Yesterday morning my colleague Spencer Manners introduced a new fielding data set, Inside Edge Fielding, that was unveiled this week at FanGraphs. This new data breaks down each play a player has the opportunity to make into six categories based on the difficulty of successfully making the play.
The category titles are pretty self-explanatory. For example, the Impossible category represents plays that are literally impossible for a fielder to make. It’s pretty straight forward. The caveat here is that this data is 100% subjective.
Inside Edge tracks every MLB play and scouts subjectively measure the difficulty of each play. This mechanism leaves a lot of room for differing interpretations between scouts on the difficulty. One scout may deem a play Remote, while another simply as Unlikely. Those that are strictly saber-minded and dismiss the more “old school” way to analyze performance may protest, but it’s an interesting new data point nonetheless.
Spencer used the data to compare the defensive value provided by two Mets outfielders, Juan Lagares and Eric Young, Jr., who are fighting for a spot in the starting lineup this season. This analysis will expand our look at the data a bit to the Mets as a team during the 2013 season.
So how do the 2013 Mets stack up? Well, first and foremost, only three players met the default qualified innings limit of 900 at a position: Daniel Murphy, Marlon Byrd, and David Wright. It’s no secret that the Mets roster churned a good bit last year, so this is no surprise. When we drop the innings limit to 700 we’re able to get eight position players, as listed in the table below. Note that the table is limited to only Remote and Unlikely plays, which are the most difficult makeable plays as deemed by Inside Edge.
| Name | Position | Remote Attempts | Remote Success | Unlikely Attempts | Unlikely Success |
|---|---|---|---|---|---|
| Marlon Byrd | RF | 5 | 0.0 % | 4 | 50.0 % |
| Juan Lagares | CF | 8 | 12.5 % | 8 | 25.0 % |
| Daniel Murphy | 2B | 14 | 7.1 % | 12 | 16.7 % |
| David Wright | 3B | 12 | 0.0 % | 10 | 10.0 % |
| Omar Quintanilla | SS | 8 | 0.0 % | 10 | 20.0 % |
| Ike Davis | 1B | 4 | 0.0 % | 3 | 33.3 % |
| John Buck | C | 7 | 14.3 % | 19 | 26.3 % |
| Eric Young | LF | 5 | 40.0 % | 3 | 100.0 % |
Byrd, Wright, Quintanilla, and Davis each had Remote play attempts and failed to convert all of them. All of the players performed better on Unlikely plays, which makes sense.
Something to note here is that there is difference in the categorical spread by position. Put simply, some positions see higher frequencies of more difficult plays than others. Roughly fifty percent of attempts by catchers are categorized as either Remote of Unlikely, for instance. Contrast that with all three outfield positions, in which roughly 90% of their plays are categorized as Almost Certain/Certain. Jeffrey Bellone at Beyond the Box Score does an excellent job of providing a breakdown of difficulty by position.
It also doesn’t do us any good to look at these numbers on their own. How did these players do in these categories when compared to their positional peers? The table below notes the ranking of each player at their position in each of these categories across the league in 2013. Remember that we’re still working with a minimum of 700 innings.
| Name | Position | Remote Rank | Unlikely Rank |
|---|---|---|---|
| Marlon Byrd | RF | 21/21 | 5/21 |
| Juan Lagares | CF | 11/25 | 17/25 |
| Daniel Murphy | 2B | 13/28 | 23/28 |
| David Wright | 3B | 27/27 | 24/27 |
| Omar Quintanilla | SS | 27/27 | 20/27 |
| Ike Davis | 1B | 23/23 | 5/23 |
| John Buck | C | 10/27 | 13/27 |
| Eric Young | LF | 1/18 | 1/18 |
Note that in some cases, like for Byrd’s and Wright’s Remote rank, over half of the players had 0%. In Davis’ case, no first baseman with over 700 innings converted a single Remote play. So even though they were ranked dead last for their position in those categories, they had plenty of company.
The data we’re using here doesn’t necessarily rank the quality of the fielder. It simply analyzes plays made or not made based on the most difficult categories. Remember, the sample sizes we’re dealing with here are very small and the data only goes back to the 2012 season. That being said, an argument could be made that, by either combining all of these categories or even just focusing on the two easier categories, we might build a somewhat accurate picture of the defensive quality of a player. If nothing else, it’s an interesting way to view which players consistently make the hardest plays.
I’m confused. What are the categories and how are they defined? Is remote something that shouldn’t be possible (perhaps named for all of the TV remotes that went flying on the play)?
So what do these new stats say about Young v.s. Lagares anyway?
Hi, Mike. The categories are as follows, in ascending order of play difficulty:
-Impossible (0%)
-Remote (1-10%)
-Unlikely (10-40%)
-About Even (40-60%)
-Likely (60-90%)
-Almost Certain / Certain (90-100%)
Impossible means just that. There is 0% chance the player makes that play and thus the scale starts at 0%. It’s a subjective way of defining the difficulty of a play. Remote is not quite impossible, but close to it. It could be thought of as the most difficult play a player can make that was actually possible to make. It’s just a way of determining which players frequently make the most difficult plays.
As far as Young vs Lagares: Lagares had more opportunities on difficult plays than Young last year, but when we start talking about such small numbers the percentages get skewed. However, the top of the scale (easiest) can show us how many of those plays a player should make. Taking all of the categories together gives us a picture of the (subjective) player’s defensive value. It might even be more interesting to just consider the middle portion of the scale when trying to make a distinction between players. It’s new, and it’s not perfect, but it’s an additional data point to consider.
So i went to Fangraphs to see if their distribution that they were claiming were truly what they were observing. These are over the last 2 years over all 30 teams, so the sample size shouldn’t be an issue.
1-10% – 9.12% (this bucket should really be 0-10%)
10-40% – 28.51%
40-60% – 57.19%
60-90% – 77.87%
90-100% – 95.43%
Ok, so if all types of plays are equally likely, which may or may not be true, then the expected success in each bucket should be the midpoint. So the expected success rate should be around 5%. Without doing any calculation, i’m pretty sure that the 9% observed value is significantly different from 5%. I’m guessing that some plays that are categorized in the 0% bucket should probably actually be in the 1-10% bucket.
When breaking the percentage down by position, i found that the OF had a higher remote success than other positions. I suspect human bias is going on here. It’s hard for fans to appreciate a good split second reaction infield catch/field but we all love a good long run in the OF that results in a spectacular dive or a HR robbing play.
As a new stat, it obviously still has its flaws to be worked out so i won’t harp on it too much, but it’s certainly interesting thing to look at.