SoxProspects News
|
|
|
|
Legal
Forum Ground Rules
The views expressed by the members of this Forum do not necessarily reflect the views of SoxProspects, LLC.
© 2003-2024 SoxProspects, LLC
|
|
|
|
|
Forum Home | Search | My Profile | Messages | Members | Help |
Welcome Guest. Please Login or Register.
Defensive WAR calculations
danr
Veteran
Posts: 1,871 
|
Post by danr on Jun 23, 2015 10:56:58 GMT -5
This is just speculation at this point by me, but I think these WAR calculations have too much value assigned to defense. Intuitively they don't look right. There are quite a few really good hitters whose WARs are low because their offensive value has been diminished maybe too much by their defensive value.
Defensive value seems more subjective than offensive value. Just because a player has less range doesn't necessarily convert into a measurable drop in value. However, his offense does convert into objective value.
If I am off-base, please someone, explain it to me.
|
|
|
|
Post by okin15 on Jun 23, 2015 12:15:27 GMT -5
Defensive value seems more subjective than offensive value. Just because a player has less range doesn't necessarily convert into a measurable drop in value. However, his offense does convert into objective value. If I am off-base, please someone, explain it to me. If you have a baseline for defensive value (a replacement level player or average MLB player let's say) then it's really not that hard to assign run values to plays that get made or don't get made, especially if you had perfect data. If 6/10 RF make a certain catch, then the guys who don't, and give up a double instead are essentially committing 2-base errors. Add these up (every outcome or expected outcome has a value, positive or negative), and you get a resulting number of runs saved or given up over a your baseline player. For example, it's pretty easy to see that if Hanley Ramirez lets two singles drop which are usually line-outs, and another two singles off the wall over his head for doubles, and another double or two down the line or in the gap that would be caught by many other fielders (all hypothetical), then he has to pay the price for that. He has less range (and a less sure glove when he DOES get there) and that factors in to the runs he gives back while in the field. I think you only have to look back at the two times we've won 3 in a row this year to see how defensive miscues can turn an inning, a game or a series. ADD2: Where you might quibble (and I don't know much about the updated metrics) is with the raw data. How accurate and reliable are the estimations (measurements?) of location, speed, and direction of each ball in play, as well as whether bizarre situations like IF and OF shifts/positioning are taking into account in a good or bad way. Lastly, like many other metrics, defensive stats are a record of what has happened, and don't necessarily correlate to accurate predictions, especially over small samples. Sorry, last thought: I believe both defensive metrics use typical values for plays made or missed, not the actual value to the game situation at that moment, so these are more like wOBA than it is like RBI's or OPS, so in that way they are trying to be more predictive and less recordative. Someone correct me if I'm wrong on this (or any other) point. |
|
danr
Veteran
Posts: 1,871
|
Post by danr on Jun 23, 2015 14:14:55 GMT -5
Thanks. That is more or less the way I thought it worked. But the difference between defensive metrics and offensive metrics is that offensive metrics measure actual events, rather than hypothetical ones.
You mention one way in which shifts might affect the fielder's performance. There are lots of things that affect a fielder's performance other than his own talent, or lack there of: Everything from the weather, sunlight and glare, a bird pooping on him, and the actions of the other players around him. to many others. He might have limited range, but he might be very good with his routes, his reflexes and his positioning. Another player might have great range but take bad routes.
Things do average out over time and fielding metrics are valuable in assessing a player's abilities. However, since they don't measure his actual defensive performance they are a different form of measurement from offensive WAR which is based on actual performance. So I think they should not be collated together but should be shown separately like a slash line: o4.2/d-.7.
|
|
|
|
Post by jmei on Jun 23, 2015 15:06:40 GMT -5
Err, defensive stats absolutely measure actual performance. A defender does or doesn't make a defensive play-- how much did it help or hurt his team compared to the average defender at that position? That's the same exercise as you would have on offense (a hitter makes an out or gets a hit-- how much did it help or hurt his team). None of the arguments you make are intrinsic to defensive performance. For instance...
There are legitimate criticisms about defensive statistics, but these aren't it.
|
|
ericmvan
Veteran
Supposed to be working on something more important
Posts: 9,027
|
Post by ericmvan on Jun 23, 2015 15:38:16 GMT -5
This is just speculation at this point by me, but I think these WAR calculations have too much value assigned to defense. Intuitively they don't look right. There are quite a few really good hitters whose WARs are low because their offensive value has been diminished maybe too much by their defensive value. Defensive value seems more subjective than offensive value. Just because a player has less range doesn't necessarily convert into a measurable drop in value. However, his offense does convert into objective value. If I am off-base, please someone, explain it to me. The other thing going on here is that we can measure with absolute objectivity the defensive efficiency of a team (BABIP allowed with errors treated just like hits), which is to say, the performance of all of the fielders plus whatever contribution the pitchers have to allowing harder ot softer contact. There is widespread statistical agreement that the pitching contribution is small. Our measures of it are admittedly inexact, and that is one source of error, but it affects all members of a team pretty much equally. When a play is not made, the metrics have to pin the blame on one or more fielders. That's another possible source of error. It's theoretically possible that Mookie is not really +6 in CF and Hanley -14 in LF, that it's actually +4 and -12, because 2 balls that dropped between them were incorrectly viewed to be Hanley's fault when they were actually Mookie's. It's very unlikely, though. The thing is, neither of these really affect the overall size of a typical defensive WAR rating. To have that happen, we would have to be very wrong about the pitcher's contribution to BABIP, and there's a lot of evidence saying we've got that very much in the neighborhood of correct. That leaves us with a pretty much correct view of the size of the defensive contribution, and hence of the variation among defenders, and hence of the size of the defensive contribution to WAR. |
|
|
|
Post by okin15 on Jun 23, 2015 16:02:10 GMT -5
Agree with JMEI. This is absolutely a record of what happened, and a much more useful one than RBI's I might add.
Tangentially, one thing I've always wondered about BABIP and defensive statistics is what do they do with HR balls that just clear the wall and are theoretically catchable. I'm guessing they don't count against a defender, but I'm assuming they do if they are caught. Does he get credit for saving a HR? Is the ball in play because it was caught, but not in play when it goes out? This is a curious (but probably irrelevant) loophole in the system.
Lastly, I'm wondering if all of this ought to be moved to an off-topic thread before it gets crazy in this hizzy.
|
|
|
|
Post by telson13 on Jun 23, 2015 17:42:00 GMT -5
Err, defensive stats absolutely measure actual performance. A defender does or doesn't make a defensive play-- how much did it help or hurt his team compared to the average defender at that position? That's the same exercise as you would have on offense (a hitter makes an out or gets a hit-- how much did it help or hurt his team). None of the arguments you make are intrinsic to defensive performance. For instance...There are legitimate criticisms about defensive statistics, but these aren't it. I think the (valid) argument is that the ultimate outcome for offense (runs scored) is definite, finite, and closed-ended. For defensive non-plays, there's a butterfly effect: you're not measuring an actual outcome of runs prevented, but rather predicting runs prevented. A non-play on defense has no finite outcome attached to it, because you never really know how many runs it led to, how much extra the pitcher had to work to get out of the inning, how much it shortened his outing, how much of the poor bullpen performance could have been avoided, etc. defensive metrics are best-guesses based on statistical averaging, rather than counting statistics. |
|
|
|
Post by jmei on Jun 23, 2015 18:45:00 GMT -5
Err, defensive stats absolutely measure actual performance. A defender does or doesn't make a defensive play-- how much did it help or hurt his team compared to the average defender at that position? That's the same exercise as you would have on offense (a hitter makes an out or gets a hit-- how much did it help or hurt his team). None of the arguments you make are intrinsic to defensive performance. For instance...There are legitimate criticisms about defensive statistics, but these aren't it. I think the (valid) argument is that the ultimate outcome for offense (runs scored) is definite, finite, and closed-ended. For defensive non-plays, there's a butterfly effect: you're not measuring an actual outcome of runs prevented, but rather predicting runs prevented. A non-play on defense has no finite outcome attached to it, because you never really know how many runs it led to, how much extra the pitcher had to work to get out of the inning, how much it shortened his outing, how much of the poor bullpen performance could have been avoided, etc. defensive metrics are best-guesses based on statistical averaging, rather than counting statistics. The way offense is incorporated into WAR calculations is also "based on statistical averaging, rather than counting statistics." wOBA isn't calculated based on how many runs actually scored-- it's calculated based on how many runs an average hit/walk/out/etc. would count for. Similarly, it doesn't really take into account how much extra the pitcher had to work, etc. These are all objections to WAR, period, not defensive metrics. |
|
|
|
Post by ethanbein on Jun 23, 2015 18:55:19 GMT -5
|
|
|
|
Post by telson13 on Jun 23, 2015 19:02:51 GMT -5
I think the (valid) argument is that the ultimate outcome for offense (runs scored) is definite, finite, and closed-ended. For defensive non-plays, there's a butterfly effect: you're not measuring an actual outcome of runs prevented, but rather predicting runs prevented. A non-play on defense has no finite outcome attached to it, because you never really know how many runs it led to, how much extra the pitcher had to work to get out of the inning, how much it shortened his outing, how much of the poor bullpen performance could have been avoided, etc. defensive metrics are best-guesses based on statistical averaging, rather than counting statistics. The way offense is incorporated into WAR calculations is also "based on statistical averaging, rather than counting statistics." wOBA isn't calculated based on how many runs actually scored-- it's calculated based on how many runs an average hit/walk/out/etc. would count for. Similarly, it doesn't really take into account how much extra the pitcher had to work, etc. These are all objections to WAR, period, not defensive metrics. True, re: WAR. I guess I'm just commenting on the way people question the "value" of defensive metrics. But you're absolutely right, as far as WAR goes. I think it's simply easier for people to accept offensive WAR vs. defensive, because they can always look at and weigh their perception of the offensive stats. It's much tougher to eyeball defensive stats. Doesn't change the validity of either projection, just the ease with which someone evaluating them accepts their intrinsic value. |
|
|
