Jump to content


Yo! You're not logged in. Why am I seeing this ad?

Photo

Youkilis or Teixeira


  • This topic is locked This topic is locked
27 replies to this topic

#1 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4,157 posts

Posted 01 July 2009 - 02:05 PM

The sport editor from the Harford Courant is looking for your opinions on who should start at first base for the American League in the 2009 All Star Game. He would like to get feedback today, so if you can get your post in here by 4:30 or 5 it would really help him out.

What say you SoSH? Youkilis or Teixeira? Or someone else?

#2 LahoudOrBillyC


  • Indian name is Massages Ellsbury


  • 3,760 posts

Posted 01 July 2009 - 02:16 PM

Kevin Youkilis had a better year in 2008, and he is having a better year in 2009. He missed two weeks, which is really Teixeira's only argument. I think Youks is the guy.


#3 rembrat


  • SoSH Member


  • 14,124 posts

Posted 01 July 2009 - 02:21 PM

It's Kevin Youkilis and it's not even close.

As a 1B:
Youkilis -.328/.461/.603
Tex - .264/.375/.547

And since that isn't really fair to Teix because of the PA difference (Youk has played a bit at 3B giving him les PA at 1B) here are Youkilis overall slash stats .317/.443/.588. Still a ton better than Teixeira's.

wOBA also gives the edge to Youkilis:

Youk - .440
Tex - .402

The people that give Teixeira the edge are the people that put a lot of stock in HR and RsBI numbers.

#4 Alternate34

  • 2,461 posts

Posted 01 July 2009 - 02:22 PM

QUOTE (Frisbetarian @ Jul 1 2009, 02:05 PM) <{POST_SNAPBACK}>
The sport editor from the Harford Courant is looking for your opinions on who should start at first base for the American League in the 2009 All Star Game. He would like to get feedback today, so if you can get your post in here by 4:30 or 5 it would really help him out.

What say you SoSH? Youkilis or Teixeira? Or someone else?


Youkilis

Youkilis - .443 OBP, .588 SLG
Teixeira - .384 OBP, .568 SLG

Youkilis also has a higher batting average. Teixeira's power numbers (especially HRs) are inflated by playing in the New Yankee stadium, though Youkilis does benefit from Fenway.

I am receptive to arguments about playing time, but it's not like Youkilis is injury prone. Defense, they are at least even, though I think Youkilis is a bit better. The last half of the previous season, they both did well.

#5 absintheofmalaise


  • too many flowers


  • 8,710 posts

Posted 01 July 2009 - 02:25 PM

After looking at a variety of numbers, BtRuns, OPS, OPS+, RC/G and a few others, I'm going with Youkilis. His OBP is the key for me. lgOBP is .347. His is .443 while Teixeira's is .384 while the lgOBP for him is .337.

#6 Wade Boggs Hair

  • 1,366 posts

Posted 01 July 2009 - 02:26 PM

I would go with Youkilis. He has better rate stats across the board (.317/.443/.588 vs. .275/.384/.568). From BP, Youk leads Teixeira in VORP (30 to 21.2) despite 60 fewer PAs. And while FanGraphs gives Teixeira a razor's edge in WPA (2.25 to 2.23), it gives Youkilis a big edge on defense (+10.7 UZR/150 to -3.1 for Teixeira). Teixeira also has a slim lead in WARP3 (6.2 vs. 6.1), but I think Youkilis' ability to match Teixeira in the advanced metrics despite time on the DL mitigates those narrow differences.

Apart from the numbers, I think Youkilis' ability to play third has added significant value to his team, given the recent injury to Mike Lowell and the need to play David Ortiz at 1B during road interleague games.

EDIT - beaten to the punch on some of the numbers.

Edited by Wade Boggs Hair, 01 July 2009 - 02:28 PM.


#7 PrometheusWakefield


  • SoSH Member


  • 5,351 posts

Posted 01 July 2009 - 02:36 PM

Lets not forget that Teixeira has moved from the best offensive park in the American League to... er... the new best offensive park in the American League. Teixeira stats show the impact:

Teixeira home .397 OBP .669 SLG 1.067 OPS
Teixeira away .373 OBP .472 SLG .845 OPS

Youkilis' stats on the road are basically equal: 1.070 at home, .998 on the road.









#8 LahoudOrBillyC


  • Indian name is Massages Ellsbury


  • 3,760 posts

Posted 01 July 2009 - 02:44 PM

QUOTE (PrometheusWakefield @ Jul 1 2009, 12:36 PM) <{POST_SNAPBACK}>
Lets not forget that Teixeira has moved from the best offensive park in the American League to... er... the new best offensive park in the American League. Teixeira stats show the impact:

Teixeira home .397 OBP .669 SLG 1.067 OPS
Teixeira away .373 OBP .472 SLG .845 OPS

Youkilis' stats on the road are basically equal: 1.070 at home, .998 on the road.


Yes, this is sort of the hammer to the argument. People looking at Teixeira's OPS+ have to realize that the park numbers BR is using are from the old stadium, not the new one. When they switch (after the season?) MT's adjusted numbers will plummet.


#9 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4,157 posts

Posted 01 July 2009 - 03:14 PM

I think an argument could be made that Teixeira is only the 4th or 5th best 1st baseman in the AL this year. Branyan, Cabrera, and Morneau are all having great seasons, and Carlos Pena and Victor Martinez (if you count him as a first baseman - 39 games there vs 38 as a catcher) don't suck either.

#10 kanga12

  • 1,400 posts

Posted 01 July 2009 - 03:17 PM

QUOTE (Frisbetarian @ Jul 1 2009, 12:05 PM) <{POST_SNAPBACK}>
What say you SoSH? Youkilis or Teixeira? Or someone else?
Not to hijack this thread, but it does bear mentioning that Teixeira's numbers aren't that much better than some of his fellow AL first basemen -- and in some cases, worse:

Teixeira (NYY): .275/.384/.568 (20 HR, 60 RBI), 146 OPS+
Branyan (SEA): .303/.400/.606 (19 HR, 40 RBI), 164 OPS+
Morneau (MIN): .313/.394/.582 (19 HR, 64 RBI), 160 OPS+
Cabrera (DET): .332/.397/.557 (16 HR, 47 RBI), 146 OPS+

Edit: Fris beat me to the punch.
2nd Edit: Corrected typos

Edited by kanga12, 01 July 2009 - 05:43 PM.


#11 joyofsox


  • empty, bleak


  • 5,303 posts

Posted 01 July 2009 - 04:52 PM

Offensive Winning Pct.

Youkilis: .776

Teixeira: .695


#12 kiransdad

  • 89 posts

Posted 01 July 2009 - 04:54 PM

Between Youkilis and Morneau; Youk has the edge in UZR (Morneau and Teixeira are comparable at -2) as well as in batting (over Morneau) so I'd go with him.

#13 yecul


  • appreciates irony very much


  • 13,845 posts

Posted 01 July 2009 - 05:14 PM

The argument for Tex would either have to be centered on his one awesome month, a total focus on HR and RBI, or to simply ignore baseball to be cynical and say how New York has a larger population to vote than Youkilis.

Others have illustrated why Youkilis *should* win, but that isn't always the final outcome.

#14 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4,157 posts

Posted 01 July 2009 - 05:15 PM

QUOTE (kiransdad @ Jul 1 2009, 04:54 PM) <{POST_SNAPBACK}>
Between Youkilis and Morneau; Youk has the edge in UZR (Morneau and Teixeira are comparable at -2) as well as in batting (over Morneau) so I'd go with him.


As always, I feel compelled to remind folks that it really takes 3 years of data before even the more complex defensive evaluation systems are close to accurate (especially for 1st basemen), and that partial year numbers can be quite misleading. When you have multiple years of data for a player it is always better to look at a player's history than his partial year. I'm not meaning to pick on you, kirins, and am fairly certain UZR's creator would tell you the same thing.

Youkilis is a very good defensive 1st baseman, Teixeira is a great one. But he's not good enough to make up for the offensive difference between the two this season, and Youkilis is the clear choice for 2009 AL All Star Team at 1st base.

#15 Beomoose


  • SoSH Member


  • 5,894 posts

Posted 01 July 2009 - 06:48 PM

.317/.443/.588? Youk. All day every day.

#16 OCD SS


  • SoSH Member


  • 6,731 posts

Posted 02 July 2009 - 01:51 PM

QUOTE (Frisbetarian @ Jul 1 2009, 06:15 PM) <{POST_SNAPBACK}>
As always, I feel compelled to remind folks that it really takes 3 years of data before even the more complex defensive evaluation systems are close to accurate (especially for 1st basemen), and that partial year numbers can be quite misleading. When you have multiple years of data for a player it is always better to look at a player's history than his partial year.


In this case though, I think you can use the small sample size numbers. We're looking at what a player has done in this half season (depending on your own All-Star criteria), not his true talent level (which is what you'd have with the three year sample).

I'd also be curious to know how +/- compares the two. IIRC the two systems can diverge a little around 1B.

#17 koufax32


  • He'll cry if he wants to...


  • 3,056 posts

Posted 02 July 2009 - 02:52 PM

Youkilis. For all the statistical reasons mentioned thus far. To pile on though, Teixeira's OPS+ is 147. Youkilis' is 160.

From the persective of who would help the AL win (since "this one counts")? Youkilis, by a complete landslide. An all-star game pits the best hitters against the best pitchers. Most "best pitchers" are power arms who have the ability to get themselves out of trouble with a strikeout at any given point of the game.

Youkilis vs. power pitchers
OPS = 1.248
OPS+ = 248

Teixeira vs. power pitchers
OPS = .871
OPS+ = 143

As others have posted, the question shouldn't be "which of these two should start?". It should be "does Teixeira deserve to even be there?"

EDIT: corrected number

Edited by koufax32, 02 July 2009 - 02:55 PM.


#18 mt8thsw9th


  • anti-SoSHal


  • 12,887 posts

Posted 02 July 2009 - 02:57 PM

QUOTE (koufax32 @ Jul 2 2009, 03:52 PM) <{POST_SNAPBACK}>
As others have posted, the question shouldn't be "which of these two should start?". It should be "does Teixeira deserve to even be there?"


Of course he deserves to be there, he isn't having a poor season a la Jason Giambi in 2004. The question is "should Teixeira be there, potentially at the expense of Youkilis?" and the answer is a resounding "no".

#19 koufax32


  • He'll cry if he wants to...


  • 3,056 posts

Posted 02 July 2009 - 03:02 PM

QUOTE (mt8thsw9th @ Jul 2 2009, 03:57 PM) <{POST_SNAPBACK}>
Of course he deserves to be there, he isn't having a poor season a la Jason Giambi in 2004. The question is "should Teixeira be there, potentially at the expense of Youkilis?" and the answer is a resounding "no".


Out of Youkilis, Morneau, and Cabrera who would Teixeira replace? I'm assuming three 1B on the roster.

FWIW I think Teixeira should be there. I just think that it would be absurd for anyone to say he was robbed if these three were named and he wasn't.

Edited by koufax32, 02 July 2009 - 03:04 PM.


#20 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4,157 posts

Posted 02 July 2009 - 03:03 PM

QUOTE (OCD SS @ Jul 2 2009, 02:51 PM) <{POST_SNAPBACK}>
In this case though, I think you can use the small sample size numbers. We're looking at what a player has done in this half season (depending on your own All-Star criteria), not his true talent level (which is what you'd have with the three year sample).

I'd also be curious to know how +/- compares the two. IIRC the two systems can diverge a little around 1B.



With defense, the sample size is quite small and the data used to calculate the assessments can be bad. This can cause partial season numbers to be quite skewed and means they do not necessarily reflect how the player has performed. I had this discussion with Mitchel Lichtman, UZR's creator, last summer, and he was quite clear that a larger sample size should always be used in lieu of a partial season.

A quick look through Inside the Book dot com's forums showed a recent conversation where Mitchel said:
QUOTE
I really hate when anyone takes a small sample within a large sample (like a player’s career or the last few years) and tried to make “sense” of it. There is absolutely no reason to be using 60 games of UZR for anything when you have 400 or 500 games. To say that someone is playing better or worse defense because the numbers are better or worse is not right. We don’t know that a player is playing bad defense because his UZR is minus. That is the whole point of using as much data as possible.


edit to add that I checked Teixeira's plus minus and Dewan, who we will be doing a chat with us next week, has him at +2 enhanced.

#21 PedroKsBambino


  • SoSH Member


  • 12,776 posts

Posted 02 July 2009 - 06:28 PM

QUOTE (Frisbetarian @ Jul 2 2009, 04:03 PM) <{POST_SNAPBACK}>
With defense, the sample size is quite small and the data used to calculate the assessments can be bad. This can cause partial season numbers to be quite skewed and means they do not necessarily reflect how the player has performed. I had this discussion with Mitchel Lichtman, UZR's creator, last summer, and he was quite clear that a larger sample size should always be used in lieu of a partial season.

A quick look through Inside the Book dot com's forums showed a recent conversation where Mitchel said:

edit to add that I checked Teixeira's plus minus and Dewan, who we will be doing a chat with us next week, has him at +2 enhanced.


MGL's comments are about true talent level, though, not about short-term performance...so OCD SS's point is still relevant.

I don't think there's any question that more data is better, when you're trying to evaluate a player. The question here is when you're trying to value performance only during a set period of time, is it better to take a larger set that (implicitly) assumes the subset of games is a perfect mirror of the larger set, or do you take what you can on the subset and decide whether it's real or not? Not sure I'd do the latter, but there's certainly a case for it in this situation.

#22 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4,157 posts

Posted 02 July 2009 - 07:35 PM

QUOTE (PedroKsBambino @ Jul 2 2009, 07:28 PM) <{POST_SNAPBACK}>
MGL's comments are about true talent level, though, not about short-term performance...so OCD SS's point is still relevant.

I don't think there's any question that more data is better, when you're trying to evaluate a player. The question here is when you're trying to value performance only during a set period of time, is it better to take a larger set that (implicitly) assumes the subset of games is a perfect mirror of the larger set, or do you take what you can on the subset and decide whether it's real or not? Not sure I'd do the latter, but there's certainly a case for it in this situation.



Your first sentence is not correct. MGL is saying that short term defensive stats should not be counted on to assess performance. To wit, "We don’t know that a player is playing bad defense because his UZR is minus. That is the whole point of using as much data as possible."

Also, from the same thread I linked above:

QUOTE
For hitting, we are actually measuring a player’s contribution to runs and wins, at least to some extent. UZR does no such thing. Just because a player has a +5 UZR does NOT mean that he contributed to saving any runs. Because of measurement error, all of those “+5 runs” could have been the easiest plays in the world. In fact, if you had watched the player on defense, he might have contributed to lots of runs for his opponents even though he had a +5 UZR. Completely different thing, UZR, from an offensive stat which reflects what “actually happened” in a runs/wins sense.


and:


QUOTE
Yes, for example, a single is a single. It describes an event perfectly. An RBI is an RBI. It is relatively easy to use that type of data to assign “MVP” value (although certainly what constitutes “valuable” is debatable).

No such analogy with UZR. It describes an amorphous series of events almost like a wave or a probability distribution for the location of a subatomic particle, or something like that.


Finally, would anyone watching Youkilis and Teixeira play 1st base through the first half of 2009 actually say they think Youkilis is worth almost 15 more runs defensively than Teixeira? Does that mean he could be 30 runs better through the full season? Seriously? That's what the partial season UZR numbers say.

#23 PedroKsBambino


  • SoSH Member


  • 12,776 posts

Posted 02 July 2009 - 08:02 PM

QUOTE
Your first sentence is not correct. MGL is saying that short term defensive stats should not be counted on to assess performance. To wit, "We don’t know that a player is playing bad defense because his UZR is minus. That is the whole point of using as much data as possible."


I don't think so. I believe I know exactly what you and he are saying and my statement is not incorrect. I think you may be misinterpreting how what MGL is saying applies in this particular context.

Now, there's no question that you want a larger dataset if possible. That's because we want to assess true talent level, not short-term fluctuation, and also because the measurement error will even out over a larger dataset.

But to answer the specific question here (what is the performance in a defined short period) our choice is as I framed it above: assume that the larger sample is also an accurate description of the smaller one, or take our chances on the smaller one being an accurate description of performance. The discussion of using the smaller dataset needs to include the reality that there's a significant error rate to that measurement (which is what MGL describes the +5 runs as potential measurement error) but I don't believe it's 'wrong' to use UZR here so much as it is even riskier than usually using defensive stats.

#24 Frisbetarian


  • ♫ ♫ ♫ ♫ ♫ ♫


  • 4,157 posts

Posted 02 July 2009 - 08:49 PM

QUOTE (PedroKsBambino @ Jul 2 2009, 09:02 PM) <{POST_SNAPBACK}>
I don't think so. I believe I know exactly what you and he are saying and my statement is not incorrect. I think you may be misinterpreting how what MGL is saying applies in this particular context.

Now, there's no question that you want a larger dataset if possible. That's because we want to assess true talent level, not short-term fluctuation, and also because the measurement error will even out over a larger dataset.

But to answer the specific question here (what is the performance in a defined short period) our choice is as I framed it above: assume that the larger sample is also an accurate description of the smaller one, or take our chances on the smaller one being an accurate description of performance. The discussion of using the smaller dataset needs to include the reality that there's a significant error rate to that measurement (which is what MGL describes the +5 runs as potential measurement error) but I don't believe it's 'wrong' to use UZR here so much as it is even riskier than usually using defensive stats.


I suppose we should just agree to disagree here, but really don't think we are far apart in our thinking on this and appreciate the repartee. The thread I found and took those quotes from was actually about Teixeira's 2009 UZR being discussed in the NJ Star Ledger, and MGL didn't seem pleased with the partial season numbers being used to show he was having a poor defensive season. One more quote:

QUOTE
Say a player has a UZR per 150 of +15 over the last 5 years - IOW, most likely an excellent defender. For the last 10 games, he has a UZR of -1 (-15 per 150). What are the chances that he actually played below average defense over those 10 games? How about 50 games? 100 games?


I have always been skeptical of partial season defensive stats of any kind, but want to make it very clear that UZR is an incredible, ingenious metric. Still, with an average 1st baseman making only about 250 plays per 160 games, and just over 90% of those being routine plays, there is not a lot of data even in a full season with which one can assess a player's defensive acumen. When you also consider the issues with the data, which include inconsistent identification of speed of batted ball/batted ball type and fielder positioning, then the potential for large discrepancies in partial season (or even single season) defensive metrics is vast. In this particular case, the chances of Teixeira being 15+ runs worse in 2009 than 2008 per UZR is unlikely, and treating these numbers as reflective of true ability level is a mistake, imo.

#25 Jnai


  • is not worried about sex with goats


  • 8,143 posts

Posted 02 July 2009 - 09:18 PM

Suppose you two took a different example using a different metric.
Josh Beckett goes out and, in his first start of the season, gives up 5 runs in one inning and gets pulled in the third.

PedroK's says: Josh Beckett was terrible in that game.
Fris says: Yeah, but he's unlikely to be terrible going forward, that might just might be bad luck.

Both of those things might be true.

Now, consider UZR.

Teixeira goes out and, in his first 80 games, plays -10 defense.

PedroK's says: Teix played very poor defense over the course of the season.
Fris says: Yeah, but he's unlikely to be that bad going forward.


But, although the arguments look similar, I assume that Fris says he's unlikely to be terrible for two different reasons.
In Example 1, we can use Josh Beckett's previous history to know that Beckett is likely to be good based on his previous career totals.
In Example 2, we can also use Mark Teixeira's previous history to know that he's likely to be good based on his career totals. But there's also an added factor: measurement error. ERA measures something unmistakably true: the run scored and it was attributable to the pitcher. No one ever miscounts ERA or wonders if the run really scored. UZR relies on the fact that a certain number of balls are hit into a particular zone and makes assumptions about the difficulty of plays, and so accurately measuring the value of UZR requires a large sample size.

The point I think is important to make is not only that Teix's defense is likely to be good going forward (which, based on a large sample size is true), but that UZR has measurement error which cannot be resolved in such a small sample. When that measurement error takes ~3 seasons to resolve, .5 seasons is simply not enough to get any sort of accurate measurement. It doesn't have to do with the fact the defense played during that time was good or bad or in line with career norms or out of line with them, it's simply that the statistic is not sensitive within that particular sample size.

So, the point is that UZR in such a small sample size is virtually meaningless. Not that it is a poor measure of Teixeira's real value going forward. Half a season of UZR is likely meaningless* for anyone, in any circumstance. What you are reading as "short term fluctuation" is probably just "noise".



*Edit: "Meaningless" is probably too strong. But, closer to meaningless than meaningful.

Edited by Jnai, 02 July 2009 - 09:22 PM.


#26 PedroKsBambino


  • SoSH Member


  • 12,776 posts

Posted 02 July 2009 - 10:25 PM

As noted before, there's clearly a difference between looking back and looking forward. There is no question that when you're looking forward you want to use something towards true talent level and thus, you really clearly want the larger sample.

However, I'm not sure that's quite the same when asking the retrospective question. To follow jnai's hypo, if I were to say that Teix was -10 for 80 games that's his defensive value thus far, Fris would have two possible responses: he could say "Teix was really still a +5 defender for those 80 games, because that's what he's averaged the prior three years" or he could say "we don't know what Teix was, the sample was too small"

I'd suggest the second of those is not useful for the specific question posed. The first is interesting, but I'm not sure it really is more accurate than the admittedly-problematic 80 game sample for the specific question of valuing 80 games of performance. Why? Because it literally throws out the 80 games. Or, it includes them as a 1/6 or so contributor, which seems wrong. We could do other things--a regression, weighted average, and so forth. Still measurement issues there, though, seems to me. This, I think, really gets at the very specific question I was asking about retrospective analysis (because I am, and have been, in agreement on prospective projection with both of you): is it really the case for a question about value over 80 past games to say that we use the three+ year old info? Not sold on that.

Jnai suggests that the half-season of UZR is essentially worthless; I think that's farther than I'd go but, there's certainly a significant error rate there due to the sample size. I hear the point on measurement, too, but I think that's a bit aggressive a description (more on that below). So, we have a problematic 80 game sample, and we then have to compare that to this: the chances that the prior three years of data on either guy are an accurate description of the first half of this year.

I can certainly see why someone would say the latter is a better set of data to be working from; generally speaking performance is reasonably linear and thus, year four won't vary all that much from 1-3, etc. There's an assumption there that fielding change is very slow and I think that's usually true, but not necessarily always the case. One thing that might impact it is when someone changes teams/parks, which impacts (among other things) how they might position. Though, less an issue at 1B. That leads me here: when we see a big change in UZR for a half-season it may well be that it's random variance, but in the approach being suggested we can't ever reach the conclusion that it is a real change and I think that's a concern.

In addition, and more broadly, I also think we should consider how what we're saying about a half-season of data here says about the general utility of UZR. My take on MGL's comments on this over, what, five or six years now that it's been out there is that he's very protective of the metric and tends to go very far towards only 'clean' data being discussed. This is why he's very insistent on the sufficient sample size. On the one hand I respect this; on the other, I think we do hit an issue where the relevance of the info is impacted by the desire for such a large sample size. For example, age becomes a significant additional variable when you're talking about three years of data and then a player's defensive value in year four (which is what we'd be doing here). Oftentimes, team/manager is a variable as well. So, while I understand from an event-perspective we want that much data I've always been concerned with that requirement's statement about the metric overall and the ability to recognize and value larger changes.

Point being, I tend to think we overstate somewhat how useful UZR is to begin with (because of the measurement error and aging and park factor issues) and thus I also think we are also overstating how much less useful it is in a half-season sample. Obviously, the small sample is going to be less useful than the large one, still. I am not as deep in the proofs of UZR over the last two or so years as you guys are, but it doesn't ring true to me that we should be discounting the half-season almost wholly, either.

#27 Bellhorn


  • Lumiere


  • 1,520 posts

Posted 03 July 2009 - 08:57 AM

QUOTE (Jnai @ Jul 2 2009, 10:18 PM) <{POST_SNAPBACK}>
Suppose you two took a different example using a different metric.
Josh Beckett goes out and, in his first start of the season, gives up 5 runs in one inning and gets pulled in the third.

PedroK's says: Josh Beckett was terrible in that game.
Fris says: Yeah, but he's unlikely to be terrible going forward, that might just might be bad luck.

Both of those things might be true.

Now, consider UZR.

Teixeira goes out and, in his first 80 games, plays -10 defense.

PedroK's says: Teix played very poor defense over the course of the season.
Fris says: Yeah, but he's unlikely to be that bad going forward.


But, although the arguments look similar, I assume that Fris says he's unlikely to be terrible for two different reasons.
In Example 1, we can use Josh Beckett's previous history to know that Beckett is likely to be good based on his previous career totals.
In Example 2, we can also use Mark Teixeira's previous history to know that he's likely to be good based on his career totals. But there's also an added factor: measurement error. ERA measures something unmistakably true: the run scored and it was attributable to the pitcher. No one ever miscounts ERA or wonders if the run really scored. UZR relies on the fact that a certain number of balls are hit into a particular zone and makes assumptions about the difficulty of plays, and so accurately measuring the value of UZR requires a large sample size.

Strongly disagree with the bolded portion - the last six words in particular. I know I've made my case on this issue before, ad nauseam, but perhaps the insight concerning the fallacy of using small-sample UZR to measure past fielding performance will help clarify my analogous position on retrospective statistical analysis in general.

The similarity between the two cases should be obvious, if you think about it. What reason do we have take as the de facto assumption that an (earned) run that scores must be attributable to the pitcher, other than that the box score accounting system encourages us to? A pitcher's performance, properly understood, consists of observed phenomena that result directly (with minimal interference from other agents or general randomness) from his own actions - the speed, movement, location, etc. of his pitches being the primary example. (You of all people should be on board with this!) Whether or not a given sample of such performance in a game results in a run scoring depends on the influence of many external factors: the efforts of the opposing hitters and the efforts of the pitcher's fielding teammates being only the most obvious. An omniscient analysis of past pitching performance might attribute an earned run entirely to the pitcher in one extreme case (he grooves a BP fastball which is hit out of the park) or hardly at all at the other end (he makes excellent pitches that happen to result in seeing-eye ground balls.)

If you take a large enough sample of game results, you can more or less assume that these extreme positions on the spectrum balance each other out, and that if Pitcher A and Pitcher B have both given up, say, 100 earned runs in 200 innings, they probably performed at about the same level over this span. (Even here, the possibility of error is high enough that one should be uncomfortable making this assumption.) But the point is that, as with UZR, there isn't a perfectly reliable correspondence between the actual physical and mental performance of the player and the game result that is tracked by the stat in question. So the "past result = past performance" notion that is, unfortunately, taken as axiomatic by so many people when it comes to box-score stats is in fact thoroughly unjustified.

You could summarize the issue like this:

1) You need a certain sample size of a game result stat in order to accurately assess a player's (past) performance.
2) You need a certain sample size of past performance (i.e. a larger sample size of game result stat) in order to accurately assess a player's talent level, which can be used to predict future performance.

The requisite sample size is smaller in 1), but it can't be assumed to be trivially small unless there's statistical evidence to suggest that it is. And of course it can vary between stats - it's entirely possible that ERA becomes an accurate measurement of past pitching performance more quickly than UZR becomes an accurate measurement of past fielding performance. But it's a difference in degree, not in kind.

#28 yep

  • 941 posts

Posted 05 July 2009 - 11:18 PM

Agreeing with Bellhorn--+

UZR echoes the problems of ERA (in the determination of which runs are "earned"). The notion of ERA as a "pure" metric frankly baffles me and I suspect might have been a typo or something on the part of Jnai.

Expanding on Bellhorn--

Everything that makes UZR a useful metric (to whatever degree it is one) is also what makes it intrinsically susceptible to SSS. There are easy plays and difficult plays and in-between plays and plays that should have been made by another fielder and plays that would never have been attempted by lesser fielders and so on. UZR actually compounds the problem of simply counting errors in small sample sizes. To whatever degree it works, it really needs a large arc.

To answer the original question, the fact that it's even close makes me happy. Tex is an outstanding player who may have contributed a few wins to this team, but 1B is a position that I would be extremely slow to lock up in a long-term, big money contract. Even for an all-star gold glove 1B. Not only because a declining 1B is basically a DH at best, but also because 1B is an easy place to park declining or semi-injured sluggers. A gimpy Bay or Drew or whoever might not be as good as Teixeira, but they are a hell of a lot more valuable at 1B than they are on the bench or the DL. Tex at 1B means Tex 1B always, and whatever replacement-level jobber you can dig up to fill in for the guy who might have otherwise spelled 1B. I.e. the value of Teixeira or any other premium 1B is the value of (Tex-(players who get injured but who could play 1B)+(whatever jobber fills in for injured player above)).

Youkilis' flexibility to play competent 3B, plus his vastly lower contract costs and obligations make the Red Sox a more nimble and competitive team, with a comperable 1B to Teixeira. So setting aside the all-star game, and factoring in the real-world contract and positional flexibility, I'll take Youkilis.

Edited by yep, 05 July 2009 - 11:23 PM.





1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users