An archive of StatSpeak from its days on MVN
June 24, 2008
Filed under fielding, OPA!
Well, I see you talked about this stuff a fair bit in your previous column. Dang real life got in the way of my statspeak reading, and I’m just now catching up. :) Still, some of those questions might still be relevant… -j
It seems like you tackle the range penalty issue quite a bit here. Have you thought about weighting the range part a bit higher to take this into account? Or determining the order of importance of all of these instead of just range and then weighting them along those lines? That might get rid of your problem.
I’d rather have a guy that can get to a ton of balls as opposed to one who would have a higher rating/percentage of sorts but with much less range.
Stabbing at the ball or slightly booting it can be the difference between runs scoring as well. With a runner on second, a grounder to leftfield could score him, whereas excellent range could keep the ball in the infield, even if the SS doesn’t end up coming up to throw (or even ends up slightly booting it).
I suppose the weights can come from getting some sort of run value on each part of the process.
The problem with the range penalty is that all my system sees is that the ball was handled by the SS. It doesn’t take into account that a ball hit right to the SS has a different out expectancy than a ball in the hole.
I’m surprised that the hands and arm scores are in the same ballpark as the range score. I thought that catching and throwing errors were much rarer than an infielder getting to/failing to get to a ground ball
Yeah, I just feel like there needs to be some type of inverse relationship, perhaps, between balls not gotten to by shortstops (and therefore not bobbled) and balls gotten to by others but bobbled/not made.
Have these outs/run expectancies been tackled elsewhere or incorporated into other systems? It would seem that would be mecca-important in terms of fixing the range penalty. For instance, we could compare the expected outcomes of all balls in the edges of certain player’s ranges against actual results.
Perhaps at the outer edges of all players with similar ranges, Everett had a better percentage of conversion, though measuring him relative to everyone outside of his peer range would make him look worse in certain aspects.
ekogan, remember that these are all “above expectation”. It is rare for a fielder to throw a ball away, but it’s fairly devastating to the out probability when he does. e.g., when a ball is on its way to the 1B’s glove, the out expectancy is something like 99.7%. If the 1B catches it, he gets .3% worth of credit. If he drops it though, he gives back all of that 99.7%. Thankfully for most 1B’s, they catch most of the balls thrown their way.
You’ll have to forgive me because I haven’t completely digested your system yet. But how does this system compare to other retrosheet-based systems like that of Sean Smith’s TotalZone (first discussed on this site when he wrote here) or Dan Fox (SFR)?
Is the primary difference that you’re breaking down fielding events with a bit more granularity into range, hands, and arm (which I love–the Jeff Keppinger stuff is dead-on in line with scouting reports on him), and then combining them again to produce a composite picture of the player? How would you expect it to differ in overall results from those two systems, if at all (which are pretty similar to one another)? Given that you’re using more event types (and thus smaller sample sizes), do you think that it’s likely to be less stable from year to year than TotalZone or SFR?
Sorry, I’m sort of a fielding stat junky and I like to think about strengths and merits of different systems. I find these retrosheet systems to be really fascinating because they have the potential to allow simple folks like me to do fielding research without relying on BIS or STATS inc for hit location data.
Have you considered taking a look at the 3rd basemen these shortstops play alongside to see if the 3rd baseman’s range impacts balls a shortstop would get to? I’m guessing it wouldn’t matter a whole lot, but in the case of a player playing alongside a very gifted 3rd baseman, perhaps it takes away enough plays for a SS’s range to be affected.
Also, what is the correlation between Adam Everett stabbing at balls at the edges of his range window vs. Jeff Keppinger stabbing at balls at the edges of his range window. The greater your window, the less likely you are to field the balls in the outer reaches of that window, however, is there a normal value that allows us to know that Keppinger actually might have better hands than Everett, especially when attempting to play a ball outside of comfortable for either player?
Justin, I’m taking a lot of my cue from Dan Fox’s SFR and Sean’s TotalZone. My guess is that the composite results won’t be all that different, and in that respect, I’m re-inventing the wheel. You’re right in that my hope is that the improvement is breaking things down a little more fine-grained-ly at least into component skills. One study I’d like to do is to see which skills translate from position to position, and maybe identify some players who would benefit from a position change. And I will be doing stability studies soon enough. This is a big huge project that I will be looking at for the next few weeks.
Spitting, that’s the problem with a lot of fielding measures. De-tangling the interactive nature of fielding is going to be hard, no matter what you do. TangoTiger’s “with or without you” framework might be a good starting point to answer that question though.
Yes, I’ve read TangoTiger, and I understand the nature of things. I do, however, think there should be a measurable in your formula that helps you determine the difference between “good hands” and “good hands at the edges of range.” There’s a huge difference between Adam Everett gobbling up what comes to him vs. what he can’t quite handle and another SS who gobbles up most of what is in his area even if he can’t get to balls Everett gets. Everett would help more by getting to balls in the hole when there are runners at first and second, but he’s not doing much good when there’s a runner at third or at simply getting the batter out. If you fine tune you’ll get more detailed results as to when which shortstops matter more.
I actually looked a bit at that topic in the 1993-1998 Retrosheet files that have the hit location data to do it. What I found was that (not shockingly) most balls right at the SS were picked up, but that the reliability from year to year on this percentage was low. This says that any variations from year to year and player to player are largely the result of chance. However, the reliability went up when looking at the zone to the SS’s left and right. Makes sense.
The part that I struggle with (and the part that dooms all Retrosheet-based fielding systems) is that we don’t have good hit location on where the ball went in the recent RS data. It also seems like you’re arguing for possibly controlling for the game state (at least for where the runners are) probably in some sort of run-expectancy way. That’s an interesting idea…
@PC, If you do look into position changes, I’d be interested to see how it relates to what the fan’s scouting report data would suggest on the same issue.
Looking forward to seeing this thing develop! -j
Yes, game state or runner positioning is what I’m talking about. It’s great that Adam Everett gets to balls in the hole, and it probably gives him a chance to create more outs overall, but where runners are positioned makes a difference as to the possibility that he makes more outs and more importantly prevents more runs than other shortstops. Like you said, there’s likely not enough data on where the ball went, but I think this is critical in coming up with who is the better defensive player.
It’s amazing how leaders in OPA! and bottom-feeders, as well, turned out to be the guys we’d expect to see in such a list.
One thing which may distort your ability to isolate components is positioning.
As a thought experiment, strong armed shortstops “should” play deeper than average to take advantage of their arms [ in theory; consider that if they played shallower than average, the ball would get to them quicker; and assuming they got to the ball, with a shorter throw and more time to make it, they would lose the advantage of their arm strength.
As I understand your explanation of OPA!, a shortstop with average range and a strong arm in reality, by playing deep would “reach” more balls, preventing more outfield singles, but not necessarily throwing out more batters. You would measure his above average skill as “range” rather than arm.
In more specialized situations (infield in; double play depth?) there might be less variation in positioning, so that you could isolate the skill better, but of course much smaller samples.
Is there anything interesting to say about grounders which are turned into outs without a throw?
OPA! does look at double play-possible-GBs a little differently, although, of course, I don’t know if the fielders were positioned differently.
Grounders without a throw (3 unassisted?) are counted as the first baseman fielding the ball and oddly enough, throwing to himself. It’s a bit of a flaw in the system that I’m not sure how to correct.
Since the article is about shortstops, I was thinking about shortstops fielding a grounder and stepping on 2nd. Same thing as your example with first basemen, where sometimes they have an option between making the toss to the pitcher covering and taking it to the bag themselves.
Interesting point there – wouldn’t you want to separate the throw on a 3-1 play as a different skill than throws to other bases? It’s often underhanded, and while accuracy still matters, timing rather than velocity is the key skill.
Just to restate my initial point, players should position themselves in some kind of equilibrium which takes advantage of their relative strengths and weaknesses (for arm and range at least) so that you cannot infer from retrosheet data what the underlying skills are.
But maybe I am wrong and you usually will be able to isolate the underlying skills. It’s a practical question, and it’s worth examining.
And if you are able to isolate the underlying skills, it may be from a counterintuitive consequence that true range aside, strong armed shortstops have a surplus of infield hits [get to more balls] , just because they play deeper.
Joe, good points all around. Now to figure out how to correct for this…
Some really interesting stuff here. People usually lose me with these kind of stats, but I have to say I understood every word of this and found it to be pretty cool.
I think other metrics have supported your point that Jeter has below average range, especially going up the middle to his left.
One thought I came up with after reading this article, is that the same weight is given to a thirdbaseman who fields an easy ground ball and makes the out at firstbase to a diving stop on a hard hit ball who also records the out at firstbase. I think there needs to be some distinction to the level of difficulty of the play. Does the OPA! system do this? If not, are there other defensive rating systems that take level of difficulty into account? I like that approach.
Xeifrank, if there’s one flaw in the system, that’s it, and if Retrosheet had more detailed data, I would so do it. I have to settle for figuring out the average value, whether it was a two-hopper right to the 3B or an amazing running/diving/leaping play. Not ideal, but… free.
[...] Statistically Speaking | MVN – a statistical and sabermetric baseball blog
Fill in your details below or click an icon to log in:
You are commenting using your WordPress.com account. ( Log Out / Change )
You are commenting using your Twitter account. ( Log Out / Change )
You are commenting using your Facebook account. ( Log Out / Change )
You are commenting using your Google+ account. ( Log Out / Change )
Connecting to %s
Notify me of follow-up comments via email.
Create a free website or blog at WordPress.com.
Get every new post delivered to your Inbox.