The visualization here would be helped by showing a difference plot so you can see exactly where the biggest effect is and probably discern the level of noise as well. Any chance you could show that?

Larry,

that is a great idea. If I have time later in the afternoon I will try to do that.

really liking these features!

Thanks, Dave. Looking forward to it. Also, what Ben said.

Great job on the graph showing the difference between the Phillies and the opposition's defense. I think that could be a worthwhile exercise for all teams.

Nice to see that difference graph. Really shows how bad Burrell was, and how good Utley+Werth in right field have been, though both may be somewhat exagerated by the left-handed power in the Phillies' line-up.

I'm curious about two more things

1) How many balls in play are hit into a typical pixel? About 4400 balls in-play for a typical team over a season, halved for home games only, means 2200 total balls in play, looks like about 400 pixels in your charts, means 5-10 balls per pixel. Is that right?

2) I thought Victorino was a good CF, but both dead center and right center show holes without too much blue to make up for it. Interesting. PMR gives the Phillies' overall CF a high rating, for comparison.

Larry,

The data I used to make the images is from all balls in play from 2005 to 2008. 2005 is when MLB started recording hit locations for the Gameday application on their website, so that is my start date. I have a total of a little over 13,000 balls in play for Citizens Bank Park, so each image should be from around 6500 balls in play.

Each pixel has around 20 balls in play, but to make the image each pixel's color is determined by the average of its run value with that of its eight neighbors.

I was really expecting the Phillies' center field to look good. Since 2005 they have had average to very good centerfielders (at least according to UZR):

2005 Lofton and Michaels
2006 Rowand
2007 Rowand
2008 Victorino and Werth

I definitely think the more sophisticated fielding metrics, like PMR and UZR, should be trusted more than this image. I am really not sure why centerfield here does not come out more positive.

Ah, forgot to note mutiple years, with the smoothing seems like enough data to trust, I was just trying to get a feel for sample size error.

What's really weird about CF is that's where the park has an unusual feature (high angled wall) and you'd think the Phillies CF would be better at dealing with it, giving an advantage. But, it seems that whatever adjustment is being made may not be productive. Looking at this sort of analysis for all strange park features would be really interesting. I've seen some discussion about "parks with weird features" yielding extra home field advantage. This might help quantify which features actually lead to advantages.

Also, I'm not criticizing these charts, just trying to understand where they diverge from other metrics, since that can be instructive. The defensive metrics usually just measure out-rates, and ignore increases in extra base hits, where these charts account for that. On the other side, your data isn't normalizing for whether Phillies batters or pitchers were better than their opponents as well as any handedness effects that might come from the Phillies having two big lefties plus a couple switch hitters.

In summary, more please!!

What really stood out to me was first base!

I disagree with all this. I've seen every game the Phillies played in the last 10-30 years and Burrell DID NOT cost them 5 games a year. 3 things: the writer is correct in that Burrell does not go back on balls hit over his head well. And that's about all he's right on. Burrell goes sideways as well as expected and comes in on the ball ok. What the writers misses is his arm. It's one of the best in the game. It's a cannon. Many of his mistakes are made up for by throwing out a runner advancing to second or trying to score at home. What the Phillies did was not take chances in tight games when they had the lead late in games. I've see much worse outfielders than Burrell playing in the majors right now.

Larry,

yeah the next step to make these better would be to control for types of balls in play. The GameDay data has if the ball in play is a groundball, fly or line drive. I should do the comparison for each ball in play type separately so I don't penalize the Phillies if, hypothetically, they had to field more line drives than the visitors.

Brent,

Yeah Howard looks really bad here. UZR has him as just above average, but David Pinto's PMR has him well below average.

Mike,

I meant to say that Burrell's defense cost the Phillies a total of five wins over from 2005 to 2008, not five wins a year. You are correct that this analysis does not value arm strength or how many runs were saved or lost due to that.

This is very neat. Not only does it show the performance of individual players, it shows defensive efficiency with interactions of players. It might be interesting to use a different scale for the infield since the out probabilities are consistently high that it's difficult to tell the difference between infields.

Isn't this also pointing out that they should have moved both their fielders more toward the lines? I would think the goal would be to minimize the green area in the graph at all. If you moved the corner fielders you would move their blobs toward the lines, do this until there is a slight bit of green between them and center and would would have eliminated a lot of green on the lines it appears.