Pitchers, MVP, Quality of opposing hitters (September 19, 2003)
BP's Rany brought up an interesting point on the quality of Loiza's opposing hitters. I estimate that the difference between Loiza's opponents and Halladay's opponents is worth about 0.30 ERA. This is a huge total, working out to about 1 marginal win for the season. Seeing how close Halladay and Loiza are, this advantage should not be overlooked.
In fact, in addition to "park adjustments" and "fielder adjustments", I would also include "opposition adjustments" to any metric. The first 2 are by far the biggest influence, but the third one is valuable in the case of Loiza and Halladay in particular.
(I know, I haven't figured out Loiza's spelling yet.)
As for people thinking about MVP and pitchers, it's really rather simple. You compare your pitcher to his peers or to the schlubs, and the impact that the top pitcher has is close to the impact that the top non-pitcher has. Forget about the 30 games v 150 games debate. The pitcher faces 1000 batters. The batter comes to the plate 600 times, and makes another 300 plays on the field.
If the pitcher is responsible for an effective 700 PA (with the other 300 going to the fielders), the batter is responsible for 750 PA (with half the 300 going to the pitcher). Certainly for a random PA, the batter has more influence than the pitcher, on average, relative to their peers. But this doesn't necessarily translate at the extremes.
--posted by TangoTiger at 03:02 PM EDT
Posted 3:17 p.m.,
September 19, 2003
(#1) -
zoobird
I think that over the course of one season, pitchers would be much more affected by quality of opposition than batters. While a batter's 600 plate appearances are typically against 162 different pitchers (ok, I'm drastically oversimplifying), a pitchers opposition plate appearances are only against 35 (or so) different lineups. So there's going to a lot more correlation among the batters a pitcher faces in one season than among the pitchers a batter faces.
Posted 3:33 p.m.,
September 19, 2003
(#2) -
tangotiger
Here's data to support your conclusion:
***
Looking at all hitters with at least 400 PA, the standard deviation of their opposition pitcher's OPS is .009.
Looking at all pitchers with at least 400 PA, the standard deviation of their opposing hitter's OPS is .017.
Concentrating only on those hitters and pitchers with 400 to 600 PAs, the standard deviations are .010 and .022, respectively.
Pitchers are much more likely to be influenced by their schedule than hitters are.
Posted 10:34 p.m.,
September 19, 2003
(#3) -
Tangotiger
By the way, the reason that pitchers will be more influenced is not necessarily the distribution of their opposing teams, but rather than there is a larger spread of true talent among hitters than pitchers on a per PA basis.
If someone wanted to, they can figure out the stdev on OPS for hitters and pitchers with at least 400 PA. I don't think it will be twice as much for hitters, but it will probably be close to it.
Posted 8:11 a.m.,
September 20, 2003
(#4) -
Sylvain(e-mail)
(homepage)
As for "adjusting to the opposition faced" the homepage links to a post I wrote for a Halladay thread: I tried to factor the difference of IP pitched agaisnt same teams between Loaiza and Halladay in order to arrive to an ERA that would be more "representative".
BTW Prospectus does a similar thing in taking into account the fact that over a season a pitcher doesn´t face his own team (see Team Pitching Adjustment http://www.baseballprospectus.com/cards/glossary.shtml)
But as Kurt points out, a major problem I faced is sample size. Are the 8 IP by Loaiza against BOS for real? Are the 26 IP pitched by Halladay against DET for real? How to regress these performances? I remember (David Smyth or Patriot) provided a chart saying by how much one should regress towards the mean depending on the PA. Can I do the same for ERA against one team: regress it towards the season ERA after taking into account the opposition faced?
For example: suppose Loaiza pitched 12 innings with 1 ER against DET (OPS 650) but has en era of 2.5 for the season (avg OPS 750)? Or do I have to do an odds ratio method first (what would be Loaiza´s season ERA had he faced a 650 OPS) and then regress the 1 ER in 12 IP towards this "mean"?
This would (I think) give a better estimation, even if a second problem would be to assess the OPS of the teams faced (hot month/injuries: one could take the OPS of the team during the week of the start considered, regress it as well...).
Please correct me if I´m wrong.
Any other ideas? Or is an odds ratio method at the end of the season in order to adjust to the OPS faced enough?
THEN would come the point of the teams that a pitched faced and not the other (interleague games/schedule).
Thank you
Sylvain
Posted 12:54 p.m.,
September 20, 2003
(#5) -
Charles Saeger(e-mail)
Tango -- a long-time thought, I'm sure I posted this elsewhere, but I'll reiterate. Could one use opposing hitter quality to get a handle on the luck factor of DIPS?
Posted 3:33 p.m.,
September 20, 2003
(#6) -
Tangotiger
Charlie,
Suppose that pitchers did not have a home park, and instead randomly played at a park for each start. They won't pitch exactly once at each park. Some parks they won't pitch in, and others they might pitch 2 or 3 times in.
As long as the distribution of where they pitch can be explained by random chance, then we don't need to consider the park factor. I think that the Central Limit Theorem would apply (though don't quote me on that).
The same would be the case with their opposing hitters. As long as the distribution can be explained by random chance, then the TRUE VARIANCE (which is what we are after) would be equal to zero. Now, I grant you, that the opposing hitters might not be due to random chance, and there is something at work here. However, I would guess that we are talking about a true variance of .001 to .002. I would be surprised if it's any higher than that.
Posted 6:23 p.m.,
September 20, 2003
(#7) -
Robert Dudek
Tango,
What about the notion that the best hitters usually bat 3rd or 4th, where they have more PA with men on base as a percentage of their total PA than the average hitter. Isn't this the same, on a smaller scale, as your leverage index for relievers. I believe that you found that starting pitchers do not deviate from 1.0 very much on the LI scale.
It's also possible that players like Bret Boone and Alex Rodriguez, who play key defensive positions and bat a lot with men on base have effective LIs that put them beyond the reach of a starting pitcher that faces 1000 batters.
Posted 6:43 p.m.,
September 20, 2003
(#8) -
Tangotiger
My guess is that the #3 and #4 hitters have an LI of 1.05 to 1.10.
As for 2B/SS, I don't think you'll find much spread in talent there compared to 3B or CF, fielding-wise.
I agree that it is tough for a pitcher to keep up, but it is not unreasonable to think a great year from a pitcher is equal to an almost great year from a hitter. Pedro and Gooden and Maddux and RJ easily equal the greats, regardless what other analysts say. Loiza/Halladay? Maybe not, but they are in the top 10.
Posted 10:33 a.m.,
September 22, 2003
(#9) -
Alan Jordan
Tango -
"As long as the distribution of where they pitch can be explained by random chance, then we don't need to consider the park factor. I think that the Central Limit Theorem would apply (though don't quote me on that)."
It's not a question of central limit theorem. I've noticed that when you invoke the central limit theorem what you usually mean is "when the sample gets large enough". I think that's what you mean here. The central limit theorem relies on the later, but it's not the later itself.
The question of whether park factors are necessary hinge on two things.
1. How important are the park factors - the larger they are, the more likely you need to deal with them.
2. How evenly distributed are the park appearances for the pitchers - this is where your comment comes into play. Ideally with enough starts randomly scattered across the parks no park adjustment would be necessary because the appearances would be approximately evenly distributed among parks. Unfortunately this would take far more than the 30- 45 starts that pitchers get. It would probably take a couple of hundred.
The idea of randomness is that with a large enough sample you get eveness of distribution, but it's much more efficient to do non random distribution if you want eveness. For example have a pitcher start one game and only one game in each park. Then park factors would be unecessary.
The best resolution of this question is to handle it on a play by play basis. That way you can factor in hitter, pitcher, park, balls in play and event.
Posted 8:55 a.m.,
September 30, 2003
(#10) -
Rally Monkey
I don't completely trust looking at opposing hitter's OPS to determine a pitchers level of competition because of platoon issues. For example, a lefty facing the Red Sox would look like he's getting a break by not having to face Trot Nixon and his 950? OPS, instead facing Gabe Kotter. If they left Nixon in, the pitcher would acctually be the one getting a break, since Trot can't hit lefties.