Baseball for the Thinking Fan

Login | Register | Feedback

btf_logo
You are here > Home > Baseball Newsstand > Discussion
Baseball Primer Newsblog
— The Best News Links from the Baseball Newsstand

Thursday, November 05, 2020

Tango: Statcast Lab: Batter-Runner vs. Outfielder

How much can StatCast tell us about the impact of defensive reputation on defensive performance?

If an outfielder has a great arm he can throw runners out as they try to advance, and we can measure that from outfield assists. But an outfielder whose skill leads the runner not to try to advance isn’t going to see it in his assist totals, or even advanced metrics.

In the linked article, Tango is trying to tease that out of StatCast data.

JBJ had 13 runners that had to make a choice, and only two of those runners went for it (and one of those was right near the go-line). Two out of thirteen is 15%, which is well below the league average of 35%. So we can conclude that runners have a good amount of respect for JBJ. JBJ is able to freeze the runners.

Link is to part 2, which in turn has a link to part 1. Both are worth reading. Part 1 shows how he derived the chart at the top of part 2, and then he takes it from there.

My own math, extending from TFA… Using the above example, if we consider an advance from 1B to 2B as being worth about 0.2 runs to the offense, then by Tango’s numbers Bradley saved about half a run just on arm reputation, just on batters not trying to advance to 2nd on a single, in the shortened 2020 season.

Half a run is… not much. OTOH, in 2020 advanced metrics are saying he saved around 3 (per Fangraphs) to 7 (per BB-ref) runs in total, without looking at this. And at the other end of the spectrum, Corey Dickerson cost his team roughly half a run because everyone ran on him (going from 1st to 2nd) when they had a reasonable chance.

What this looks like in a normal full season, and with other scenarios included (batter 2B to 3B; runner 1B to 3B on a hit; runner tagging up on a flyball), might end up being material. Some of the other scenarios might happen less often, but with much bigger run impact. We’ll see.

Anyway, I thought this was interesting, and another example of StatCast allowing us to understand the defensive impact of all these things we talk about as significant but haven’t measured. Even if arm reputation is found to be immaterial in general, that’s a plus, because then we can stop discussing it as though it is. And if it is material, all the better to measure it. I’m very much looking forward to seeing where this goes.

villageidiom Posted: November 05, 2020 at 09:07 AM | 13 comment(s) Login to Bookmark
  Tags: advanced metrics, defense, defensive metrics, outfield, statcast, tangotiger, throwing

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. Pat Rapper's Delight (as quoted on MLB Network) Posted: November 05, 2020 at 10:50 AM (#5987363)
Does this tell us anything that the Advanced Fielding metrics at bbref (via Retrosheet) don't tell us? With so few examples cited for each OF (13 for JBJ, 11 for Merrifield, 11 for Harper, 4 for Dickerson, 5 for Stewart), it would seem to be highly susceptible to small sample sizes where flipping one or two results (unusually fast or slow subset of runners, Angel Hernandez as 2B ump, etc) would have a big effect on the conclusions.

BBREF's data shows baserunners hold 44.5% of the time when presented with the opportunity to take an extra base on the CF, but they hold 46.6% of the time on Bradley and he's had 3 seasons at or over 50% hold rate. That implies the same thing (runners respect Bradley's arm), but even with over 100 samples per season the numbers are still volatile (he's also had 2 seasons below 40% hold rate mixed in with those 50+% hold rates).
   2. SoSH U at work Posted: November 05, 2020 at 10:58 AM (#5987364)
This is along the lines of the argument I've made that in perfectly operating baseball universe, outfield assists should be relatively* equal over time between the strong-armed and the Damoned. Outfielder arm strength should be factored into the "send" equation, and if players/coaches read the situations correctly, including the breakeven point on a given play, JBJ and Juan Pierre would end up with similar assist totals.

* Not completely, given the occasional force play situation would favor the Jesse Barfields of the world.
   3. Rally Posted: November 05, 2020 at 11:17 AM (#5987368)
Does this tell us anything that the Advanced Fielding metrics at bbref (via Retrosheet) don't tell us?


Using Statcast will get you meaningful data quicker. All singles are not created equal. A single off the right field wall, where it's fielded cleanly and the batter is slow enough that he doesn't try for second, should be very easy for a runner on first to get to third. A liner right in front of the right fielder would be tougher. Statcast can adjust for this. Retrosheet/PBP doesn't, but the hope is that things will even out over larger sample sizes.
   4. Walt Davis Posted: November 05, 2020 at 03:38 PM (#5987405)
Weird some of the stuff my brain has stored away (and stuff it's decided I don't need to remember) but ... a Cub TV moment from the Hawk years. Some pretty fast runner on first when there's a single to right. Comes chugging around second fully expecting to motor to third. Just as he's rounding, the camera switches to him and ... you see this look of confusion on his face (probably as he picks up the 3rd-base coach telling him to hold) then the dawning realization of who is in RF and he screeches to a halt and goes diving back into 2B just in case.

But sure, this is going to be very much in the margins -- few opportunities and the impact of any success (from the OF's perspective) is frequently non-existent (runner would or would not have scored later anyway)

Re #2: In his rookie season, Vince Coleman had 16 assists.** He matched that one other time and had a couple other seasons over 10. FWIW, Lonnie Smith also had a run of years in the mid-teens but I don't recall how weak his arm was per se vs. just taking bad routes. Bradley's career high is 13 assists; Dawson's is 17 but his season totals look similar to Coleman's; Barfield though had a run of high teens, low 20s. On the other hand, Damon only made it to 10 assists once but possibly the model falls apart at the extremes where the arm is literally a noodle.

** Or maybe not. At b-r, his overall OF assist total is listed as 16 but he's credited with just 9 in LF, 0 in CF and 2 in RF and 5 seems like a lot of assists to not know where he was at the time. Of course being Coleman, maybe he was lined up in foul territory. :-) But Barfield had two seasons where he had one more assist in RF than he had overall in OF.
   5. SoSH U at work Posted: November 05, 2020 at 04:11 PM (#5987410)
On the other hand, Damon only made it to 10 assists once but possibly the model falls apart at the extremes where the arm is literally a noodle.


Like I said, that's what you'd expect to see in a perfectly operating world, where baserunners and third base coaches were much better at evaluating the breakeven point of a given play, which would factor in the arm strength of the outfielder. That's not the baseball world we inhabit.
   6. Rally Posted: November 06, 2020 at 08:03 AM (#5987510)
Lonnie Smith made a ton of mistakes but was excellent at recovering. Bill James once wrote about his excellence at damage control. A lot of his assists were probably something like "ball gets away from Smith. Runner is trying for second. Here's the throw, and he's out!"
   7. villageidiom Posted: November 06, 2020 at 08:27 AM (#5987513)
Using Statcast will get you meaningful data quicker. All singles are not created equal. A single off the right field wall, where it's fielded cleanly and the batter is slow enough that he doesn't try for second, should be very easy for a runner on first to get to third. A liner right in front of the right fielder would be tougher. Statcast can adjust for this. Retrosheet/PBP doesn't, but the hope is that things will even out over larger sample sizes.
Not to keep beating the same drum, but you want things to even out over larger sample sizes if you want to use it to project. You don't care if there's some random variation in their actual accomplishments if you're measuring actual accomplishments. Like, if Mike Trout has in a given season a disproportionate number of catches that rob someone of a HR, well, that's far greater value than if he's catching normal fly balls, and it's even greater value than if the balls went over the fence for home runs. You can't project he's always going to get a disproportionate share of HR-robbing opportunities, so yeah, you want that controlled for in projections, possibly by treating those catches the same as any other. But if you want to see how much value he actually produced in that season, it's all about what he did in the opportunities he had.
   8. Ron J Posted: November 06, 2020 at 10:37 AM (#5987531)
#6 Or runner assumed he'd mishandle the ball. Surprise. Not this time.
   9. Rally Posted: November 06, 2020 at 10:37 AM (#5987532)
I think you need a large sample size on this kind of data (using PBP) to infer actual accomplishments. I don't know how big is big enough, but let's say Trout in a small sample (like a 60 game season) has 15 singles hit to him with a runner on first and all 15 times the runner makes it. That looks like he was well below average on holding runners by PBP data. But Statcast might be able to tell you that based on the location of hits relative to where Trout was playing, he had no chance to stop any of them, and even Roberto Clemente or Jesse Barfield would not have been able to throw them out. In that case, Statcast will correctly say he didn't cost (or add) any runs for his team.

What I just wrote applies even if you care not a bit about projections.
   10. Mefisto Posted: November 06, 2020 at 01:30 PM (#5987561)
The amazing thing about Clemente is that even after he'd been in the league long enough for everyone to know that it was suicidal to run on him, he still piled up huge numbers of assists: 22 at the age of 23 (18 at 20, 17 at 21); 19 at 25; 27 (!) at 26; etc. Even at the age of 32 he had 19 (or 17 -- BBREF is weird here).
   11. villageidiom Posted: November 06, 2020 at 01:51 PM (#5987566)
I think you need a large sample size on this kind of data (using PBP) to infer actual accomplishments.
How many home runs does Trout need to hit before they are called home runs?

(Yeah, yeah, rejected Bob Dylan lyrics, whatever.)

We might be talking past each other. I mean, yeah, I agree in the sense that you need a lot of PBP data to see what the typical attempt rate is on those kinds of hits, and the success rate as well, in order to assess any one single to center. I'm just saying once you have all that PBP data, you can assess actual accomplishment on a sample as small as one hit. That one-hit sample tells you nearly nothing about how often he's likely to prevent runners from advancing in general (which is what we'd want in a projection), but it does tell you whether someone advanced.

If you mean that there's no material value in that one observation, well, yeah. In that same sense, there's nothing material in one stolen base. Someone who only plays in one game, appearing as a PR, stealing 2B, and doing nothing else, will have a WAR of +0.02 on the year. Hundredths of WAR aren't material. But +0.02 is still the measure of value from his actual accomplishment. He did steal that base.
   12. Walt Davis Posted: November 07, 2020 at 06:40 PM (#5987760)
Trying to referee here ... what Rally is saying is that, based on heaps of historical statcast data, we (should) have a more accurate model. That more accurate model allows us to better categorize each future event which then better allows us to evaluate what happened in small samples. I'm not seeing that projections have anything to do with it but sure, the next quetsion is how large does that small sample have to be before we start drawing conclusions on how good a defender a particular player is.

Basically all I read Rally as claiming is that, using pbp data only, we might say something like "on average, only 65% of runners advance to 3B on singles to CF but this year 80% did against Trout so he had a bad year" but using statcast data we'd say "on average, given the speed/trajectory of singles relative to his positioning and speed of runners that Trout encountered, runners would make it to 3B 75% of the time so Trout was basically average this year." That can be boiled down to one play obviously.

On projections, the better your "model", the smaller the sample you need to draw inference. Data offering more information (variables) and higher measurement accuracy vastly helps build better models. I think all that Rally was claiming is that "the better the data, the smaller sample ("more quickly") you need to draw valid inference." That's kind of tautologically true. It's possible that statcast is adding so much more noise than signal greatly increasing the chance of user error -- that would be true if it was (a) adding more variables but those variables aren't actually useful or (b) the variables it adds are so poorly measured as to be pretty useless.

Basically all we really had before was a human assessment of what zone the ball landed in and a rough categorization of how it was hit. Statcast gives us those things much more precisely plus (hopefully) precise measures of where the fielder was positioned before the ball was hit and the speed of the runner (in terms of how fast he ran on this particular play or how fast he is in general). The superiority of the latter data helps us whether we are assessing the value of this specific play, the seasonal value of the fielder or the historical value of the fielder for use in a projection. Arguably you get the value of the single play just as "quickly" regardless of what historical data you use but the better data will, on average, give you the more accurate evaluation of that single play.
   13. Walt Davis Posted: November 07, 2020 at 07:01 PM (#5987765)
#6 ... maybe yes, maybe no. Can we tell the difference between "Lonnie Smith is really good at recovering" (he had a lot of practice!!) vs. "runners take extra chances against Lonnie Smith because he stinks?"

For an obvious extreme example ... In 1970, Bench threw out 30 base-stealers; in 1996, Piazza threw out 34. We wouldn't conclude from that that Piazza's good footwork helped him overcome his crappy arm (or vice versa). We'd coonsider that a pretty silly conclusion. We'd note that Bench threw out nearly 50% of the attempted steals while Piazze managed it just 18% of the time (staggering really, 155 steals allowed).

Chances are Lonnie had a reasonable numbrer of assists because twice as many tried to advance on him as on an average LF. The question/issue I was raising is that this might occur if the player has a crappy arm or because they are slow or because they take a bad route or because they frequently boot the ball (although this probably shows up as an error) or some combo thereof.

#5 ... on this sort of thing, I think we are reasonably close to a "perfectly operating world." Lots of poor fielding OFs get high assist totals precisely because teams run on them. One thing we have to factor in that we usually don't is the score -- even with Lonnie in LF, there's not much point risking it when you're up/down 5.

You must be Registered and Logged In to post comments.

 

 

<< Back to main

BBTF Partner

Dynasty League Baseball

Support BBTF

donate

Thanks to
Vegas Watch
for his generous support.

You must be logged in to view your Bookmarks.

Hot Topics

NewsblogNBA 2021 Playoffs+ thread
(1785 - 12:05am, Jun 21)
Last: rr: cosmopolitan elite

NewsblogJuneteenth Weekend OMNICHATTER, for June 18-20, 2021
(133 - 11:29pm, Jun 20)
Last: ERROR---Jolly Old St. Nick

NewsblogCollege baseball pitcher Sang Ho Baek dies after complications from Tommy John surgery
(8 - 11:28pm, Jun 20)
Last: Tulo's Fishy Mullet (mrams)

NewsblogRays call up top prospect Wander Franco
(1 - 10:32pm, Jun 20)
Last: Walt Davis

NewsblogThe Padres Need a First Base Upgrade
(2 - 5:25pm, Jun 20)
Last: Joyful Calculus Instructor

NewsblogOT - Catch-All Pop Culture Extravaganza (June-August 2021)
(3 - 2:46pm, Jun 20)
Last: Ben Broussard Ramjet

NewsblogOT Soccer Thread - In Which Euro 2020 Is Played in 2021
(215 - 2:16pm, Jun 20)
Last: AuntBea odeurs de parfum de distance sociale

Sox TherapyAnd Breathe
(16 - 10:32am, Jun 20)
Last: John DiFool2

NewsblogBaseball Reference Adds Negro Leagues Statistics, Rewriting Its Record Book
(151 - 9:03am, Jun 20)
Last: ERROR---Jolly Old St. Nick

NewsblogHow baseball’s response to RFK’s death inspired some players to fight back and sit out
(1 - 8:45am, Jun 20)
Last: willow19

NewsblogFrom the spitball to Spider Tack: A brief history of foreign substance use by pitchers in baseball
(2 - 12:46am, Jun 20)
Last: the Hugh Jorgan returns

Newsblog2021 NCAA College World Series schedule, live stream, time: Vanderbilt eyes repeat as eight-team field is set
(1 - 8:46pm, Jun 19)
Last: Cblau

NewsblogHe Made Sticky Stuff for MLB Pitchers for 15 Years. Now He's Speaking Out.
(22 - 5:03pm, Jun 19)
Last: Captain Joe Bivens, Elderly Northeastern Jew

NewsblogEmpty Stadium Sports Will Be Really Weird
(13243 - 10:40am, Jun 19)
Last: Never Give an Inge (Dave)

NewsblogPrimer Dugout (and link of the day) 6-17-2021
(5 - 2:24pm, Jun 18)
Last: snapper (history's 42nd greatest monster)

Page rendered in 0.3142 seconds
48 querie(s) executed