All Forums > SimLeague Baseball > MLB > Defensive Metrics
9/5/2013 12:46 PM
is that because the 2013 trout has no official numbers yet, for 2013? so his 2012 year is better, because it's official for the sample size? or u are still bent on getting people to agree with u, so that u can get the monies to acquire the bis data?

because nobody is agreeing with u that value is something other than money-related. how much do u need to acquire the numbers so that u can be absolutely sure that they have value?

sooner, before the season ends, and new numbers become official, negating yesterdays numbers in your evaluation of valuable. please. 
9/5/2013 12:49 PM
hurry. theres a classroom here of number geeks, and the bus to las vegas leaves tomorrow. help them beat the bookies, will u? student loan repayments are hanging in the balance.
9/5/2013 1:39 PM
Posted by bad_luck on 9/5/2013 12:20:00 PM (view original):
Posted by MikeT23 on 9/4/2013 8:14:00 PM (view original):
Posted by bad_luck on 9/4/2013 7:35:00 PM (view original):
Posted by MikeT23 on 9/4/2013 6:55:00 PM (view original):
Posted by bad_luck on 9/4/2013 6:01:00 PM (view original):
Are you asking why sample sizes play a large role in the reliability of a stat?
You know exactly what I'm asking.  Don't play dumb.

Per you:
1)  We need three seasons of fielding data to get a good "feel".
2)  But not with hitters because they get 700 PA while fielders get 150-200 chances.

So, if it's a numbers game, and you're saying it is, do we need three years of fielding data for all positions?    If so, why?   As I recall, some positions get a lot more opportunities to field the ball.    Is it 600 chances?   What's the number?
1) We need roughly three full seasons of UZR to reliably conclude that the number is a good indicator of the player's true talent. That three seasons isn't exact, it's just a guideline. 

2) When Taylor Teagarden posts a 1.200 OPS over 50 plate appearances, do you expect that to continue? Or do you consider it a fluke due to the sample size?

2b) Assuming that you consider it a fluke, would you stop using OPS because it doesn't reliably indicate a player's true talent when the sample size is small?


1)  So now it's "roughly" three seasons?    So maybe it could be two for some guys and four for others?   How do we differentiate who needs how many seasons?

2)  We've agreed, somewhat, that 700 PA can sort of tell us who can knock the **** out of the ball and who can't.   Why are you now using 50?

2b)  Please see 2.
1) I don't see why you need an exact number. For OPS to reflect true talent you need roughly 600 PA (and even then it should be regressed 30% if you have just that 600 PA). Batting average needs over 900 PA (or roughly one and a half years). Everything is roughly. There are no absolutes. And I'm fairly certain that if I tried to tell you there were, you'd **** yourself.

2) If a guy only has 50 or 200 or 400 plate appearances, you wouldn't take his OPS or batting average or whatever offensive stat as his true talent level. You'd regress it to the mean by the appropriate amount because your sample is too small. At 600 plate appearances, OPS is fairly reliable (you'd need to get to over 900 for batting average), but the year to year correlation is still well below 1. To get to a similar correlation with UZR, you need three years of data because defensive opportunities don't happen as often as plate appearances.

Do you really not get it or are you playing dumb?

1)  Aren't you the one trying to put a number on pretty much everything?   Why is this suddenly different?

2)  So it's not opportunities?    And it's not longer "roughly" three seasons?
9/5/2013 1:42 PM
Posted by MikeT23 on 9/5/2013 1:39:00 PM (view original):
Posted by bad_luck on 9/5/2013 12:20:00 PM (view original):
Posted by MikeT23 on 9/4/2013 8:14:00 PM (view original):
Posted by bad_luck on 9/4/2013 7:35:00 PM (view original):
Posted by MikeT23 on 9/4/2013 6:55:00 PM (view original):
Posted by bad_luck on 9/4/2013 6:01:00 PM (view original):
Are you asking why sample sizes play a large role in the reliability of a stat?
You know exactly what I'm asking.  Don't play dumb.

Per you:
1)  We need three seasons of fielding data to get a good "feel".
2)  But not with hitters because they get 700 PA while fielders get 150-200 chances.

So, if it's a numbers game, and you're saying it is, do we need three years of fielding data for all positions?    If so, why?   As I recall, some positions get a lot more opportunities to field the ball.    Is it 600 chances?   What's the number?
1) We need roughly three full seasons of UZR to reliably conclude that the number is a good indicator of the player's true talent. That three seasons isn't exact, it's just a guideline. 

2) When Taylor Teagarden posts a 1.200 OPS over 50 plate appearances, do you expect that to continue? Or do you consider it a fluke due to the sample size?

2b) Assuming that you consider it a fluke, would you stop using OPS because it doesn't reliably indicate a player's true talent when the sample size is small?


1)  So now it's "roughly" three seasons?    So maybe it could be two for some guys and four for others?   How do we differentiate who needs how many seasons?

2)  We've agreed, somewhat, that 700 PA can sort of tell us who can knock the **** out of the ball and who can't.   Why are you now using 50?

2b)  Please see 2.
1) I don't see why you need an exact number. For OPS to reflect true talent you need roughly 600 PA (and even then it should be regressed 30% if you have just that 600 PA). Batting average needs over 900 PA (or roughly one and a half years). Everything is roughly. There are no absolutes. And I'm fairly certain that if I tried to tell you there were, you'd **** yourself.

2) If a guy only has 50 or 200 or 400 plate appearances, you wouldn't take his OPS or batting average or whatever offensive stat as his true talent level. You'd regress it to the mean by the appropriate amount because your sample is too small. At 600 plate appearances, OPS is fairly reliable (you'd need to get to over 900 for batting average), but the year to year correlation is still well below 1. To get to a similar correlation with UZR, you need three years of data because defensive opportunities don't happen as often as plate appearances.

Do you really not get it or are you playing dumb?

1)  Aren't you the one trying to put a number on pretty much everything?   Why is this suddenly different?

2)  So it's not opportunities?    And it's not longer "roughly" three seasons?
1) What's different?

2) What?

I now see that you aren't playing dumb.
9/5/2013 1:54 PM
1)   Your intentions on putting one solid number on everything.     "We must measure his defensive value with a number!!!!"     Now, when asked for a number, you hem and haw like some hillbilly trying to explain his still.

2)  Is it opportunities(like PO/A/E) or three seasons or "roughly" three seasons?   How much data do we need for UZR not to be useless?
9/5/2013 2:13 PM (edited)
1) What number are you asking for?

2) One season of UZR is useful if you understand that future performance is likely to vary by a lot. Just like OPS after 1/3 of a season of plate appearances is useful if you understand that future performance is likely to vary by a lot.
9/5/2013 2:19 PM
Here's a problem I have with this -

When I see someone's OPS, let's say it's .900...even if it's just 2 months worth, I'll be confident in saying "he's playing well offensively.  He's on base 40% of the time, he's hitting some homers and doubles...he's playing well."

If I see a UZR of 8 after one season...I don't know if he played well, or if randomness and the fact that the coaches positioned him properly and if he's better at moving back on a fly ball rather than in and got more fly balls hit behind him, etc, etc came into play and really he played an average outfield that year.  Where I have much more confidence that the guy who had an OPS of .900, even after 50 games, was very productive for those 50 games.

Does that make sense?
9/5/2013 2:31 PM
Posted by burnsy483 on 9/5/2013 2:19:00 PM (view original):
Here's a problem I have with this -

When I see someone's OPS, let's say it's .900...even if it's just 2 months worth, I'll be confident in saying "he's playing well offensively.  He's on base 40% of the time, he's hitting some homers and doubles...he's playing well."

If I see a UZR of 8 after one season...I don't know if he played well, or if randomness and the fact that the coaches positioned him properly and if he's better at moving back on a fly ball rather than in and got more fly balls hit behind him, etc, etc came into play and really he played an average outfield that year.  Where I have much more confidence that the guy who had an OPS of .900, even after 50 games, was very productive for those 50 games.

Does that make sense?
That makes sense, but we're dealing with two different things - the talent level of the player vs the results of the game.

When you see an OPS of .900 after 2 months, you can say he played well offensively. But you can also doubt whether or not his performance, similar to the way you doubt UZR, was an aberration of the small sample. Did he get lucky on balls that dropped in? Is he batting 8th in the NL and walking a ton because the pitcher bats after him? Did he fluke into a triple or two? It wouldn't be hard for an average OPS player (.750) to get lucky enough to OPS .900 for a couple months.

A guy who puts up a UZR of 8 in one season still played well enough to get that 8, even if it was due to positioning or luck or catching deep fly balls instead of bloop singles. At the same time, his true defensive talent level may be much closer to average than great.
9/5/2013 2:41 PM
Posted by bad_luck on 9/5/2013 2:13:00 PM (view original):
1) What number are you asking for?

2) One season of UZR is useful if you understand that future performance is likely to vary by a lot. Just like OPS after 1/3 of a season of plate appearances is useful if you understand that future performance is likely to vary by a lot.
1)  When can we trust UZR to provide a valid number?

2)  How much is "a lot"?
9/5/2013 2:43 PM
Right, except I don't trust the results of UZR, where I do trust things like OBP.  OBP tells you how often a player got on base, how well he succeeded at his job of getting on base.

I would love a defensive model that can judge whether or not a player should have gotten an out based on where the fielder is standing.  If it can't do that now, then I also have questions about how accurately it judges things like the trajectory and velocity of a ball hit as well.  And I don't know if I can trust that stats that say "in this situation, it's an out 55% of the time" if the stat has difficulty with the idea of fielders standing in certain places, and not one universal place as a center fielder.
9/5/2013 2:46 PM
As I said many moons ago, offensive stats aren't subjective.   You got a hit or you didn't, you're on base or you're not.    Not so with the new defensive metrics.
9/5/2013 2:52 PM
UZR is sort of the equivalent of an offensive stat that judges what should have happened based on what you did while batting, and not what actually happened after you hit the ball.  So "A ball that comes off the bat at 100 MPH at a trajectory of 75 degrees off the ground is a home run 78% of the time, so your "adjusted slugging" for that atbat is 3.12."  Or "The ground ball hit at 76 MPH directly between third base and second base is a hit 13% of the time, so you are rewarded with .13 hits that atbat."  A) I wouldn't trust that stat without more information on how "they" come to these numbers, and B) I don't know how interested I'd be in those results anyway.  It's a results-driven game.
9/5/2013 2:53 PM
Posted by burnsy483 on 9/5/2013 2:43:00 PM (view original):
Right, except I don't trust the results of UZR, where I do trust things like OBP.  OBP tells you how often a player got on base, how well he succeeded at his job of getting on base.

I would love a defensive model that can judge whether or not a player should have gotten an out based on where the fielder is standing.  If it can't do that now, then I also have questions about how accurately it judges things like the trajectory and velocity of a ball hit as well.  And I don't know if I can trust that stats that say "in this situation, it's an out 55% of the time" if the stat has difficulty with the idea of fielders standing in certain places, and not one universal place as a center fielder.
True. OBP is an actual record of a plate appearance. It isn't subject to any sort of judgement. But in a small sample, it's still fluky. It doesn't tell you how good he is at getting on base. You need roughly 1 full season of OBP to see that.

I understand not trusting UZR. But let's say you had video of every defensive play of every player along with HITfx data that showed the speed and trajectory of every ball hit, don't you think you could sit down and evaluate a player's defense fairly well?
9/5/2013 3:05 PM (edited)
Posted by bad_luck on 9/5/2013 2:53:00 PM (view original):
Posted by burnsy483 on 9/5/2013 2:43:00 PM (view original):
Right, except I don't trust the results of UZR, where I do trust things like OBP.  OBP tells you how often a player got on base, how well he succeeded at his job of getting on base.

I would love a defensive model that can judge whether or not a player should have gotten an out based on where the fielder is standing.  If it can't do that now, then I also have questions about how accurately it judges things like the trajectory and velocity of a ball hit as well.  And I don't know if I can trust that stats that say "in this situation, it's an out 55% of the time" if the stat has difficulty with the idea of fielders standing in certain places, and not one universal place as a center fielder.
True. OBP is an actual record of a plate appearance. It isn't subject to any sort of judgement. But in a small sample, it's still fluky. It doesn't tell you how good he is at getting on base. You need roughly 1 full season of OBP to see that.

I understand not trusting UZR. But let's say you had video of every defensive play of every player along with HITfx data that showed the speed and trajectory of every ball hit, don't you think you could sit down and evaluate a player's defense fairly well?
OK so here's the thing.  The guy with the OPS of 1 for a year had a fantastic offensive year.  He'd be in the MVP discussion.  Now, if you say "but his BABIP is .398 and 9 of his homers were only out at Yankee Stadium and he's lucky, etc" that's fine, but he still put up the numbers and contributed to the team.  

With UZR, you're getting a number that essentially tells you how well a player did based on what others have done in the past, based on where a ball is hit, and doesn't take into consideration positioning of the fielder.  It's "that's a bad play, he really should have caught that ball." It's subjective and I'd like to see more data and learn more about how this data is generated before I trust it as much as you and many others do.  
9/5/2013 3:15 PM (edited)
Posted by burnsy483 on 9/5/2013 2:43:00 PM (view original):
Right, except I don't trust the results of UZR, where I do trust things like OBP.  OBP tells you how often a player got on base, how well he succeeded at his job of getting on base.

I would love a defensive model that can judge whether or not a player should have gotten an out based on where the fielder is standing.  If it can't do that now, then I also have questions about how accurately it judges things like the trajectory and velocity of a ball hit as well.  And I don't know if I can trust that stats that say "in this situation, it's an out 55% of the time" if the stat has difficulty with the idea of fielders standing in certain places, and not one universal place as a center fielder.
the bat is the defensive model of defensive metrics. the 0-7's are explained very well thru many pages of this thread.

however, i see your point about on-base percentage. if getting on-base is of importance, then the highest percentage of an event like that is by batting safely. ichiro suzuki bats safely at a higher percentage than the walks that also allow him to get on-base. on the flip side, barry bonds same on-base percentage should reflect that he is on-base more often by walks, instead of batting safely.

the overwhelming cause of the defensive switching, which loaded the first base side of the infield, was not done so to keep barry bonds from walking to first, even though they added extra defense. they actually had barry bonds in a situation where the pitcher was going to walk barry bonds anyways. why the defensive switch even after the decision to walk him, by not giving him anything to hit?

point is, what will the bis data numbers tell everyone here, when applied to defend against a batter, with certain known tendencies, when that batter is preferred to walk in the first place? will the numbers show-up as again meaning defensive metrics must always give us a batter equivalent, basically? basically, as it says on page one of this wonderfully valuable thread, that gives offensive equivalents to defensive metrics?

we already know hgh and stem cells will also be incorporated into your findings, but that is for another thread, once this one closes.

so, tell us what it costs to get the proprietary numbers, and lets party as we close this thread with a climactic finale. i'm game for that.
of 23
All Forums > SimLeague Baseball > MLB > Defensive Metrics

Terms of Use Customer Support Privacy Statement

Popular on WhatIfSports site: Baseball Simulation | College Basketball Game | College Football Game | Online Baseball Game | Hockey Simulation | NFL Picks | College Football Picks | Sports Games

© 1999-2014 WhatIfSports.com, Inc. All rights reserved. WhatIfSports is a trademark of WhatIfSports.com, Inc. SimLeague, SimMatchup and iSimNow are trademarks or registered trademarks of Electronic Arts, Inc. Used under license. The names of actual companies and products mentioned herein may be the trademarks of their respective owners.