Yeah, it’s gotta be Grassko’s Range Adjustment on Batted Balls: GRABB.

Gassko ... my apologies for misspelling.

as long as his metric shows that Jeter is a horseshit fielder, it’s OK by me

The system has merit and I like the side-by-side comparisons with other metrics out there, but do we have to thump our chest every other line?  Cut out the self-aggrandizing and this is a great article.

More valuable in the field, Ron Oester or Kevin Bass.

More valuable in the field, Ron Oester or Kevin Bass.

false

but do we have to thump our chest every other line? Cut out the self-aggrandizing

what he said; the author is going to tear his rotator cuff patting himself on the back

Shouldn’t the results of a system be given in proportion to the precision of the results? IOW, this system is using a reasonable, but certainly imprecise estimate of chances. So why have 1 guy at +11 and another at +12? Why not simply use 5 labels, such as bad, below avg, avg, above avg, and excellent?

what he said; the author is going to tear his rotator cuff patting himself on the back

The Jets are interested in signing him to back up Vinny.

The outs to run conversion is wrong.  It should be at around 0.8.  For those who have never seen the explanation, here it is:

============
Suppose a team with Ozzie at SS gives up on average 12 non-HR hits, and 2.6 walks every game (which of course is 27 outs). Applying .50 runs per non-HR hit (I know it should be closer to .55, but I just want to keep it basic), and .30 runs per BB, and -.10 runs per out, and we get 4.08 runs scored per game. And per game, we see that Ozzie’s team faces 41.6 batters (again, let’s not worry about DPs, etc).

Now, let’s say Ozzie was traded for Spike, and let’s say for every 41.6 batters faced, there is one ball that Ozzie gets to that Spike doesn’t. So, for those 41.6 batters, Spike’s team records 13 non-HR hits (1 more than Oz), 2.6 walks, and 26 outs (1 less than Oz). However, there’s still one more out to go! Since Spike’s team gives up 13 non-HR hits / 26 outs, we can estimate that this team will give up 13.5 non-HR hits, 2.7 walks, and 27 outs per game ( a total of 43.2 batters, a remarkable 1.6 MORE batters than Oz). Anyway, applying our LW constants, and we see that Spike’s team gives up 4.86 runs per game.

This number is .78 runs MORE than Ozzie. This is the result of Ozzie getting to one more hit than Spike. .50 runs for the hit, and about .30 runs for the out gives you the .80 runs.
====================

***

I didn’t get a sense that park factors were being handled.  For example, fielders at Coors have a tough time converting BIP into outs.

***

I like the basic premise to the concept.  You are trying to establish some sort of context for all the players.  There are other things in play of course, and I think those can also, possibly, be handled.  For example, a GB pitcher will get more outs per GB than a FB pitcher.  So, you can use the team level GB/FB ratio to tweak that. 

***

I’m not sure how the LH/RH split is being done, though, I guess I’ll have to read Charlie’s article.  But, basically, a LH hitter won’t necessarily hit proportionately the same number of groundballs and flyballs to the same side of the field. 

***

Anyway, David’s idea is to try to establish as much context as possible using as much of the data at hand.  In that respect, it will be a worthwhile effort, and would be a step ahead of Charlie and Clay, and a step below MGL.

So the Natspos made the right move in dumping that mediocre, Cabrera for that awesome stud, Guzman.

Using the method Dial came up with to approximate Defensive Runs from ZR, the average out made by a SS would be .75 runs—outfielders would be more (a CF is around .84), but it’s always around .8, consistent with Tango’s theoretical example.

So if Jeter was -27 assists, he should really be at -20.25 runs, not -16.  (FWIW, PMR pegged him at about -4.57, and IIRC UZR had him right around average.)

I agree that there is merit in this approach, with some tweaks and further study.

A question:  are the UZR statistics you cite the actual UZR numbers, or the UZR prorated per 150 games?  Jay Payton played 1027 innings in CF last year, equivalent to 114.1 games—he was really +34 runs in that few games? 

Because the PMR Runs you cite are the actual numbers, not per-season, and that seems true of your own RAA numbers as well.

(And not to be a jackass, but where do you get the PMR-to-run conversions?  I happen to know because, um, they come from me.  But someone coming into the conversation cold might wonder—and it should probably be acknowledged that Pinto didn’t come out and say “X player is Y runs above average,” but that the runs reflected by PMR are an estimate based on other research in the field.)

I don’t think park factors will say anything meanigful about an infielder’s range. We need to know how thick the infield grass is ( Cubbies IF is bound to show more range than the Coors infield ), and that is prone to change year to year.
do we have access to data as to how many seeing eye singles are hit?
if not, Maybe a way to approximate infield grass depth is :
p : fraction of hits that are singles
seeing eye singles = ss ~ (#singles - p*#ld)
park average : ss / (ss + #groundouts)
normalise the park averages to get park factors.

as long as his metric shows that Jeter Clayton is a horseshit fielder, it’s OK by me

MGL published the infield and outfield factors in his last UZR column on this site.

This number is .78 runs MORE than Ozzie. This is the result of Ozzie getting to one more hit than Spike. .50 runs for the hit, and about .30 runs for the out gives you the .80 runs.

Tango, isn’t this like punishing the SS because of how the rest of his team gets outs...or doesn’t? 

What if a player gets to 1 ball less per game but on a team that only allows 9 non-HR hits a game and 2 walks a game?  Is he a better fielder than Spike simply because his team doesn’t allow as many hits?

The way I prefer is to do this:
RE = run expectancy

Add the RE of a single with 0 outs, 1 out, and 2 outs. 

Subtract from that what the RE would have been had the player made the out (add the RE with no men on and 1 out, no men on and 2 outs, and no men on with 3 outs (zero, obviously, but that’s just to make it easier to follow)).

Take that number and divide it by 3 to get the average run value.

And there’s your multiplier.

I’m curious as to what you think, Tango?  Is there something wrong with that that I’m missing?

Cut out the self-aggrandizing and this is a great article.

Didn’t mean for it to sound that way. I’ll just take the compliment as meaning my self-aggrandizing was justified. ;)

Shouldn’t the results of a system be given in proportion to the precision of the results? IOW, this system is using a reasonable, but certainly imprecise estimate of chances. So why have 1 guy at +11 and another at +12? Why not simply use 5 labels, such as bad, below avg, avg, above avg, and excellent?

David, each metric (even UZR!) has pretty big error bands. With a full season of UZR, maybe a difference of a 3/4 SD is telling you that there is some difference between two players. With my metric, maybe it’s 1 SD. But are just more easily quantifiable, which is why I use them.

The outs to run conversion is wrong. It should be at around 0.8. For those who have never seen the explanation, here it is:

Tangotiger, maybe you could correct me here, but I thought that for assists and (to a lesser extent) putouts, you don’t want to use straight-up out values because one assist above average is not necessarily one out created above average. I used .59 and .69 as my values for assists and putouts. If I’m wrong, I’l fix that for the next installment.

I didn’t get a sense that park factors were being handled. For example, fielders at Coors have a tough time converting BIP into outs.

They’re not. I was going to use MGL’s park factors but I’m worried that as we’ve seen a few new parks come in since then, the park factors would be different. My specific worry is San Diego and the NL.

I like the basic premise to the concept. You are trying to establish some sort of context for all the players. There are other things in play of course, and I think those can also, possibly, be handled. For example, a GB pitcher will get more outs per GB than a FB pitcher. So, you can use the team level GB/FB ratio to tweak that.

That is essentially what I’m doing. I’m assuming that ground balls should be converted at a league average rate (which I think is .272 BA) and fly balls should be converted at a league average rate (~.275 BA), and same thing for liners, etc…

I’m not sure how the LH/RH split is being done, though, I guess I’ll have to read Charlie’s article. But, basically, a LH hitter won’t necessarily hit proportionately the same number of groundballs and flyballs to the same side of the field.

I thought of this but did not adjust for it. I’m not sure how true this is, but more importantly, I could not look into this w/out PBP data. Maybe MGL can chime in on this.

(And not to be a jackass, but where do you get the PMR-to-run conversions? I happen to know because, um, they come from me. But someone coming into the conversation cold might wonder—and it should probably be acknowledged that Pinto didn’t come out and say “X player is Y runs above average,” but that the runs reflected by PMR are an estimate based on other research in the field.)

Your site is listed in the resources and references. Would you like to be credited directly in the article? I can do that in my next installment if you’d like.

A question: are the UZR statistics you cite the actual UZR numbers, or the UZR prorated per 150 games? Jay Payton played 1027 innings in CF last year, equivalent to 114.1 games—he was really +34 runs in that few games?

Yeah, that was a mistake I noticed after the article was published...it shouldn’t really matter for most players though.

Two more things: I don’t want to use infield factors (I don’t think they’re significant) and I really need more name ideas (though I’m not anti-GRABB).

My past experience:

(Author refers to method as “my method") = BAD method

(Author refers to method as “this method") = GOOD method

This is based on a small sample size, yet I’ve found it to be a reliable rule-of-thumb.

You could always call it something unassuming yet catchy and reminiscent of a statistic from another area. Using your initial would work well. Perhaps Gassko Putouts and Assists, or GPA? No? Not good for you?

“I’m assuming that ground balls should be converted at a league average rate (which I think is .272 BA) and fly balls should be converted at a league average rate...etc...”

Could you refine this to calculate a conversion average for each position?  For example, it seems possible the conversion rate for 3B is not same as for 2B.  And if dataset is big enough, you could even do LHB/RHB versions (GB-3B/LHB=.279, etc.).

Devil,

That’s what I did!  My illustration is a long-complicated way to get to what RE does.  The run value of a hit will be around +.50 runs.  That is, the change in run expectancy of adding a hit, will on average, add +.50 runs.  The run value of an out will subtract .30 runs.  That’s a .80 run swing.

but I thought that for assists and (to a lesser extent) putouts, you don’t want to use straight-up out values because one assist above average is not necessarily one out created above average. I used .59 and .69 as my values for assists and putouts. If I’m wrong, I’l fix that for the next installment.

That’s a good point.  You’re saying here that you can get an assist from, say, a relay throw?  For an OF though, wouldn’t the putout be the same as a single out?

Your site is listed in the resources and references. Would you like to be credited directly in the article? I can do that in my next installment if you’d like.

Oh, no, there’s no need for that.  But the References & Resources only contains one link; it’s to this BTF thread, wherein MGL gives the UZR for several players.

Blackhawk, David did include a link to your blog in the original article, but it was taken out.  Probably my fault.  Sorry about that.

No worries—it’s not about me (I didn’t really do anything but combine the efforts of others), but I do think someone reading it might wonder where those numbers come from, as they’re not on Pinto’s site anywhere ...

(Btw, Studes, thanks for the link from your HBT a few weeks ago when Steve Finley was benched!)

Devil,

That’s what I did! My illustration is a long-complicated way to get to what RE does. The run value of a hit will be around +.50 runs. That is, the change in run expectancy of adding a hit, will on average, add +.50 runs. The run value of an out will subtract .30 runs. That’s a .80 run swing.

Okay.  I think part of my confusion (besides being dumb ;)) was that I was looking at the run value of a single specifically, not all non-HR hits.

That’s a good point. You’re saying here that you can get an assist from, say, a relay throw? For an OF though, wouldn’t the putout be the same as a single out?

Yes. Also, I think you want to regress the run value towards zero based on how sure you are that your system is coming up w/ the correct number of plays above/below average a player made. So for example, if you use something like fielding runs, maybe the run value for an assist is .3, with CAD, it’s .5, and with this system, it’s around .6. For UZR, of course, you would use .8. Does that sound right?

Could you refine this to calculate a conversion average for each position? For example, it seems possible the conversion rate for 3B is not same as for 2B. And if dataset is big enough, you could even do LHB/RHB versions (GB-3B/LHB=.279, etc.).

That’s basically what I’m doing. To find xAssists or xPutouts, I find how many assists (or putouts) a player at 3B (or w/e position) is expected to get with a left-handed batter hitting a ground ball and how many he is expected to get w/ a right-hander hitting a ground ball. Overall, the numbers will add up to outs/GB = .728, or whatever.

“I don’t want to use infield factors (I don’t think they’re significant)”

I would strongly disagree. I would imagine there’s a significant difference for infielders that play on field-turf, for example. If you are going historically, I don’t see how you can’t adjust for at least artificial turf if you are doing the 1980s.

I would also think that ground balls hit in Coors Field come at a much greater velocity, which would certainly impact 3B significantly (as well as 2B, SS and 1B to a lesser extent). Not to mention grass height, etc..

If it can be done fairly easily, I think this is pretty important.

Joe, MGL’s infield park factors ranged from .97-1.02. Coors was .97. That’s pretty insignficant if you ask me (remember you would have to regress 50% towards the mean to adjust fo the fact that a team plays half its games on the road, making the spread .99-1.01).

True David. But with the number of plays involved even 1.5% (.97 would go to roughly .985 possibly a little higher/lower depending on the number of teams in the league) is significant. You are talking a 2.5% spread from worst to best park, over 400 chances that’s 10 plays. Using .69 as the run value of a play, you could have an 7 run spread right there. I guess that not huge, but it could make a significant difference in rankings. I mean the #5 2B in the AL was only 12.4 RAA, having the park impact as much as 7 runs seems significant.

It’s kind of like how a .9 HR factor (not adjusted for playing 1/2 the games on the road) isn’t a big deal (maybe 8-10 HR a year for a typical team), but it’s huge for a 1B (50 singles) factor. The more events, the larger the impact of small park factors.

“But with the number of plays involved even 1.5% (.97 would go to roughly .985 possibly a little higher/lower depending on the number of teams in the league) is significant.”

Actually, it won’t be lower than .985, misspoke there. It will be a little higher than .985 depending on the number of teams in the league . . .

I would strongly disagree. I would imagine there’s a significant difference for infielders that play on field-turf, for example.

Maybe it makes more sense to compute infield factors based on turf type rather than on park.  Even if you get that the new types of turf are 1.02, grass is .98, and old AstroTurf is 1.04, that’s worth doing.

I agree that the individual park factors probably have as much noise as signal.  But if you group the parks by turf type, you greatly increase the sample size, while keeping most of the real differences (or, to go to a cliche, you manage to measure the baby while throwing out most of the bathwater).

MGL’s park factors were based on 10 years of data and are already regressed so I’d assume there’s much less noise than I otherwise would . . .

FWIW, MGL’s park factors are in this article.

Blackhawk - how about a link to your blog.

Another factor though I don’t know how significant this is.
When a ball is hit into deep short or the center field hole, and the fielder dives to keep the ball in the IF, while not recording an out, but potentially saving a run. Is there a way to factor this into the range calculations? To me, this is often the most graphic/visual evidence of range, but something I think this model doesn’t capture.

For Rauseo:  here’s where I explain how I attempted to convert the PMR figures into runs.  The sidebar has links to the results for all positions.

Another factor though I don’t know how significant this is.
When a ball is hit into deep short or the center field hole, and the fielder dives to keep the ball in the IF, while not recording an out, but potentially saving a run. Is there a way to factor this into the range calculations? To me, this is often the most graphic/visual evidence of range, but something I think this model doesn’t capture.

No, but I doubt it’s significant. How many times does this happen in a year? No more than a couple…

What’s the point here?

David, do you think a system using traditional fielding stats can be more accurate than zone rating, which is publicly available?

No, but I doubt it’s significant. How many times does this happen in a year? No more than a couple...

Almost every game you see an infielder make a play on a ground ball without making a throw because he decided he wouldn’t have gotten the runner in time. Not all of these save runs directly, but I think it’s a good measure of *range* as opposed to strictly measuring runs saved.

David, do you think a system using traditional fielding stats can be more accurate than zone rating, which is publicly available?

Yes, though I’m not claiming that this metric is. I don’t know either way until I do some tests. What I am saying is that a non-PBP metric can compliment a PBP metric, especially if it is not perfect, i.e. PMR and especially ZR.

Almost every game you see an infielder make a play on a ground ball without making a throw because he decided he wouldn’t have gotten the runner in time. Not all of these save runs directly, but I think it’s a good measure of *range* as opposed to strictly measuring runs saved.

Either way, this isn’t something that as far as I know anyone keeps track of.

Almost every game you see an infielder make a play on a ground ball without making a throw because he decided he wouldn’t have gotten the runner in time.

The question is whether that’s a result of good range or poor positioning; it could be either.

-- MWE

Any system that considers Alex Cora a good fielder must be flawed. I saw him all of 2004, and he looked terrible.

The question is whether that’s a result of good range or poor positioning; it could be either.

It could be due to any combination of range and positioning, not an either/or thing.

Does anyone have a link to Charlie Saeger’s article that is referenced? The one in David’s piece doesn’t seem to work and a BTF search comes up empty.

I’m surprised this did not get a link

Gassko’s 2005 Gold Gloves

I can’t find it, but I could of sworn that there was a discussion somewhere on BTF concerning it.