Article: A New Metric for Shortstop Fielding

[Thrylos]

Recently when I was looking for an objective way to numerically describe how good Engelb Vielma's glove for my Minnesota Twins' top 40 prospect list, which is fairly easily understood as a concept, I came up with a simple metric: The percentage of putouts that resulted in a double play.---

This is an edit to clarify a few points that were made previously that may have been confusing.

I did a bit of research to see whether it will pass the stink test, and so looked at Omar Visquel's Gold Glove seasons (1994-2001). Visquel's percentage of putouts that occurred during double plays (DP/PO) was 40%, so I concluded that Vielma's 39.6% was indeed encouraging. Furthermore, I used this metric as supportive of what I have seen with my eyes this season to suggest that Jose Polanco, with a 52.3% DP/PO rate, despite the rumors, is a very good fielding shortstop. This resulted in a major upset on the top 10of my Twins prospect list, and a hearty discussion of the metric, among other things, here.

 

Conceptually it is very simple metric: An effective shortstop will turn more batted balls into outs than a less effective shortstop. It is affected by many things like range and arm, but it is not perfect. It misses the number of chances for double plays as a normalization, and it does not help describe how the shortstop was with the glove when there were no putouts. So I did three things :

 

a. When I first thought of this, I thought that putouts were the way to go, because for some reason it helped tell more about a shortstop than assists. After a bit of discussion and noodling, this is not really valid. I was wrong to use putouts for the denominator. I think that total chances (TC) are a better denominator, so that is it. Instead of percent of putouts that were DPs, I am using percent of total chances that resulted in double plays (%CDP, from the DP/TC formula)

 

b. To add something into the measurement that describes a shortstop in a non-double play situation, I went back to an old (and tired) friend and gave it new life by marrying it with %CDP. This is fielding percentage (FP), which by itself is inadequate to wholly describe fielding, but is a very simple conceptual metric: Errors over chances. So this compound measurement is simply: The percent of total chances that resulted in double plays multiplied by fielding percentage, or FP. Because that is a mouthful, I am calling it shortstop fielding effectiveness, or SSFE. (The name is similar to the other metric I devised to simplify pitching effectiveness: Pitching Effectiveness.) For the equation-inclined: DP/TC x FP=SSFE.

 

c. To normalize for the chances of a double play, or what percentage of total chances were with a man on base, I normalized against the league for a full season, assuming that the chances for a double play are pretty much the same for all teams over the course of 2280 games (152 times 15). A league normalization would be good enough. So I calculated the average SSFE (which was 13.5) and divided each player's SSFE by that average, resulting to a normalized value, which I call nSSFE. A nSSFE of 1 is average, everything above 1 is above average and everything below 1 is below average. To make it look numerically a bit more familiar (think ERA+ and OPS+,) I multiplied by 100, making the average 100, like those other two metrics, creating what I call nSSFE+ (still a mouthful). 139 players played shortstop in the bigs in 2014.

 

Does it pass the stink test?

 

Here is the nSSFE+ for all MLB shortstops in 2014 with more that 200 innings at short. In blue are the above average shortstops (nSSFE+ 110 or more) and in red are the below average (nSSFE+ 90 or less.) Since this is a Twins-focused blog, the Twins' players are in bold.

 

 

 

I think that it does pass the stink test if you look who are in the blue categories (JJ Hardy, S. Drew, et al) and who are in the red (Derek Jeter, Jimmy Rollins et al.)

 

Is it a perfect metric? No; because there is no such a thing. But I think that it is easily understood and can be valuable. And it is better than the "eye" alone. The two together may be awesome. Could it be translated to other positions? I will try to play around, but feel free to play and tell me I will eventually look to see how the average moves with history, and potentially refine it, but this is the first attempt.

 

Originally published at The Tenth Inning Stretch

 

Click here to view the article

[John Bonnes]

You're taking on a challenging task: to come up with a fielding measurement that uses numbers fairly accessible for the minors, and do one must work with a very limited set of data. One also must make a lot of assumptions about the noise that is inherent to such an approach. I think a lot of those assumptions are a little optimistic.

 

That said, the results are encouraging. From a 10,000-foot level, it seems to identify (reputed) good vs (reputed) bad shortstops last year in MLB. My questions:

 

1. Have you done so for other years (to make sure this methodology didn't match your data as opposed to vice versa)?

 

2. Are the scores for a particular player fairly consistent from year to year? 

 

3. If you take players whose defense has slipped do to age (say from their mid-20s to their early 30s), do you see a negative trend?

 

Thank you for sharing this Thrylos. It does make me wonder, even with all the noise surrounding double plays:

1) are they an underutilized indicator of fielding2

2) is the ability to turn them an underappreciated (and potentially very impactful) skill, like pitch framing?

[old nurse]

It is pretty much out of the shortstop's control if there is someone on base so the put out is a DP or not. The higher the % of plays a shortstop makes that are a DP could indicate there is a groundball pitcher or  pitchers who put a lot of runners on base. I don't think it reflects on the shortstop.

If there is a baserunner and there is a play involving the shortstop, be it put out or being in the middle of the double play, the % should be near 100 for success rate.

[Seth Stohs]

Correct, old nurse. Consider that if the shortstop fields the ball and makes the throw to 2B, the throw is much shorter, many times even underhand. If he gets the put out, it means he had to catch the ball and throw it to first, not having to field it at all.

 

As you've said, the other key is that someone has to be on base and the pitcher coaxes a ground ball. 

[Thrylos]

You're taking on a challenging task: to come up with a fielding measurement that uses numbers fairly accessible for the minors, and do one must work with a very limited set of data. One also must make a lot of assumptions about the noise that is inherent to such an approach. I think a lot of those assumptions are a little optimistic.

 

That said, the results are encouraging. From a 10,000-foot level, it seems to identify (reputed) good vs (reputed) bad shortstops last year in MLB. My questions:

 

1. Have you done so for other years (to make sure this methodology didn't match your data as opposed to vice versa)?

 

2. Are the scores for a particular player fairly consistent from year to year? 

 

3. If you take players whose defense has slipped do to age (say from their mid-20s to their early 30s), do you see a negative trend?

 

Thank you for sharing this Thrylos. It does make me wonder, even with all the noise surrounding double plays:

1) are they an underutilized indicator of fielding2

2) is the ability to turn them an underappreciated (and potentially very impactful) skill, like pitch framing?

 

Thanks.   About those questions:

 

As I indicated, I need to further look into more data historically to see how that metric looks from year to year and even more importantly, how the league average Percent Chances that are turned into double plays looks year to year (and league to league.)   Is there portability?  Is that number fairly constant?  How does it look in the minors?  (Alas, we do not have very good league total chance data for the minors and it is a pain in the neck to assemble)  And I do suspect that individuals will change from season to season in both ways.  I'd love to look at Jeter or Smith or someone else for example.

 

Regarding the double plays:

 

a. The normalization against the league average Percent Chances that are turned into double plays takes care of the ground ball pitcher/man on base factor.  It assumes that the probability for a shortstop to encounter a chance for a double play is the same at any PA across the league.  To give you an analogy (and we are talking about 1500 innings.) Let's imagine someone has exactly162 days to bike from NYC to LA.  This is 2776 miles, which averages out to 17.1 miles a day.  Of course, because of external factors (not much different than groundball pitcher/man on base) during some days someone might do 40 miles, some days someone might rest and some days might hit 17.1 exactly.   The mathematical problem with these assumptions is that, while you are able to see the big picture and say that someone's rate of > 17.1 miles a day is better than the average, if you a measuring just a part of that (like from Chicago to Des Moines) you do not know whether it is significant.   I am not sure what sample is big enough for this to discuss with some certainty, but I think that a third of a season (500 innings) or even a quarter (350) might be fines.  Still need to do the math...

 

b. About the double plays and why double plays.   I am using them as the numerator, but I am measuring baseball skills that turned into results.  It is a bit of an abstract concept, but if you are looking at batting average (Hits/ABs) you are using hits as a numerator, but you are measuring the ability of a hitter to make productive contact utilizing his contact skill.  So the assumptions are: turning more double plays than another player requires better fielding (a combination of range, arm, instinct, accuracy, whatever -cannot differentiate) skills.  So if someone turns more double plays than another, he must be a better fielder somehow.  Because, if you cannot turn the double play, you'd likely have a hard time making a diving catch on a line drive.   Also the metric is further refined by accounting for errors in routine situations.   Does that make any sense?  I am really counting double plays, but I am measuring fielding skills, and the ability to turn double plays are a way to measure those skills. 

 

As I indicated, I just wanted a simple objective (read one that is based on data) measure that will give me a bit of confidence about someone's fielding and it is easily calculated (good luck getting pluses or minuses or UZRs or RZRs for minor leaguers) with the data we have.  I think that we rely too much on scouts (who are great btw, but they too have a SSS since they watch someone for about 5 games on average) for fielding evaluation...   Perfect?  No by any means, just another factor in evaluating someone.  

[Eris]

One item that is missing in this evaluation is the importance of the second baseman in converting DP.  If this is the case, an above average 2B should elevate a SS DP metrics and conversely.  Can you test for this?  

[Thrylos]

One item that is missing in this evaluation is the importance of the second baseman in converting DP.  If this is the case, an above average 2B should elevate a SS DP metrics and conversely.  Can you test for this?  

This is accounted a couple of times: in the normalization process, the same way of accounting for a ground ball pitcher, for example, and in the denominator, using all shortstop chances (i.e. by broadening the container.)    I am actually looking at the second (and third) basemen data with a derivative measure.  Would be interesting to see how much correlation there is.

 

Definitely a standard deviation & sample size issue (a better infield.)

Edited January 11, 2015 by Thrylos

[ashbury]

My snap reaction would be to reach out to people who have been working on defensive metrics for a long while.  Two names that come to mind are Chris Dial, and Vince Gennaro.  Both are SABR members so I could put you in touch with them if you wanted.  The purpose would be to ask if this is a computation that has been looked at in the past; it's not good to be trapped by the orthodoxy of the "experts" but neither is it good to ignore their experience.

 

Just as things like BABIP are in principle supposed to even out but sometimes don't, I think a year's DP and TC totals could be skewed for an individual player if the total opportunities are far from the norms, and I don't know how to easily get the number of situations where there is a man on first and fewer than two out.

[USAFChief]

I don't think it's safe to normalize chances for a double play. Just looking at the AL for 2014, Texas allowed 2015 runners via hit or walk, Oakland allowed 1675. That's around 2 more runners per game, which means a lot more DP chances.

 

Nitpic, same subject: shouldn't you be using 162 games times 15 (2430) rather than 152 times 15?

[Thrylos]

I don't think it's safe to normalize chances for a double play. Just looking at the AL for 2014, Texas allowed 2015 runners via hit or walk, Oakland allowed 1675. That's around 2 more runners per game, which means a lot more DP chances.

 

Nitpic, same subject: shouldn't you be using 162 games times 15 (2430) rather than 152 times 15?

 

Typo.  I am using league totals, so it is 162 times 15

 

I am all ears regarding alternative normalizations

Edited January 11, 2015 by Thrylos

[DuluthFan]

You may need to make adjustments to this equation. A double play is more of a team stat than an individual stat.

While a shortstop may initiate and participate in a double play, the completion of the play usually involves the fielding of one or two other players. The most common is the 6-4-3 (short to 2nd to 1st) double play. While the shortstop is the initial player to touch the ball in this play, the completion of the play also relies upon the fielding prowess of both the second baseman and the first baseman. You would have to remove their contributions to the play to use it as a stat for just the shortstop.

[Thrylos]

You may need to make adjustments to this equation. A double play is more of a team stat than an individual stat.

While a shortstop may initiate and participate in a double play, the completion of the play usually involves the fielding of one or two other players. The most common is the 6-4-3 (short to 2nd to 1st) double play. While the shortstop is the initial player to touch the ball in this play, the completion of the play also relies upon the fielding prowess of both the second baseman and the first baseman. You would have to remove their contributions to the play to use it as a stat for just the shortstop.

 

That's what the normalization does.  As FIP assumes that the pitcher has a league average defense behind him, this assumes that the ss has league average fielding partners...

[jorgenswest]

Thanks for the work. Every attempt at making sense on fielding data gets us closer to understanding the impact on the game. I hope to do some crunching on with the help of play index when time allows.

[Hosken Bombo Disco]

... Instead of percent of putouts that were DPs, I am using percent of total chances that resulted in double plays ...

 

MLB defines a fielding chance as a putout, assist, or error. As long as you are building something from the ground up… It might be time to bring the infield single into the equation too. None of the usual counting stats account for this yet, as far as I know. Even better, assign every ground ball to an infielder, as Bill James roughly suggested (I know I read that somewhere).

[Thrylos]

MLB defines a fielding chance as a putout, assist, or error. As long as you are building something from the ground up… It might be time to bring the infield single into the equation too. None of the usual counting stats account for this yet, as far as I know. Even better, assign every ground ball to an infielder, as Bill James roughly suggested (I know I read that somewhere).

 

That work has been done and it is the "zone" ratings.  These days are the Revised Zone Rating (or RZR) and the Ultimate Zone Rating (or UZR).  The latter chops the field in 75 pieces and assigns each one to a particular fielder.  And it looks at what happened to balls hit in any of those pieces.

 

Fun.

 

But nothing simple.  Nothing intuitive and nothing that one can calculate with a cell phone.  And we have no data whatsoever about that in minor league play.

 

(Bill James invented the much maligned "Range Factor" and then improved it with plus/minus in his fielding bible.  RF is something fairly easily calculated, too...)

Edited January 12, 2015 by Thrylos

[clutterheart]

What were Dozier's & Plouffe's percentage in the minors?

[BK432]

I just wanted to say I like what you're trying to do; as a previous commenter stated, any effort to improve defensive metrics helps. I can't offer any help or helpful critique, but thank you.

[Paul Pleiss]

most of this metric talk is way over my head after the first couple of paragraphs, but from what I understand, i like the new metric combined with other metrics we have. I like the outside the box thinking. I wonder how it works at identifying solid SS candidates from years past; does your metric continue to pass the smell test going backwards>?

[diehardtwinsfan]

Couple things I'm wondering.

 

1) Would it make more sense to look at the number of double plays as a percentage to the number of opportunities (i.e runner on 1st with less than 2 outs)? The number of opportunities should be distributed somewhat on a bell curve (at least for the guys playing the position all the time),but I'd think this would smooth out some of the irregularity.

 

2) What about heavy GB pitchers? I would naturally expect more double plays from a staff that has a GB focused staff. I'd think that would overstate the defensive effectiveness of those SS due to the opportunity issue.