesnt mean only he had a chance at that. Its quite likely the LF or RF also had a similar chance at that. I mean not really but if they had his speed they would have a chance...
right?
Broxton didnt play an entire season in CF (or anywhere else). He played 1028 inn. Doesnt that mean if we were to extrapolate to an entire season we'd have to add another 30% or so? Not sure cause not sure what that statcast number means ...
well lets see. He played 511 inn. That's what 38% of a full season? Going with some rough math here: lets say there's about 20 fly balls in a full season for each of the 3 toughest boxes, remoteunlikelyeven. That seems to be what you're saying when you say that most outfielders are only dealing with about 55 balls per year... Lets call it 60 to round it off. Going with the percentages you give I count 13 more balls that he would have caught in 2016 if we extrapolate to a full season.
Ok that's about 15 runs. Not bad.
Jacoby Ellsbury weeps silently.
I think you left out Reached on Error.
Willie Mays weeps silently.
ROE is an AB.
Bobby Tolan put in 2 seasons in the early 70s where he outperformed avg CF by 0.2 chances per game (I think its '70 and '72, he had an offseason injury in '71 I think)
Dom DiMaggio outperformed avg CF by 0.2 (all of these per game) from age 2532. From age 3032 he outperformed them at 0.25+. A nice 3 year peak. Of course he missed 3 seasons to the war so there's that. He also droped a little at age 33 and a lot at age 34, exactly what you would expect despite whatever lack of confidence you feel these stats have. He had a very good arm for his entire career ending at age 35, he always had a good amount of assists. In a 162 game season he could probably save 40 runs above avg on range and a handful more on his arm.
Richie Ashburn. In his first 11 seasons (starting age 21) he has 8 seasons where he is at 0.5 chances above avg; 3 seasons where he was 0.1 or 0.2; I dunno if he's not getting discretionary balls or he's hurt or what. He also has more assists than DiMaggio and sometimes in the 1820 neighborhood of Clemente or Sierra. His drop off at age 32 is clear and understandable. The data seem to support what people saw, there's little reason to think we need 3000 put outs to establish some certainty about this.
TZ gives him about 1620 runs positive per year (usually closer to 20). Even if he's getting a lot of discretionary chances, he's still got to get credit for what 0.3? That plus his arm strength you can make case for him saving 50 runs a year. Not saying for certain but there's evidence. It shows limitations of TZ.
Mays is usually averaging about 0.15 to 0.25 for most of his early years. He had some Ashburn type years in the early 60s in Candlestick not sure what is going on there. His assists are really good up to age 29. I can see him getting to 30 balls a year above avg for the first decade of his career and his assist numbers probably provide 810 more runs.
HIs hold numbers were always very good. His career in CF he holds runner about 4% better than avg. The first decade, doing some quick math, it looks like he's about 6% above average. So with 200 runners, that's 12 extra bases, or about 3 runs there.
So a young Willie Mays you can make a case for 40 runs saved (30 + 8 + 3).
Also fly balls to the OF have been steadily dropping at least since the 1940s. I really detest the present TTO trend. i like a game that tests players on multiple dimensions of talent.
Paul Blair; under age 30; he has about 4 seasons where he's 0.3 above average; and age 30+ he has 3 more seasons with about 0.2+. He has good assist totals throughout. In those 7 peak years he's about 7.5% better at holding runners, so a peak Paul Blair might be:
almost 50 (40 or more + 5 + 4)
Curt Flood: from age 2331 he's about 0.25 above avg; he has fair assist totals but nothing outstanding I think he's avg or maybe a bit above and holds: his about 2% below avg. So assuming the 0.25 is a good number we get:
40 + 0 1 = 39
so again there's discretionary chances issues, which we cant be entirely sure but I think going across all these great CFers we can see enuf congruence to make some assumptions about how many runs an elite CF can save you..
How is using PA as the denominator going to give you more useful information? Taking an extreme case, two players from the 2004 Giants:
Barry Bonds BA .362 H/PA .219 TB/PA .491
Marquis Grissom BA .279 H/PA .259 TB/PA .417
What is it about H/PA that tells us that Grissom was a better hitter than Bonds?
I think traditional batting average tells us that Bonds was an elite hitter, while Grissom was slightly above average. You have to look at OBP to get the full picture, but that would be true of H/PA also. To get the complete picture, we would also have to look at SLG.
TB/PA at least tells us that Bonds was the better hitter, but does not indicate the massive difference in their values, 10.6 WAR to .6 WAR.
OBP and SLG, whether you add them, multiple them, or look at both separately tell you much more than either BA or H/PA.
Yeah, that's right  they took steroids, it dramatically increased their performance, putting them in position to potentially earn hundreds of millions of dollars ... and then said to themselves "I won't do that again".
I know this was a throwaway comment, and I tend to be skeptical of the Brady Anderson PED theories for the reason you noted. But I find it pretty hard to believe this guy was clean. And Gonzalez did have a good 5+year stretch of improved batting and SLG  it wasn't just a single season.
Anderson was certainly ripped, and I wouldn't be shocked to find out he was on something, but he did look more natural to me. But I give him a bit more benefit of the doubt.
You must be Registered and Logged In to post comments.
<< Back to main