THE BOOK--Playing The Percentages In Baseball

Tango,

How much does modern play by play defensive ratings use these hang times? I assume that it could help with outfielders to a small degree to quantify how hard a ball is hit. Could it be used as a how many SD away are they within each range of each batted ball type of thing?

Batted balls would probably need to be defined better in the future. Batted balls aren’t as clear as the 24 states, but some combination of GB, LD, FB, etc and hang time could be interesting to help define batted balls and defensive value.

(BTW, I am loving the book)

They aren’t used at all, at least by any of the defensive stats you’ve heard of.  There may be some teams that track hang time and keep a confidential defensive system.

If they don’t already, I’ll guess somebody around here will convince the Mariners to track it.

Ideally, for judging fielding of fly balls, you’d have information on where the player was standing before the ball was hit, information on where the ball landed, and how long the ball was in the air.

That way you could start to examine players (primarily outfielders) on how well they’re positioned and their range (considering their read of the ball). We could get these nice graphs where we show what the realistic range of a player is to each different direction, and where they should ideally play depending on batter/pitcher matchups, and all sorts of other fun stuff.

For ground balls, it would be much much harder, but I don’t see any reason that the above information couldn’t be captured for each play. Someone with more brains than I could probably write computer software to track the ball and location it lands from with a few fixed cameras (or possibly even the cameras used for broadcast).

And man the charts would be cool.

Sal, I was just thinking along similar lines about range: that it should be a function of where the player started and where the ball landed/was caught.

Starting position is probably something that could be tracked with the right cameras and/or people sitting in the right locations at the park, but I’m not familiar with BIS to know whether they’re set up to do this right now.

Colin did produce some great fielding graphs a few months ago, and I have a feeling that they’re a good start for this research.  For example, maybe we can use the “epicenter” of a player’s data points as his assumed starting position?
otherfifteen.blogspot.com/2008/04/can-we-measure-fielder-range.html
otherfifteen.blogspot.com/2008/04/derek-jeter-vs-troy-tulowitzki.html

Post 4 was marked for moderation and is now open.

Tango - since distance and slice are already tracked by BIS and Stats, how about starting your own project to get readers/the public/Retrosheet people to start tracking hangtime? Just create a simple form that allows anyone to upload the hangtime for each batted ball in a given game. All that anyone needs to track hangtime is a stopwatch and a DVR. To reduce errors, you’d ideally receive info from multiple trackers per game and then average the results.
Even if you don’t get data for every game in year 1, you’d still get some useful data, or at least get an idea of how feasible the project seemed.

The time to manage such an effort is outside my scope.  I’m hopeful that one day I can get a Wisdom of Crowd approach going to track batted balls and fielders movement.

I’ve had this idea in the back of my head a while.  I’d include analytics to evaluate the reliability of the observation and observer, to rule out intentional or unintentional bias and to weed out contributors who aren’t trying hard enough to be accurate, and some form of recognition for the best contributors, i.e. points for every submitted observation that is within X% of the eventual “accepted” value.  This to get people enthused about becoming regular contributors.  It’s a great idea, and I predict within 3-5 years it will exist.

I’m with you Greg.  I expect some sort of Palm or phone app that people would use.  A feedback score, ala ebay, is a necessity.  I do this with my Fans Scouting Report, whereby I throw out any ballot below a threshhold. I get about 2% junk ballots. 

I expect something similar, whereby the more extreme your data point is from the median, the less your data point will be weighted, and it also acts as your score.

We could do alot of good stuff, if we didn’t have to worry about mortgages and healthcare costs.

My idea - and it’d have to be a team doing it - is to simply have a camera that takes a picture at a designated point in the pitch sequence - either at release or when the hitter makes contact. Mount it somewhere in the stadium with good sightlines. Have one guy operating it. It should be relatively simple thereafter to figure out where each fielder was situated at that moment during the at-bat.

(I know there are some further issues with this - I’m not sure if something like a PC or a tilt-shift lens would take care of the problems of perspective distortion. You should be able to get pretty good results if you calibrate the camera to the field first, however. Have a guy walk around with a tape measure and set yourself up some reference marks.)

For hangtime - a ball caught in the air will necessarily have a shorter hangtime than a ball struck in the same fashion that isn’t caught, right? How do you control for that?

Tango #9: Amen.  Personally, I wish someone would figure out how to eliminate the need for sleep, I must spend 4 or 5 hours a night not thinking about baseball…

Colin,

I’ve got a model about 75% built to do pretty much what you described.  I pick a large number of reference points around the perimeter of the playing field, such as the wall angles, or fence height changes, i.e. readily identified spots.  Also, the bases.  Then I connect each point to each other point, until there is a dense spider grid all over the field.  Next, identify the XY coordinates of each reference point on an image taken from the fixed camera.  Then when an image is captured from the fixed camera, mark on the image the XY coordinates of the ball, or the fielder (I use their right foot), or whatever you want to locate.  Then I wrote some code to figure out which of the huge number of “cells” created by the spider web the mark is in (on the image), and then by knowing which cell it’s in, figure out where that cell is on the field.  It works quite well, but I just haven’t had time to polish it. 

I also came up with a way to figure out where the camera is just by marking the three bases, using their apparent relative positions compared to their true relative positions.  Just in case you can’t figure out where exactly the camera is. 

Like Tango said, so many ideas, so many other things to worry about…

...I wish I was smart.