Dave,
You might want to take the expected DER to also include the park DER, which I have provided somewhere on my site.
Posted by tangotiger on 01/14 at 01:59 PM
DER Responsibility
January 12, 2004
A quick graphical look at David Pinto’s fielding analysis.
During the offseason, David Pinto has conducted a series of very good defensive analyses called the Probabilistic Model of Range (that’s a mouthful), and MGL has also finished his Ultimate Zone Ratings for 2003 . These are two excellent reviews of individual fielding skills, based on play-by-play data.
I’ve been spending some time with Pinto’s analysis, because it focuses on Defense Efficiency Ratio (DER), which can also be applied to Win Shares. In fact, DER is a key component of the Win Shares calculation for splitting runs allowed reponsibility between pitching and fielding.
David’s work analyzes the team and individual fielding by calculating the difference between expected outs and DER (based on where and how hard the ball was hit, and league-average fielding) and actual outs and DER (based on what actually happened on the play). If a team’s fielders converted a higher ratio of those batted balls into outs, they registered a positive number of outs vs. the league average, and a higher actual DER.
And if a team’s fielders didn’t have the range to convert those into outs, they registered a negative number of outs vs. the league average, and a lower actual DER.
So, my first question was: is DER a good indicator of good fielding? Here’s a graph of the good fielding teams (as measured by Pinto) vs. actual DER:
The better fielding teams (to the right on the “X” axis), do tend to have better DER. The R squared of this data is .33, which is okay. As you can see however, teams such as the Blue Jays were better than their DER indicated, and teams such as the Athletics were not as good as DER would indicate.
MGL’s assessment of these two teams, by the way, appears to be consistent with Pinto’s.
Now, here’s a graph of each team’s expected DER vs. their actual DER:
As you can see, the relationship is a lot tighter here. In fact, this is a pretty good graph. It shows the very strong relationship between the nature of the batted ball and the actual outcome, and it shows the relative impact fielding can make. The teams that are above the fitted line are the good fielding teams, and the poor fielding teams are below the line.
Here’s the lesson I take from this graph: the difference between a high DER and low DER is not likely to be just fielding. In fact, fielding does not even account for a majority of the DER variance among teams.
Fielding only impacts a maximum of 15 to 20 points of DER. As an example, take the Tigers and Indians. They both had an expected DER of .690, according to Pinto’s analysis. But their actual DER’s were .683 and .698, respectively. A difference of 15 points, which is one of the widest gap of any similar teams. The biggest gap between similar teams appears to be the one between the Yankees and Braves—about 20 points.
But actual DER varies by quite a bit more than twenty points—it ranges from .675 to .720, or 45 points. So, how should DER be used to evaluate the split between fielding and pitching? Very carefully.
I think Bill James was very aware of these issues when he created Win Shares. He did use DER to allocate Win Shares between pitching and fielding—higher DER results in more credit given to the defense—but he muted the impact. Actually, the manner in which he muted the impact makes it really hard to analyze this aspect of Win Shares.
But I plan to keep struggling with it. And I hope to propose an alternative or two (including one sent to me by Charlie Saeger) over the next couple of weeks. Stay tuned.
Thanks, Tango. I didn’t include your park factors for this analysis, because I wasn’t sure how to integrate them with Pinto’s analysis. Most of the time, I do include them.
Posted by studes on 01/15 at 08:01 AM
According to the above link, you would need to subtract .025 to Colorado’s expected DER. In your chart above, Rockies’ expected DER is .685 (which assumes they play in a neutral park).
Playing in Coors, that drops it down to .660.
Their actual seems to be around .678, making them +.018 per BIP or about +60 runs.
Pinto has Colorado as below average, and MGL has them as above average.
But, nowhere near +60.
Can you detail again, using the Rockies, how you got the expected DER?
Posted by tangotiger on 01/15 at 10:36 AM
Expected DER is quoted directly from Pinto’s analysis. My impression was that park factors were already included in his analyses, and that expected DER does not assume a team plays in a neutral park.
I’m sure there are significant differences in his home park approach, compared to MGL’s. I’m guessing that’s one of the biggest reasons for differences in their outcomes.
Posted by studes on 01/15 at 10:55 AM
You’re right, park is included in Pinto stuff.
The 4 largest problems with Pinto model are:
1 - Pinto uses 1 year data to establish “true effects”, where UZR uses multi-year… so, if a park happens to have looked like it played as friendly to fielders, but historically it’s not, then you have a problem. Just because it looks friendly based on limited sample doesn’t make it so.
2 - I think Pinto has a problem of “comparing to himself”. His maximum likelihood estimation probably includes the player himself in establishing the chances of making an out.
3 - No grid locations used, especially problematic in the OF (CF), where you’ll find the biggest discrepancy between Pinto and UZR. Pinto hopes that the rest of the parameters will capture “hit deep”, but I’m sure they don’t.
4 - Too many slice locations, and they are treated separately. There should be a “distance” parameter that shows that slice S and slice T are next to each other. Currently, he can have the possibility of slice R,S,T of having an out rate of .60,.40,.60 (assuming all other parameters the same). That really makes no sense. There should be another parameter that includes distance between slices as a way to smooth the probabilities.
Posted by tangotiger on 01/15 at 04:13 PM
Btw, Pinto’s model is a great first step, but it should be treated as a work in progress.
Posted by tangotiger on 01/15 at 04:14 PM
Page 1 of 1 pages
Commenting is not available in this channel entry.
| <<Previous Article: "Morgan and Those Who Followed" | Next Article: Baseball's Entertainers>> |