THE BOOK--Playing The Percentages In Baseball

I might be able to use this.  The question is the expectation.  I’d probably need to do something similar to what John did, and figure the average run change per out and run change per non-out by position.

However, it has NOT been shown that what John does is better than the way you and MGL do it.

For example, if Chase Utley has a disproportionate number of great player with men on base than bases empty, then he will come out look really good.  Is it luck that he did so well with men on base?  Who knows.

What about an OF who makes great plays with a runner on 1B and 0 outs?  One would think that it’s irrelevant, but Dewan’s process will give that play more run value than if it was bases empty 2 outs.

In a “value” sense, this is correct.  But in an “ability” sense, it’s irrelevant.  Depending what you are after, it will lead you to different methodologies.

However, it has NOT been shown that what John does is better than the way you and MGL do it.

How does MGL figure runs?  I’ve been trying to make sense of some of the numbers on Fangraphs that seem very odd to me. For Example Ryan Church in 2007:

Year--Team-----POS-G--PO--ExO-Range_Runs_-UZR/150
2007-Nationals-LF---91-196--174----4.1-------5.5

How can a +22 in PO over expected outs only be worth 4.1 runs?  That’s about what you would get if you were figuring the value of a saved out only, instead of the difference between an out and a hit.

Are those numbers coming from the same source?  Maybe MGL is excluding shallow flies that could also count as popups - I think STATS ZR does this, and is why their plays made will not match up to official putout totals.

Right.  There is a big line on the Fangraphs page, which I’d essentially read it as: “official MLB stats”, and “stats that MGL uses for fielding”.

So, if the Nationals LF have 174 expected outs, and if they are +4 runs, then you have to presume that MGL is counting as if they had 179 outs.

The other 17 outs MGL simply discards is irrelevant for UZR purposes.

If you wanted a presentation a bit more complicated, you could have the “official” MLB PO,A,E,Inning numbers, and then the “MGL uses” PO,A,E numbers.  Whether this is better or worse depends on the kind of reader you have.

I thought it was interesting that Dewan values each extra OF play at about .2 runs less than a play made in the IF.  I would have guessed it was slightly more, based on higher # of doubles and triples.  Is this because runners can advance even on an out to the OF?  Does that really make OF plays 25% less valuable?

Also, the Pujols rating really looks like an outlier (not to say it’s necessarily wrong).  Any thoughts on why Dewan rates him at about twice as many runs saved as UZR?

Don’t forget that Dewan uses “enhanced” plays, which if memory serves, is the number of plays plus the extra bases.  So, he should be counting it as .75 for the extra plays, and .25 for the extra bases, which averages out to .58 or something.  Something like that I think.

What about an OF who makes great plays with a runner on 1B and 0 outs?  One would think that it’s irrelevant, but Dewan’s process will give that play more run value than if it was bases empty 2 outs.

Tango, if I’m reading you correctly, that’s not what John did.  He used a standard run value per position per enhanced plus/minus value.  He specifically says that he took out the impact of “clutch fielding” if you will.

And you’re right that Dewan applies his run values per base instead of play made, which makes the interpretation of his values different than systems that apply them per play made.

Agreed with Studes, on both counts. 
Check out the last couple paragraphs of the excerpt.  The 24 states analysis was used to come up with the average run value for each play made (and play missed).  The run factor is multiplied by the plus/minus (or enhanced plus/minus) total to get Runs Saved.

Thanks guys.  I guess I read it too fast, and presupposed what he would be using the RE matrix to begin with and read it along those lines.

I suppose then that that whole section was simply to justify the conversion process of plays to runs.

Studes, can you give us the highlights, specifically compared to the adjustment MGL does with UZR:
1. bat/pit hand
2. base/out
3. park
4. GB/FB tendency of pitcher
5. location (slice/distance)
6. speed
7. trajectory (FB, GB, LD, fliner, pop)

(I think those are it.)

In the first version, Dewan only considered #5 and 7.

For catchers they are looking at ER prevented for pitchers in addition to SB/CS.  I thought that as best we can tell, catcher ERA differences are mostly noise.

A big problem, if they are just comparing catchers to other catchers with the same pitchers, is that even if this is real and you can say one catcher is better than another, you can’t make the jump and apply these across the board.

For example, the Angels used 2 catchers for about the same amount of time last year.  Mike Napoli comes in at -3, and sure enough, Jeff Mathis is +3.  But how can we jump to a rating and assume we know how much better or worse each is compared to Ryan Doumit?

Amazon just emailed me and told me not to expect my copy of The Fielding Bible that I preordered 2 months ago until the end of March or beginning of April.  Are there copies out there or have they just not finished it yet?

First off, I think the book is terrific—better than the first one (and I liked the first one too).  The plus/minus system and Defensive Runs Saved are only part of what’s in the book.  Lots of other good stuff, too.

John doesn’t always go into the detail I’d like to see, but from what I can tell he looks at the following for plus/minus: Vector, trajectory, speed (hard, medium, soft), holding runners or not (for first basemen), whether hit and run is on (for 2B and SS), bases saved (for 3B,1B and outfield) instead of just outs, wall balls for outfielders.

Rally, John talks about his catcher rating extensively, and he admits it’s only a beginning. He does regress to the mean, though he doesn’t call it that.

Peter, I don’t know about Amazon, but the book is in the ACTA offices.  If you want books quickly, best thing to do is to order directly from ACTA.  That probably beats Amazon by one or two weeks.

I hope I’m not hijacking the thread here, but I wanted to circle back to a discussion about Soriano we had and his relative value at 2B and LF. According to the Fieldng Bible he went from being a historically bad 2B to an at least above average LF’er. It seems in this particular case we have a player much more valuable (and better suited) by moving down the defensive spectrum.

Ryan Braun too.  Upton too I imagine.  Lots of players.

Any particular point?

Rally had said:

Using the above defensive numbers for Soriano:

Say he’s +30 over replacement on offense.

As a LF, +1 Def, -7.5 Pos, +30 = +23.5
As a 2B, -10 Def, +2.5 Pos, +30 = +22.5

------------------------------------------

I’m not trying to play gotcha here at all. This post struck me because the values he attached to Soriano were so close even as a lousy 2B.

Then Tango said:

Rally’s got it right.

Generally speaking, if you move anyone, anywhere, you’ll always be close to equilibrium.

It’s the extreme players that are most affected.  A really good fielder, who gets moved to SS (most opps on the field), will get his value increased (in a similar way as moving Pujols from hitting 8th to hitting 4th).

And, on the opposite side, moving a really bad fielder from SS/CF to corner OF/1B (or DH) will increase his value, much the same way as moving a really bad hitter from cleanup to 8th will increase his value.

-------------------------------------------

Which makes sense, but would we expect a really, really lousy 2B to be such a good LF? That would imply that any 2B would be a great LF and would be just as valuable at LF - which on the face sounds ridiculous.

Maybe my point is that the concept fungibility of defensive skills is suspect. Some player, maybe many, have particular positions that they are best suited for.

Matt

On average, of those players who are moved from 2B/3B to LF/RF, they go from being something like -5 runs in the IF to +5 runs in the OF.  Sometimes you will get -10 IF become +5 OF.  Sometimes you will get 0 IF become +5 OF. 

But, if you find someone who is -20 or -30 as an IF become +5 as an OF, he is an exception.  That’s because we have plenty of data that shows otherwise.

Obviously, some players can hide certain skills in the OF.  OF don’t deal with many GB.  They don’t deal with accurate throws 120 feet away, with no bounce, and with a speeding runner.  So, if Ryan Braun has problems with quick and accurate throws at 3B, you put him in LF.

No one expects a really lousy IF to be a good OF.  It happens.

Matt, when a player plays more than one position or is moved from one position to another, like Sori, or Braun, they may do especially better or worse, for whatever reason or reasons. It could be noise (in either the original or the new position), or they could be particularly suited to one position or the other.  Who knows?  All we know is the average difference in position-specific run value from one position to another based on all the players who play different positions.  And even then, we are making educated guesses because of sample size, aging, and selective sampling issues/problems.  In general, as Tango says, unless we know that a player is particularly suited to a particular position, if we move him to another position, his value should remain about the same unless he is especially good or bad at one position and it behooves us to allow him to have more or fewer opportunities at another position.

Wondering why a player does a lot better or worse or one position as compared to another is like wondering why a player does a lot better or worse in one lineup slot as compared to another. Who knows?  Noise?  Maybe.  Something else?  Maybe.  All we know is that players probably do around the same in whatever lineup slot they are in, although, as with defense, players are more suited for some lineup slots than others, not because we expect them to do any better or worse (although that is possible), but because we can leverage certain skills and profiles in one slot as opposed to another one.

Tango, something I added this year in UZR is that I treat bunt ground balls separately (like another batted ball category).  I am also going to tweak it a little soon to take into consideration “wall balls”.

What do you guys mean when you say he used “extra bases” (or whatever) in computing the run values?  I simply use the average run value of a hit in a certain location with a certain type of ball (line drive, fly, etc.) and the value of a generic out in that location given that kind of ball (in fact, I think I use the same value for an out no matter what kind of ball it is and the location).

The thing that I do differently from other pbp fielding folks is that when a ball is caught by anyone, no one else gets docked any runs.  If a ball is not caught, then everyone who ever catches a ball from that “bucket” gets docked something.  Some of the other systems dock every player who does not catch any ball.  I think that is wrong.  When a ball is caught by another player, it suggests that it was actually hit closer to that player and further away from all other players.  It is not correct, for example, if a ball in location X is not caught by the LF’er but caught by the CF’er, to dock the LF the same number of runs as a ball hit in the same location but not caught by anyone.  Some systems do that.  The fact that the first ball was caught by the CF’er suggests that it was hit closer to him and further away from the LF’er than a ball (even though it is the same kind of ball in the same location - presumably, according to the data) that was not caught by anyone.

People do not realize how inaccurate the data actually is.  Every time someone compared the data from one company to another or to video, they always remark how bad they think the data is. Not that the results of these pbp systems are not any good.  They are (good).  It is just that the integrity of the data leaves a lot to be desired.  There is only so much you can do when you are hiring 10 dollar an hour people to record what they see on video with no real reference points.  That is true even when you have 2 or 3 people doing the same thing.  Do you really think that a person can nail a line drive that falls in the middle of the field to within 10 feet of distance from home plate and within a few degrees in terms of the vector?  No chance.  For example, we found SD between hit tracker data and the BIS data to be around 10 feet in distance.  That is a lot.

MGL: let’s discuss this “not docking if ball is caught”.  It depends.  If you have Adam Everett at SS, and if he catches a ball, then you don’t ding the 3B.  If he doesn’t get that ball, then both he and the 3B get dinged.

On the other hand, if you have some random Yankee veteran of 14 years, and he doesn’t get to any 5/6 balls, then both SS and 3B get dinged.

So, the 3B gets dinged less, if he plays next to a great SS.  I don’t know that that necessarily makes sense.

In illustrative numbers: if you have 100 balls hit to 5/6, and the 3B gets 30 outs, Everett gets 60 outs and the Yankee-who-will-not-be-named gets 20 outs and the average SS is 40 outs, then this is what it looks like:

With Everett: 10 hits allowed, split as 57% for SS.
With Yankee: 50 hits allowed, split as 57% for SS

It doesn’t add up, if you are trying to keep the 3B and SS independent.

I don’t ding any players when an out is made, even in years where I have project scoresheet zones.  In fact, I think I was one who helped convince MGL to do it that way.  I go with the philospohy of no harm no foul.

In the Yankee example, I don’t see it as necessarily true that A-Rod will look better if he’s playing closer to the shortstop.  He may cut off more balls going to the 5-6, but by doing so he’s opening the line.

I’m not sure if this is really an issue in UZR though, (correct me if I’m wrong) because the zones are fine enough that there shouldn’t be many shares zones in the infield.  For TotalZone it can be a problem, I think Chase Utley is not getting full credit for his shutdown of the 3-4 hole and some is spilling over to Ryan Howard.  Everett probably helped Biggio’s rating a bit.  But there’s only so much you can do when all you know is which fielder picked up a hit or made the out.

"In general, as Tango says, unless we know that a player is particularly suited to a particular position, if we move him to another position, his value should remain about the same.”

I would think that the position differences in opportunities would mean this isn’t quite right.  It seems to me a good fielder will decrease his value when moved down the spectrum (fewer opportunities), while a poor fielder will increase his value with the same move (again, fewer opportunities). 

**

I think the difference in required skill set between IF and OF is fairly substantial, such that a lot of players are significantly better at one than the other.  Think about 3B and CF.  Offensive production is very similar, suggesting the scarcity of the two skill sets is virtually identical.  But does anyone think Rolen, Glaus, or Lowell could be above-average CFs?  For the IF/OF move, I think the potential for selective sampling is huge when you look at players who make that switch.  Within the IF and within the OF, I’d have much less concern.

More Fielding Bible excerpt:
http://www.hardballtimes.com/main/article/what-makes-chase-utley-so-good/

***

Guy: I don’t disagree with you necessarily.  We’re just talking about degrees.

And Peter replies at BallHype to Dewan’s findings, which really echoes alot of what we are saying in posts 19-21:

The reason Utley plays so far to his left, and he does, is that he has to cover the part of the 1st baseman’s normal zone that Howard can’t get to.  Utley made 58 plays in what would normally be considered the 1st baseman’s area of responsibility.  A normal 1st baseman would only have allowed 35 balls to go through and be fielded by the 2B.  So that’s 23 plays that Utley gets to make that other 2B’s wouldn’t have an opportunity to.  Also, the shortstop makes 10 more plays in Utley’s area of responsibility than he would normally make if everybody hadn’t shifted toward first.  Thats a plus 33 plays that Utley makes, not because of his above average skill, but because of Howard’s less than average skill.  Utley is still a great 2B and saved more runs for his team than any other 2B in 2008.  But he doesn’t project to be the best 2B per opportunity.

So, we need to be very careful in treating players in isolation.  A WOWY approach, with Utley/Howard and other2B/Howard would be VERY instructive.

I don’t think it would be very useful, because there isn’t much data on 2B not named Utley playing alongside Ryan Howard.

More useful than Ripken vs. non-Ripken Oriole shortstops, but still a small sample size.

I did some research a while ago using the fine location data that suggested that there is not much overlap in terms of balls that more than one player can actually get to in the IF.  That is not to say that positioning, such as in the Utley situation above, is not important in terms of processing the data.  It is certainly possible that adjustments can be made in UZR (or other PBP systems) to take into consideration the skill of adjacent fielders.  The “adjustments” would be a proxy for position, I guess, with the assumption being that a fielder plays closer to an adjacent fielder who is not that skilled and further from one who is more skilled.

To be honest, I am a little skeptical of that construct and I am a little skeptical of the Utley data and presumptions above.  That is especially true for 1B and 2B.  Don’t forget that a 1B is not supposed to come too far off the bag to field a ball even when he can get to it.  He is supposed to leave those for the 2B if he thinks that his 2B can get to it.  So I don’t think that a 2B is going to change his position much depending on the range of his 1B.  And the 1B simply does not get many balls to field in the first place. 

With the SS and 2B, they are so far apart, that I really don’t think that the range on one affects the positioning of the other.  I just don’t see that.

I didn’t mean to question the relative positional values. I looks good as far as my limited statistical chops take me.

This may have been conflated in my own mind, but average is not the same as typical. For every Soriano there’s an Ellis who would certainly be far less valuable at any other position and whose arm probably only plays in LF or 1B. Of course, I don’t know that for sure since he will never be moved off a position where he’s playing Gold Glove caliber defense. Only the Yankees do stuff like that. 

By the way, ARod is playing a supposedly easier position at 3B at only an average level while he was a Gold Glove SS. Too much time has past for the excuse of positional unfamiliarity to hold any water.

Maybe my only point is that positionally flexible players are the exception rather than the rule. And maybe I am only stating that for my own benefit since I read too much into the idea of “average fielder” in the first place.

Matt

Matt, no doubt that there are skills and skill sets that are more suited to one position than another and that only some players can play multiple positions.  For us to assume that if a player moves from one position to another (even though he has not moved) that the positional adjustments will automatically apply, is presumptuous.  However, if a team does in fact move a player from one position to another (such that they think that he has the skill set to play that new position), then I think it is a good assumption that the positional adjustments will apply, on the average.  After all, that is where the adjustments came from in the first place - from players who were actually moves or who play multiple positions.

It is like MLE’s.  They only are derived from players who are deemed good enough and to have the skill sets to be promoted to the majors.  To think that they would apply to players who have not yet played in the majors and are not touted to play in the play in the majors or have been intentionally kept in the minors, is a bit presumptuous as well.  However, if a player is in fact promoted, then their MLE’s give us a pretty good idea as to what to expect in the majors (again, on the average), given the histories of all players who played at both levels.

Guy, I think that’s right about leveraging opps, but, for example, there is not much of a difference between SS and 2B opps.

Is there a difference between the “enhanced” plus/minus (i.e. on Bill James’ site) and the plus minus ratings in the excerpt you posted? Also, is one of plus/minus or UZR significantly better than the other?

I’d probably weight it as 50% MGL, 30% Pinto, 20% Dewan, but I’d really have to see the methodology.

I’d like to see some discussion of the Defensive Misplays/Good Plays system of Bill James, which was mentioned in the first Fielding Bible, but is prominently featured in the new book.

It’s ‘fashionable’ to look at James as somewhat of a ‘relic’, but I happen to like this approach. If you are going to have trained observers looking at every play on video, I think this is a better way to maximize their contribution than just looking at the UZR type stuff.

Whether the implementation is good enough is up for discussion. As well, it’s possibly likely that the system rewards skills over range disproportionately, but that pretty much depends on how good the raters are at evaluating the range-style Good Plays…

I have always thought that the subjective ratings for each play would be real valuable as an addition to the objective data (for example, let’s say that there is a ball to a certain part of the field that the SS catches 90% of the time, but on a particular play, it takes a bad hop and no SS could have caught it).

Also, there are people/teams who are looking at video and doing evaluations on plays other than batted balls, such as relay throws, throws from the outfield (like overthrowing the cutoff man), tags at bases, dropping an easy pop fly, and the like.  Boston, for example, has been doing this for a while, as far as I am aware.

I have always said that we are 90% there as far as defensive evaluation goes.  The other 10% is more granular data like hit f/x as well as other ways of looking at defense other than batted balls being caught or not.

Keep in mind that there are two ways to improve defensive (or any, for that matter) evaluation:  One, getting rid of some of the noise so that we can get more accurate numbers with smaller sample sizes, and two, getting better inputs and coming up with better methodologies such that we are more accurate (in describing true defensive talent) no matter what the sample size.

Oh, BTW, I got my Fielding Bible from Amazon a couple of days ago!

Oops, MGL, you’re not supposed to get it from Amazon, you’re supposed to get it from ACTA at a higher price.

For me, I like this concept because it coincides with the experience of watching an actual game. When I see the 3bman double pump and miss out on a double play, I want to see him docked for it. When I see the SS dive and keep a hit in the infield (not giving baserunners a chance to advance further), I want him to get credit for it.

The main problem I see is that everything is given the same weight. A Good Fielding Play where the SS gets to a ball that looked like a sure hit to the observers is given the same credit as a failed but ‘reasonable’ throw to the plate on a SF where a runner takes an extra base.