THE BOOK--Playing The Percentages In Baseball

From Ben:

5:35: On 95% of flies, it was readily apparent whether the fielder would catch the ball for the duration of the play. Only rarely was the outcome truly in question. Jeremy runs through each of those borderline plays, showing the trajectories of each primary fielder colored to indicate the out probability at each given interval.

95% is way higher than I would have guessed.  A corner OF has a bit close to 500 plays that he’s involved in.  If it’s only 5% where skill is involved and the other 95% is just noise, that means there’s only 25 plays that requires skill, meaning that a guy with great skill (Ichiro) to a guy with no skill (Dunn) would be worth +/- 12.5 plays at most.  And that’s presuming there’s no play that Dunn would make that Ichiro would miss (i.e., Ichiro is perfect in all the plays that requires skill and Dunn is useless).  Probably close to true.  Anyway, if we are more realistic, it’s +/- 10 plays, or +/-8 runs from absolute best to absolute worst.

I’m not buying it.  I think the gimme (noise) level is at best 90%, and maybe it’s even 85%.

If there’s one thing that should be easy enough to come out of the FIELDf/x, is to get the % of plays at each position that are gimmes.  That is a critical thing to know.

BRILLIANT! Three LOL lines:

SCOUTf/x........... Max Marchi

6:36: Here comes Max Marchi, who flew in all the way from Italy to deliver this talk. Hear that, John Walsh? When you live in Italy and you’re delivering a talk in San Francisco, you actually go to San Francisco. It’s just common courtesy. I swear—some people.

...

6:49: Sabermetrics sounds way better with an Italian accent.

...

6:55: Max’s final slide is a map with his air route from Italy overlaid. John Walsh should be ashamed of himself. Max visited Buffalo and Cleveland in between landing in New York and taking off again for San Francisco, so it’s safe to say that he’s seen some of our nation’s finest cities, as well as Buffalo and Cleveland.

I wouldn’t put too much stock in any specific conclusion reached from the Field F/X data that we had for the Summit. It was 13 half games’ worth, all in one park. So that 95% thing - yeah, that strikes me as high, too.

Max also hit Chicago and South Bend. I believe he’s in Sacramento now.

The 95% figure was from something I probably said in my presentation. That number is, of course, meaningless.

Will the Field F/X data be freely to the public? How large are those datasets?

WHoy: Today we can’t even be sure that Field F/X will ever exist outside SF and if it does then it might not be freely available. They won’t say who is going to pay for it. I read that the datasets are very large (15 rows for every second of every game)! It’s an exciting project.

Ben’s recap was fantastic – funny and really complete, at least of my talk.  He basically wrote down nearly every single thing I said, but it was funnier when he wrote it.

As for coming to California in person, he’s right, but I was running a little short of funds this summer, since my beach villa in Porto Cervo needed some roof work and my backyard clay tennis court needed to be resurfaced.  Maybe next year.

Agree that Ben’s recap is fantastic.

Alex/7, it’s 15 rows of data for each of the 16 to 19 players, umpires, and coaches on the field for every second of the game.

That’s hard to get the mind around.  A 3 hour game has 10,800 seconds.  We can discard the between innings stuff I’m sure, but still quite a bit.  And then for 16-19 individuals.

Even if that became available, I don’t know what I could do with it.

I mean 10,800 seconds 16 individuals at 15 rows per second is around 2.5 million lines....just for one game lol.

I think we’d need to devise a way of extracting only the data that’s needed for a particular type of query. Let the entire set of data on their servers, find of way of identifying and then selecting for download lust the relevant data.
for example, a few quick ideas for fielding
1. The location of everyone on the field at the time the ball hits the bat.
2. ditto when the ball hits the ground
3. location ball hits ground, with hangtime
4. when the ball is retrieved, time from catch to throw
5. for throw, identify location of target
6. length of time the throw was in the air

At each of these points in time (balll events), record everyone’s location

As I said, quick thoughts

Peter Jensen displayed some tables he derived from the data that drove his graphics

Dan Hennessey has a recap here:
http://knuckleballsblog.wordpress.com/2010/08/30/2010-pitchfx-summit-recap/

Rob Neyer has a recap here:
http://espn.go.com/blog/sweetspot/post/_/id/5041/fieldfx-is-going-to-change-everything

I plan to write a recap in the next day or two, also, though I’m still having a hard time digesting everything I heard at the summit.  There was a lot to think about, both from the presentations and the informal conversations.

Brian,

I was able to scan through the chat yesterday, but I wasn’t able to respond then, so I guess here and now seems like a fine place.

The purpose of evaluating routes and jumps and stuff during plays isn’t to come up with an all-encompassing metric like UZR. The purpose is to evaluate routes and jumps. The result doesn’t change whether the fielder had good or bad process, but that’s not the point. That information is useful for evaluating the player going forward. Yes, it would be ideal to present that information in terms of runs, but I think trying to accomplish that would have been beyond the scope of my pay level and my 20-minute presentation.

Jeremy, I understand now. I was a little confused at the beginning of your presentation when you were also displaying the remaining time in flight. A few minutes later it was much clearer what you were doing. I think my mind had already jumped forward to what Greg was doing in his presentation in judging range.

My comments about defensive run prevention are in the context of an overall measurement of defense, which is not what you were trying to do.

We did have some follow up comments in the chat, concerning your presentation on routes

1. that the more hang time on the ball, the more time the fielder has to move around under the ball without affecting the results. He might have a rational explanation for a circuitous route.

2. But, when time is short on a sharply hit ball, the fielder doesn’t have that luxury and must make best speed to his estimated landing point for the ball.

The second is the situation that would likely be the most informative of the outfielder’s ability to take a good route to the ball.

Like timing a batter from home to first - he won’t always be running hard all the way to the base. We need to select x amount of his best times to do the metric.

Brian, I used both the time since the ball was contacted and the time until the ball landed. I hope that addresses your concerns.

Jeremy, it’s good. I jumped the gun the first minute or two of your video, was confused on what you were showing. One of the other chatters explained to me.

Where did the presentations go? No longer in the download section of the sportsvision web site.

They put them up during the presentations but then they were down.....they said they’d be back up within a week tho.

Dave Allen’s recap:
http://baseballanalysts.com/archives/2010/09/pitchfx_2010_su.php

I had already updated the main thread, but thanks for noting it, because I forgot to tell everyone.

2010 Summit presentations have now been posted at http://baseball.sportvision.com/summit