Fielding Stat Theory

[sansterre]

I'm opening this discussion by stating that OOTP 11 and UZR will make this discussion relatively obsolete; that said...

For my league I've tried to develop an effective fielding metric/stat. Range factor and error percentage are both important, but each terribly incomplete. I've attempted to create my own metric - I'm pleased with it but I would appreciate feedback to consider flaws in my reasoning, or possible improvements. Here's my methodology:

Step 1: Plug my league's stats into TangoTiger's Run Expectancy Matrix generator to determine relative linear weights to different events (1B/Out, etc.) This is especially important for me, in light of the fact that I'm playing in 1902.

Step 2: Take away a certain number of putouts for each position. An assist is a play that the fielder makes. A putout is either 1) a play the fielder made or 2) a play another fielder made that this fielder simply had his foot on the base for. First Basemen, for example, have sky-high putout totals, for which they merit very little credit. I feel that fielders deserve credit for the first kind of putout, but none for the second. For this reason, I have attempted to remove these plays from each fielder's total. (Catching the ball in the second kind of putout is a skill, but a failure shows up in errors, so I'm not too worried about that.)

The table used is one I mostly guessed at - actual data on which position gets how many of each kind of putout is not something I know how to locate, and would be a bear to create. Listed are the number of POs actually credited to each position from the listed total:

1B: 5.4% of total POs
2B: 49.6%
3B: 82.6%
SS: 67%
P: 43.2%

Again, these rely on guesswork, and I'm open to alternate guesses, but these seem at least like a step in the right direction. Once you take the player's assists and add to them the modified PO total, you have the player's adjusted Range, or aRange for short.

Step 3: Player fielding doesn't consider the team environment. There are three major variables in this regard: BABIP, Ks and GB%. The higher a team's BABIP is, the more chances the team is getting to make plays. To use an exaggerated example: a player with an aRANGE of 4.00 on a team with a BABIP of .200 is getting far fewer chances to make the same number of plays as a player with an aRange of 4.00 on a team with a BABIP of .400. And so on. Furthermore, a player on a team that strikes out a lot of opposing batters will get fewer opportunities to make plays, because more outs are already at the plate. GB% I'm aware of, messing with it just seems more difficult than its worth for the moment.

Anyhow, so you multiply each players aRange by (1- Team BABIP) and divide by (1 - League BABIP) (penalizing for bad fielding), then multiply by the league's average chance of a non-strikeout out (1-K/IP/3) and divide it by the team's average chance of a non-strikeout out.

Step 4: From here, you give each out created (adjusted aRange) the value the linear weights from step 1 gave an out, and penalize each error the value of one single (plus a bitty bit, most errors seem to be singles, but some are doubles.)

Step 5: This is just the boring stuff - divide by innings played, multiply by 9 for the arbitrary 'game' rating. Determine league average for the position from there, then take the difference between a player's 'game' rating and the league average, multiply it by their number of innings and you get the estimated value above average for their fielding that year. Good times.

Conclusion: So yes. Very interested in ways to make this better, though I am quite pleased with it. Groundball data, better estimates on the adjusted POs... I'm all ears

And I may well have not explained things well. If that's the case, my apologies.

[knockahoma]

Hey, Sansterre.

I thought your post was interesting. I don't know if anything I say will have much value. I played shortstop in high school and college in the late 70s and early 80s, and in competetive softball until last year. I'm not overly familiar with some of the new fielding sciences. As you and Bill James have agreed, some skills are hard to judge by the numbers.

Step 1: Plug my league's stats into TangoTiger's Run Expectancy Matrix generator to determine relative linear weights to different events (1B/Out, etc.) This is especially important for me, in light of the fact that I'm playing in 1902.

Not familiar with that program.

Step 2: Take away a certain number of putouts for each position. An assist is a play that the fielder makes. A putout is either 1) a play the fielder made or 2) a play another fielder made that this fielder simply had his foot on the base for. First Basemen, for example, have sky-high putout totals, for which they merit very little credit.

There's some good sense, here, IMO. In fact, I got to thinking, " What if you took away putouts altogether for the other infielders? At first, my inner arguer objected. What about Brooks-like dives, or what about over the shoulder catches on texas-leaguers that take away hits? But, both of these examples tend to rely on quick first steps and reaction [read as: range], which should show up on grounder numbers. And for every great catch on a dive, or pop-up, you're looking at a couple dozen ho-hummers that might actually obfuscate our real search.


At any rate, just as a counter-weight, it might be interesting to isolate grounders from line-drive catches and pop-ups to see what happens to your numbers. That also would help deal with the gb% issue a bit.

From a real life perspective, just think of a coach at infield practice. You bat the ball around and make sure every ball is a grounder. After a period of time, just using grounders, could you evaluate the range of one player over another pretty accurately? I think so.

I imagine someone already has some formula dealing with just groundballs. It wouldn't be a holy grail stat on fielding, but it could be a parsing stat one could compare to forumae that count grounders, line-drives and pop-ups.

For example, what if using formula 1 (grounder+line drive + pop-up) you came up with this finish for shortstops:

1. Anderson
2. Bleier
3. Jones
4. Wilson

But, what if using only grounders (taking away line drives and pop-ups), you get this:

1. Jones
2. Bleier
3. Anderson
4. Wilson

Would that make it easier to adjust for team gb percentage when shortstops are only being counted on fielding grounders, instead of grounders/linedrives/popups ? In fact, what if you performed the same study on only line-drives, and only popups?

I'm just coming at this from a guy who played infield all his life and watched other guys doing it, too. Maybe it doesn't help with the math, but thought your ideas were interesting and so wanted to chat about it.