THE BOOK--Playing The Percentages In Baseball

I asked this question on Devil’s post, but I thought I would ask it here as well.

I haven’t done much reading on catcher defense. Does anyone know if any work has been done on quantifying the catcher’s ability to frame pitches? It seems that with Pitch f/x data, we could measure called strikes outside the zone/balls inside the zone, and assign a run value to each catcher.

Maybe someone has already done this type of study, and I missed it.

#1,

Dan Turkenkopf has tried to. Here’s one of his pieces on it: http://www.beyondtheboxscore.com/2008/4/5/389840/framing-the-debate

RJ, I think Dan later admitted that was most likely just random variation among umpires rather than an actual catcher skill.  When I ran a similar study for pitchers, I got a similarly high variance:

http://www.hardballtimes.com/main/blog_article/measuring-the-umpires-affect-on-the-game/

Not sure what happened to my first post, or if it got snagged by the spam filter, but I’d also like to suggest that maybe catchers should be judged by SB/Baserunner.  CS% doesn’t properly reward catchers that are good enough for opponents to know not to steal against them.

Well, that chart sure matches up with my lying eyes when it comes to the three Yankees catchers.  JoPo we know is below average, sometimes flat-out bad (but that bat, that bat!!).  Molina, to me, has declined defensively, which is why I went nuts when Girardi decided to give him playoff starts (offensive black hole, merely so-so glove).  Cervelli looked really, really good out there (the bat, though...), which matches his rep from the minors.  I’d rather he start than Molina (Burnett also did very well with Cervelli catching, IIRC, so if one must make decisions based on a handful of starts, I say choose the best guy: Frankie C!).

Good stuff.

Is anybody else troubled about adding the linear weight value of a successful steal to the value of a caught stealing to calculate the run value of successfully throwing out a runner?  Doesn’t seem right to me.

Valuing discouraging the attempted steal is important because of the positive run value when a runner is moving on a hit ball. 

These two comments probably cancel each other out to some extant. Devil fingers rank order of the catchers is probably close, but the run values are probably a little off.

Before I go any further, I should make clear that for the purposes of this post that I assume that it is well-established that the differences between catchers with regard to gamecalling (as expressed in, e.g., “catcher’s ERA") are negligible, and thus such issues will not be discussed in this post.

Until someone can point out a methodological error in Tom Hanrahan’s study (http://www.philbirnbaum.com/btn2004-11.pdf.) I don’t think anyone should assume that it is “well established” that gamecalling skill differences are negligible.

Peter I was troubled when I first read it, but I didn’t bother to try to prove whether he was right or not.

Let’s try to do that.  Peter is talking about this:

So we total the CSctch +SB to get total stolen base attempts (SBA) and then to total CSctch/total SBA for the lgCS rate. We use the weight of .63 runs for each caught stealing, which represents the average linear weight of the caught stealing (.44 runs) plus the weight of the stolen base not achieved (.19 runs). The formula for runs above/below average for each catcher is thus (CS - (lgCSrate) * SBA) * 0.63.

First, let’s do it the right way, and then let’s see if the above works equivalently.

Let’s say that for every 1000 runners on first base, there are 100 SB and 50 CS.  The league average would be 50*.44 - 100*.19 = +3 runs per 1000 runners.

So, a guy with 100 SB and 0 CS would be -19 runs, and a guy with 0 SB and 50 CS would be +22 runs.  And relative to average, it would be -22, and +19 respectively.

Devil is saying:
(CS - .333*(SB+CS))*.63

Which is:
(.667* CS - .333*SB)*.63

So, the guy with 100 SB and 50 CS would come in at 0.

The guy with 100 SB and 0 CS would come in at -21.

The guy with 0 SB and 50 CS would come in at +21.

Seems to work fairly well…

I think the formula devil fingers was using in that case came from my post:
http://www.basement-dwellers.com/2007/11/player-value-part-3c-fielding-catchers.html

I agree that it’s sort of an odd way to present it--when I went back to this recently to implement the method for another project I did a double-take.  I thought I had worked out at the time that the math was equivalent, though, and thus saved a step vs. working it out separately for SB’s and CS’s.

I don’t really have time to look into it now to see what I did to verify this, but Tango’s check suggests that it’s not equivalent.  If so, I’m happy to do it the other way.  Makes no real difference to me.
-j

@Aaron/4, slyde (at redreporter) and chuckb (at BtB) did this earlier this summer, referring to it as “rep runs.” I’ve never been comfortable with it, though.  If a team is running on a catcher, but they’re not being successful above the break-even point, it’s actually to the catcher’s team’s advantage that this team is running.  Running or not running is sort of irrelevant in my view, the question is whether they are successful or not when they do run.

I’m willing to be wrong about this, but so far I haven’t seen an argument for reputation runs saved that makes sense to me.
-j

I agree.  If you have superman catching, no one will run, in which case, it’s a disadvantage to the fielding team.

Ideally, all teams would be stealing at the breakeven point and above, so that the superman catcher will look better than all catchers.

But, this doesn’t happen.

And, generally speaking, catchers with reputations for great arms (e.g. I-Rod as a young catcher) have awesome runs saved numbers, despite the fact that “no one” ran on them.  It seems to work well, at least qualitatively.
-j

"But, this doesn’t happen.”

We really don’t know if that is true, because none of the analyses of SBAs (that I’ve seen) deal with the problem of busted hit-and-runs, which account for a significant proportion of CS.  These analyses are measuring the main cost of failed H&Rs, but not capturing the benefits (runners advanced, DPs avoided). 

Similarly, to fully assess catchers I think you would need to try to measure their impact on base advancement and DPs.  Maybe using something like Dan Fox’s baserunner metrics (there may be others just as good—I just happen to know his).  potential value of a “shut down” catcher is fewer runners advancing to 3rd on singles, less scoring on doubles, and more DPs—not just the CS total.

Guy: good points.  I guess I should say that my guess is that this doesn’t happen.

Or, the recorders can mark plays as straight steals and busted h/r.

Peter/6:  The Hanrahan studies are interesting.  He finds that team ERA is higher for a 4-7 year catcher than a rookie catcher, and then a similar pattern looking at matcher pitcher-catcher pairs.  But I wouldn’t consider it definitive either.  A couple of thoughts.

For the matched pairs, I think age is potentially a very serious problem.  A pitcher who is still with same team 4-7 years later will tend to be a young pitcher (because many pitchers have short careers, and changing teams is more likely after age 27).  So in 4-7 years these young pitchers will be in their prime, and thus better.  All his examples of big improvement—Koufax, Short, Hudson, Mulder—were 22-24 when the catcher was a rookie. 

I don’t totally understand his prior-year study, but I’m not sure it proves anything.  Any group of pitchers who throw 100 innings in back-to-back years will tend to get worse in year 2, because of selection of bias (those unlucky in year 1 don’t return). 

The fact that teams with rookie catchers are weak teams, and therefore tend to improve in subsequent years, is another complication.  Perhaps the teams are improving defensively.  In any future studies, it would be good to break down pitcher performance into components, so we can see if any improvement comes in Ks and BBs (plausible catcher influence) or BABIP (less plausible, likely defense). 

One possibility is that he’s measuring pitcher-catcher familiarity, rather than catcher age.  This would still be a kind of catcher ERA “talent,” but one that any catcher could gain with experience with a given pitcher.  So it would be interesting to see if the same pattern occurs for a veteran catcher who learns a new staff.

Tango - How are the recorders supposed to be able to differentiate a straight steal from a hit and run?

I guess they don’t need to differentiate.  You can note the point at which the runner took off.  And note if he even looks back to the batter.  If the batter takes a pitch in the strike zone, or close to the strike zone.

He doesn’t have to pass judgement whether it was a hit and run or not, but at least tell us what he sees so that we can infer what the runner believed the play was.

Yadier Molina is by far the best fielding catcher in baseball...Laird is 2nd, more than likely, but even then, he’s not really that close to Molina…

I didn’t remember posts 15 & 16 when I did my post to THT Live on the hit and run:
http://www.hardballtimes.com/main/blog_article/yadier-molina-hit-and-run-plays-and-ideas-for-a-study/

You can note the point at which the runner took off.  And note if he even looks back to the batter.

I did not find either of these things to be helpful in identifying the hit and run.  Based on what I saw, I find Peter’s comment at THT quite plausible:

I have speculated, as have others, that the conventional hit and run, where a batter is required to swing at the pitch, has largely been replaced by what used to be called a “run and hit”, where the batter knows that the runner is attempting to steal but only swings at pitches at which he would normally swing.

I did see a couple weak swings at pitches well outside the strike zone by batters trying to protect the runner, but I did not see any obviously bad jumps by the runners, and no runners looking back at the plate.  I do think that the traditional hit and run may well have been largely replaced by something where the runner is expected to get the best jump he can without being picked off.

The important issue to me is not whether a play resulting in a CS or SB was a hit and run, a run and hit, or a straight steal.  Peter may well be right that the first category has largely disappeared, and that there’s no meaningful way to distinguish the latter two.  What’s important is finding a way to identify the plays on which the runner goes with the pitch and the ball is put into play, and to integrate that into analysis of basestealing and catcher defense.  THAT is the missing element.  If we do that, I think we will learn that teams employ the “Runner Sent” strategy—which should be our unit of analysis, not the “stolen base attempt”—in a much more rational way than most prior analyses have indicated.  Sending the runner reduces DPs and leads to more base advancement, and that’s what we need to measure.

For catchers, it’s potentially the reverse:  increasing DP rates and preventing runner advancement.  Since Molina became their regular catcher, StL has ranked 1st, 4th, 7th, 4th, and 2nd in DPs in the NL.  They must be 1st or 2nd overall for those 5 years.  StL OFers have also had a bit above average Arm ratings, suggesting some success in preventing runner advancement.  This obviously doesn’t prove anything, as DPs are influenced by pitchers and infielders, and runner advancement by OF’s arms.  But it suggests to me there may be something new to discover here about the value of catchers shutting down the running game.

Excellent points, Guy/19.  Somehow I feel like that post is going to turn out to be important, a la Simon Cowell “I have a horrible feeling that song could be a hit” to General Larry Platt re “Pants on the Ground”.

Has nobody looked at the running game that way, and how can we pretend to understand the “breakeven point” for the running game if we haven’t?  And with what year’s data did Retrosheet start including the “runner sent” designation?  That data is all there!

I have done the analysis.  It is quite involved.  I have not published it, but I did mention some of my findings several years ago in a thread here.  People seemed skeptical.  The runner in motion information goes back to 2004, but I think that it was not consistently applied to parks until 2005 if I remember correctly.

When I said I have done the analysis, I should have said that I have done it from the perspective of the runners.  Guy’s point that it also affects our evaluation of catchers is an excellent observation and I have not looked at the data that I got in that way.

I have, though, incorporated my findings on how we evaluate extra bases taken by runners and outfield arms where having the runner in motion also has a significant effect.

"Has nobody looked at the running game that way, and how can we pretend to understand the “breakeven point” for the running game if we haven’t?”

I’ve been suggesting this for a few years, and watching for someone to do it, so I’m fairly confident it hasn’t been done.  Or rather, hasn’t been published—I hope Peter is able to share his work. 

Separating out the catchers’ impact on DPs or runners advancing will clearly be challenging.  Maybe this is a place where Tango’s WOWY method could shed some light?  But until a catcher has a long career, we may not be able to separate his contribution for that of his pitcher and fielders very well.  But we could probably get a good approximation just by studying the average impact of runners going on the pitch, in terms of DPs prevented and additional bases/runs gained.  Since we can measure how much more or less frequently runners take off on any given catcher (controlling for handedness of pitcher), that should allow us to draw pretty good conclusions about each catcher’s impact on baserunners.  (This is all in addition to SB and CS, of course.)

Note to self:  When Peter or Guy post something, pay attention.

You can also click the “see all posts” next to Guy or Peter’s name, and see everything they wrote in this blog on one page.  It makes searching pretty easy.

http://www.insidethebook.com/ee/index.php/site/comments/a_simple_way_for_teams_to_come_up_with_proper_stealing_strategies/

This is where we had this discussion last year.

Thanks, Peter. 

I think most people share MGL’s assumption in that post, that what really matters in evaluating baserunning is SBs and CSs, while running on BIP is only a minor consideration.  But I doubt that’s true.  Peter, can you easily tell us about how many times a runner goes on the pitch on a BIP in a season (compared to total SBA)? 

Just doing some rough math, let’s say that the batter swings on 40% of SBA.  For every swinging strike, an average hitter will have about 2 BIP.  That would mean that for every 10 SBAs we observe, there might be another 8 runner-going plays we are ignoring.  So we could be missing almost half of the relevant plays!  Now, the impact of this BIP running is smaller:  for one thing, at least 40% of these BIP become flyball outs or HRs, and so the runner going has no impact.  And on 2-out GB outs it won’t matter.  Still, it’s not hard to imagine the average benefit of these plays is enough to substantially affect how high the success rate needs to be on the SBAs to make sending the runner a good percentage move overall.

Peter, where do you find data on runners in motion?  Is this in the retrosheet pbp?

Rally/29, there is an X code in the pitch event data on Retrosheet for each pitch on which a runner was in motion.  Thanks to Peter for earlier bringing this to my attention.

For 2005-2007 data. Not on 3-2 count with 2 out and not including bunts, there were 5581 runner in motion events. 1359 of these were FB or LD outs.  Leaving 4222 plays where there were potential gains by being in motion.  Or about 1400 a year.  I calculated a revised break even point.  I don’t have the exact number in front of me, but if I recall correctly it was somewhere down in the mid 60s% range.  That’s the generic value, it varies a lot by baseout state, game state, lineup position, etc.

That’s a lot of plays, though less than I would have expected.  How confident should we be that the Retrosheet scorers detect the runner going in all cases?  Seems like a hard thing to pick up while also watching the hitter and the batted ball trajectory.  (But I’m fairly ignorant about the details of recording the retrosheet pbp data.)

I hope you find the time to publish this at some point Peter—it’s great data.

Guy - The data is recorded by the Gameday stringers at the stadium.  They have a great view of the field.  At least at SF where I watched one work they do.  I think they would be aware of when the runner was in motion as the pitch was being made.  They do have a lot to record with little time in which to do it.  Whether they are 100% accurate is impossible for me to know, but the one stringer I observed was very focused and seem well trained in her job.  She was doing nothing else but observing the game and recording the information.

I will have to take another look at what I have already written on the article and see how much work it would take to complete it.  Thanks for the encouragement.

Peter, how many of those LD and FB runners in motion result in a DP because the runner was in motion?

I have been talking about this for years (that the value of a runner in motion is a combination of the SB/CS and extra advancement, fewer DP on GB, fewer forces at second, extra DP’s on LD and FB, PLUS the negative value of the hitter swinging at bad pitches on the hit and run).

As far as valuing the catcher is concerned, two things are in order.  One, using updated run values for a stolen base attempt that incorporates all of the above considerations.  Two, assuming that for all catchers, a certain minimum number of hit and runs will occur, regardless of the catcher’s throwing ability, although it still could be related to the catcher’s throwing ability to some extent.  We also have to assume that the penalty for swinging at bad pitches on hit and runs only applies to those hit and run attempts.

So, for example, let’s say that we have a catcher who is so good that no one runs on him but there are 10 hit and runs against him per year.  Well, the value of those “stolen base attempts” (they are really hit and runs) include the penalty for swinging at bad pitches. 

Now, let’s say that we have a catcher who is good, but not quite as good.  There are still 10 hit and runs against him (maybe 12, since opposing teams might be a little more likely to execute and hit and run against a catcher with a worse arm - I am not sure), but then there are another 20 legitimate SB attempts.  So, for 10 of the 30 runners on the move, we apply the penalty for swinging at bad pitches to 10 of them, and not to the other 20. And of course, all the other considerations apply to the 30 total runners in motion.

Anyway, that is how we handle everything from the perspective of valuing catcher’s arms.  We either look at the PBP data against that catcher and try and determine how many hit and runs were executed against him (that is difficult of course), or, as I said, we assume a certain number of hit and runs against all catchers, with that number being either the same for all catchers or a little more the worse the catcher’s arm is.

And of course part of the equation is trying to determine one, the penalty on a hit and run for swinging at bad pitches, and two, the gain from “opening up a hole.”

And BTW, there is no such thing anymore as a distinction between the old hit and run and run and hit.  I have mentioned this before.  Hit and runs are executed exactly the same way as a straight steal.  The only difference is that the runner should not take such an aggressive lead that he gets picked off and he is supposed to look back at the hitter (although not everyone does and lots of runners look back on a straight steal as well).  You are never taught to leave “late.” What is the point of that?  There is none.

As well, on a hit and run, the batter MUST swing at the pitch unless it is in the dirt.  Most batters cannot tell whether a low pitch is going to be in the dirt, so they swing anyway.  The batters never has an option.  On a straight steal, on the other hand, sometimes the batter will take a pitch in order to let the runner steal, although obviously if the runner is not a great base stealer (high SB percentage), that is a dumb strategy.  And certainly, you should only do that if the pitch is not a really hittable pitch.  There is of course a BE point where it is correct for the batter to take the pitch 100% of the time, no matter where it is pitched.  There is also a BE point for the batter to take a borderline pitch that he might otherwise swing at.

MGL - Parts of your post are confusing to me.  You say that “there is no such thing anymore as a distinctiontion between the old hit and run and run and hit”.  And then you go one to mention differences in the way a runner approaches a hit and run and a straight steal.  Perhaps it is just differences in the way my coaches defined these plays and the way your coaches did.  For the college and high level teams that I played on there were always three separate signs, but the distinction was in what the batter did; the runner did the exact almost the same thing on each play, he tried to steal the base. He only looked back if he heard the crack of the bat in order to know whether to hold up on a fly ball, return immediately on a catchable line drive, or whether he was going to have to slide or not at second on a ground ball.  On a straight steal the batter was expected not to swing at the pitch, on a run and hit he had the option to if he would have swung in normal circumstances, on a hit and run he was required to swing.

It is my belief from what I have read in interviews with players and managers, and with supporting evidence from the statistical record, that modern MLB practice has put more emphasis on the run and hit and green lighting their better runners.  On both of these plays the batter makes his decision to swing or not as he normally would.  On a run and hit the runner will be trying to steal, with a green lighted runner the batter knows that he may be stealing at any time.  I think that most teams have gradually come to realize that requiring the batter to swing on a hit and run or a straight steal is not productive except in very limited circumstances and have almost entirely abandoned this practice.

MGL - Is your understanding different than this?

Peter/33:
Do you know (approximately) the number of SBAs on which the batter swung in your sample?  It seems to me that the number of other RIM (runner in motion) events should be about twice as large. 

In other words, the frequency of RIM when a batter puts the ball in play should be at least as high as when the batter swings and misses (if there’s any difference at all, the batter should be making more effort to make contact with a runner going).  If the retrosheet data shows a lower RIM rate when balls are put in play it would indicate the stringers miss some of the RIMs.  In that case, would be interesting to look at the RIM rate in 6 categories to try to isolate the problem:  GB/out, GB/hit, LD/out, LD/hit, FB/out, FB/hit.  I would expect the RIM rate to be roughly the same in all cases, so if it isn’t we would know where the under-counting mainly occurs.

"there is an X code in the pitch event data on Retrosheet for each pitch on which a runner was in motion”

Do you mean the pitch sequence field?  My db has an X when the pitch results in a ball put into play.  I do see a > in some of the strings, is that runner in motion?  My file was created with bevent.  Maybe the codes are different with cwevent.  Ran that last night (took awhile to figure out the command line prompts needed), though it was too late to do anything with it.

Yes, Rally, I’m sorry. I meant >, not X.  I was thinking about runner in motion on a ball in play, which has both a > and an X, and got confused.

Guy - There were 2635 CS on pitches to the plate in 2005-2007.  199 of those were on pitchouts which would make the chance for a hit ball in play very small.  Retrosheet has no hit balls in play on pitchouts during this time.  Of the 2436 non pitchout CS, 714 were on swinging strikes.  The problem for the type of analysis that you are proposing is that it is impossible to separate the successful stolen bases on swinging strikes into those that occur where a run and hit was called, or where the runner was running on a green light, or where a straight steal was called.  There may be some individuals, or possibly even some teams, that still believe the batter should swing and miss on a strike to protect the runner. 

I really doubt that the stringer is underestimating by a significant amount if at all. 

MGL - There were 201 fly out or line out double plays when runners were in motion.  My best estimate is that 48 of those would have occurred even if the runner had not been in motion.  So having the runner in motion resulted in 153 additional double plays which goes into the debit column.

Thanks Mike.  Great stuff Peter.  Gives me a lot more things to play around with in retrosheet.

Rally - Another useful thing to know about the pitch sequence coding is that * means that the following pitch was in the dirt.

Thanks Peter.  How complete is that?  If I wanted to see how many pitches in the dirt each catcher dealt with, do you think I’d have 90% or more?

Do you think it’s reasonably consistent between stadiums?  Or maybe some scorers mark that near 100% but others only 50%?

That would make a much better denominator for evaluating catchers on WP/PB if it can be trusted.

Rally/42, if the data source for pitches in the dirt is MLBAM, the data quality is relatively poor, in my experience.  I believe the data source for Retrosheet is MLBAM; Peter can correct me if I’m wrong.

A much better source for WP/PB evaluations is the PITCHf/x trajectory data (i.e., solve the trajectory for where in x,y it passes through z=0.)

The source for Retrosheet data is MLBAM, but not the identical data that is in the Gameday XML files.  The runners in motion is only in the XML files for foul balls.  I have not compared Retrosheet balls in dirt to Gameday’s so I don’t know whether any changes were made.  I agree with Mike that z=0 location is a better parameter for judging passed balls and WPs.  I would think that from some pressbox locations it might be difficult for the stringer to directly observe whether the ball hit the ground before it hit the catcher’s mitt or not because of being blocked by the umpire and catcher and possibly the batter.

Peter, right, I should not have said, “there is no distinction,” and then listed some differences. We are in partial agreement.  I cannot say for sure what major league teams do, but my understanding from watching many games is that on a hit and run the batters must swing.  You still see that all the time.  That is the ONLY difference between the straight steal and hit and run, other than the runner looking to see where the ball was hit.

There is no sign that I am aware of from the third base coach that tells the batter NOT to swing because the runner is stealing.  The runner can go on his own, in which case the batter does what he normally does.  Also, the runner can signal the batter that he is going to steal on the next pitch, although I think this is rare.  The coach also has a “don’t go” sign and a “steal if you can” sign (I think). I don’t think that in the major leagues, there is a “steal on this pitch sign” like you have in amateur ball.

I do know for sure that there is still the “you must swing” hit and runs.  In fact, I think they all are.  That is why it is a hit and run.  If the batter has the option, it is not a hit and run and it would be crazy to send a runner who is not fast if the batter has the option. You see batters throwing the bat at the ball all the time on those plays.

Good questions to ask a major league player or coach/manager.

Peter:  Thanks for the data (39).  That is generally consistent with my expectations, except the swinging strike percentage on CS (29%) is a bit lower than I would have guessed.  So in broad strokes, for every 4 SBAs there are about 2 other RIM plays, 1 of which provides a potential benefit to the offensive team (GDP avoided, bases advanced) and 1 which does not (including a cost of 153 extra DPs on air balls).  I’ll be interested to see your estimates of the value of these additional RIM.

Pitches in the dirt, by year from retrosheet”

2005 10323
2006 10880
2007 10919
2008 11297
2009 12367

Looking it by ballpark for 2009, the range is from 510 in Washington to 295 in Seattle.  There might be some scorer difference there but the spread seems reasonable.  (I’d consider it useless if one park had 15 and another had 750)

Yovanni Gallardo bounced the most last year with 114. John Lackey had 129 in 2005, the most of any pitcher the last 5 years.

Gotta say, I was very surprised to see Russell Martin did so poorly at blocking pitches. That’s generally been his strongest defensive ability.

Rally, as best I can tell, Gameday captures about 40% of the balls in the dirt and notes them as such.  For example, of the 4664 pitches that hit the dirt in front of home plate in 2009, Gameday assigns 1779 of them as Ball in Dirt and 162 as Swinging Strike (Blocked).

I haven’t investigated by park to see if there is any effect, nor have I looked to see whether there is some other systematic reason that some pitches get noted as being in the dirt and otherwise identical pitches do not.

My rough estimate is that there are around 30,000 pitches in the dirt each year in MLB, i.e., around 4% of all pitches, or a dozen per game.

Here’s the fraction of pitches that Gameday records as being in the dirt, based on where they hit the ground, as measured in front of or behind the point of home plate.

The plate extends from y=0 to y=1.4 ft, and previously I’ve estimated from images that a typical distance for the catcher’s glove when he is set up to receive a pitch is around y = -4 ft.

In #50, “where they hit the ground” means where the pitches would have hit the ground had they not been caught.

Mike - Most of the time Retrosheet (Gameday) doesn’t bother to record a pitch in the dirt unless there is a man on base.  With nobody on whether the catcher blocks a pitch or not makes no difference, and there is no incentive for him to expend energy and risk injury by doing so.  I have Retrosheet recording 10877 balls in the dirt with men on base in 2009. What is your corresponding number for men on base where z=0 and y>-4?

Mike, is there anything on the net that gives us individual catchers’ blocked percentages updated through 2009?

Peter/52, good point.  Here is the same graph for pitches with runners on base.  It looks like Gameday stringers get about 85% of the pitches in the dirt in that situation.

MGL/53, I’m not sure if this is what you are wanting, but Dave Allen posted a list of leaders and trailers here:
http://baseballanalysts.com/archives/2009/12/pitchfxing_past.php

He used a more simplistic analysis in some ways (taking the pitch height at y=1.4ft rather than finding the z=0 point) and more complex in others (accounting for pitch type and horizontal location).

85% is pretty good.  Enough for us to identify the good and bad catchers at blocking pitches.  At some point I’ll update the catcher ratings on my site, at least 2005+, to use balls in dirt as the denominator for WP/PB instead of innings.  But I don’t expect the ratings to change much. 

The way I measure catcher defense takes some short cuts (splits by LHP/RHP instead of WOWY) and people don’t think they are any good.  I still see “We have no idea how to measure catcher defense” posted in a lot of places.

Yet compare my numbers to Devil Fingers’ WOWY approach for the leaders/trailers and what do you get?

Laird 13/14
Johjima 9/9
Ross 7/7
Hanigan 7/6
Molina 7/10

Montero -7/-8
Hundley -7/-8
Bard -7/-7
Olivo -9/-6
Napoli -10/-8

I could tell you which column is mine and which is WOWY, but why bother?

MGL, if you don’t do it yourself I could send you my counts of balls in dirt/PB/WP by catcher.

We don’t have any idea how to measure game calling ability or pitcher coaching effects for catchers, and we’re only at the beginning of looking at pitch framing skills.  We also don’t take into account the effect of hit-and-run/runners-in-motion on catcher baserunning defense.

So there is a lot of room for improvement.  But no, “we have no idea how to measure catcher defense” is pretty far from the truth, too.

Agreed.  I’m only measuring a portion of catcher defense, and remain agnostic on the whole issue of gamecalling.  To this point we still aren’t sure if there’s a skill here.

As long as we’re upfront about them being limited to blocking pitches, the running game, and errors, I think the catcher ratings are more reliable than the various fielder range ratings.

"MGL, if you don’t do it yourself I could send you my counts of balls in dirt/PB/WP by catcher.”

Please, thanks.

“As long as we’re upfront about them being limited to blocking pitches, the running game, and errors, I think the catcher ratings are more reliable than the various fielder range ratings.”

Right, exactly. We are probably a little shaky on measuring catcher fielding skills on bunts and squibs.

Mike - Just for fun I went back and looked at the frame by frame Pitch f/x videos that I have from my Hit f/x research and looked at a couple of low pitches from the A camera or side view.  Fortunately the catcher’s mitt just makes it into the frame.  If the batter is standing on the back line of the batter’s box the catcher sets his target about a foot behind the back line.  This would be about Y = -3.3.  But if it is a low pitch the catcher will extend his arms to catch it nearly at the back line (Y = -2.4 or so).  This corresponds nicely where your graph starts to fall off to the lower percentages.  The other thing that I noted is that the catcher will move up on batters that also move forward in the batter’s box so that his target sometimes is right on the back line.  In this position he could catch balls at Y= -1.5. 

There was only one ball in the dirt in this game recorded by Retrosheet.  It crossed the front of home plate at Z= .268.  The catcher also successfully block strikes in the dirt that crossed the front of the plate at Z= .363, .324 and .491 and they were recorded correctly.  There was a wild pitch on a ball that short hopped the catcher hitting the back line of the batter’s box after it crossed the front of the plate at Z= .432. The catcher caught a ball not in the dirt that crossed the front of home plate at Z= .828 and several in the .9 range.  The stringer got every call correct in this game.

MGL/53

I’ve updated my Catcher Block Percentage, which is based on the Gameday stringers versus the pitch F/X z=0 point over at Beyond the Box Score.

2010 Marcel Projections:
http://www.beyondtheboxscore.com/2009/12/28/1220699/2010-catcher-block-percentage

2009:
http://www.beyondtheboxscore.com/2009/12/21/1208572/2009-catcher-block-percentage

2008 and before:
http://www.beyondtheboxscore.com/2008/12/1/675791/2008-catcher-block-percent

I’ve looked into using the z=0 point with Harry Pavlidis, but we weren’t able to come up with a good value for what were actually in the dirt.  I’ll see if I can find the results of the stringer versus z=0 comparison.

My gut feel is that we need to use a curved line to determine whether the pitch hit the ground rather than a straight one at a given x value.

DAn/61 was marked for mod and is now open.

Dan & Mike - Mike, I don’t know how this affects your table or how it effects your research Dan, but balls that are in the dirt and not blocked by the catcher are not recorded by Gameday/Retrosheet as being in the dirt.  So wild pitches can be balls in the dirt that are not blocked and go to the backstop and are just marked “B”, all balls not in the dirt whether or not they are blocked by the catcher also just marked “B”, or balls in the dirt blocked by the catcher but still allowing the runner to advance that are marked “D” or “*B” in Retrosheet.  Or the similar categories of swung and missed strikes.

My gut feel is that we need to use a curved line to determine whether the pitch hit the ground rather than a straight one at a given x value.

Dan, I don’t understand what you’re saying here.  Can you clarify?

Peter/63:  I’ll go back and check to see whether my data set includes them as opportunities or not.  Worst case scenario I need add them back in which shouldn’t affect things tremendously I don’t think.

Mike/64:  My guess is that the actual shape of the “in-dirt” line as seen from above is a curve of some sort instead of a straight line at Y=-3.5 (for example).

I think you’ll see the Y value of balls in the dirt be closer to 0 as you move further in or off of the plate.

Peter/63, the count of wild pitches + passed balls is of the right magnitude to account for about 15% of pitches in the dirt, so perhaps the explains the discrepancy in my graph.  I have not made a detailed accounting to check the agreement, but it seems that now that I understand in which situations the stringer will potentially mark a pitch as in the dirt, the data may be of much better quality than I had supposed.

Dan great work, other than this:

”After the jump we’ll get into the full set of results, but I’ll share some observations first.”

You did that just to annoy me, right?  I don’t see a “jump” by the way. Is this one of those for the RSS feed?

Dan, a couple of questions, after the jump that is…

“Jump”

1) Are you taking into consideration where the ball bounces or are you considering all pitches in the dirt as the same?  If the answer is yes, you are considering where it lands, why would someone catching Wakefield be any different from anyone else?  In fact, I would think that a 70 mph knuckleball in the dirt would be a lot easier to block than a 90 mph fastball or even an 83 mph curveball or 87 mph slider?

2) Are you using any regression for your Marcel?  If yes, what are you using for the regression (basically how many opps or pitches of league average blocking ability)?  If you are, how did you determine how much to regress?  You would have to do some kind of regression analysis (y-t-y or correlation or ICC) to determine how much to regress given each catcher’s number of opps.

Do you have the yearly data available for each player do that we can do our own projection?

Thanks!

MGL: 

I actually wrote that before your piece on “The Jump”. The reason it’s there is that SBN’s editor actually includes a jump in it.  But yeah, it’s a little cliche.

1) I’m counting all balls in the dirt the same - simply using the stringer’s indication of the ball being in the dirt.  I need to go back and double check the data now based on Peter’s information about the treatment of wild pitches.

2) I’m adding roughly 342 league average opportunities for regression purposes.  That’s the seasonal average of opportunities for the league from 2007-2009.  I’m doing 5*2009 + 4*2008 + 2*342 for the opportunities.  Since the entire measure is above/below average, the league average opportunities are 0.

I can post the spreadsheet somewhere if you’re interested.

"I can post the spreadsheet somewhere if you’re interested.”

That would be great.

You’d get rid of a lot of noise if you took into consideration where the ball bounced as in Mike’s graph in #54 above. Otherwise you’ll get pretty much the same results as anyone using retrosheet.

Plus, you have to figure out how much regression (how many league average opps to add) by running an ICC or a year to year (or any time period) correlation.  You can’t just guess at it.  It could be 300 and it could be 1000.

A couple of comments related to this:

TT@10: I wound up evaluating SB and CS separately because of this. Turns out Bill James, in his Catching Win Shares article, took the same tack. Steals are as a percentage of runners on first, caught stealing are as a percentage of steal attempts.

Hit-and-run, general: count me in the camp who attributes the decline of the true hit-and-run to be the reason why OCS% went from 45% in 1915 to about 25% now. In fact, we could check the more recent data for evidence of this. How well do players bat after a caught stealing versus after a stolen base? After a failed hit-and-run, the batter has swung and missed, and the runner is probably caught stealing. After a pure stolen base attempt or a run-and-hit, the runner is probably safe, and the batter took, not necessitating another strike (in fact, the pitcher probably threw a ball slightly outside if he anticipated the runner going). Thus, if there are plenty of hit-and-run plays, batting stats after a caught stealing will be much worse than after a stolen base.

"Thus, if there are plenty of hit-and-run plays, batting stats after a caught stealing will be much worse than after a stolen base.”

Interesting!

MGL:

My office is being painted right now, so my computer with the spreadsheet is inaccessible.  I’ll post it once I’m able to set the machine back up.