PDA

View Full Version : New Pitcher Performance Metric?

RedsManRick
04-04-2008, 03:25 PM
This might not even make sense, but bear with me. When looking back on a pitcher's season, we often look at his component performances to get away from the variability inherent in run scoring based on defensive performance, environment, etc. Metrics like FIP try to isolate how well the guy actually pitched as opposed to how many earned runs were attributed to him -- itself a fairly convoluted calculation.

Standard Bill Jamesian game scores are nice, but use hits and runs as modifiers -- events in part beyond the pitcher's control and thus not real accurately reflective of how well the guy pitched.

The basic premise here is that pitchers don't have much control over WHEN at bats events occur. In the long run, the pitchers who do the things that tend to lead to fewer runs are better than the pitchers do the things that lead to more runs.

Within a given outing, the run scoring variability due to defense and random timing of events can be huge (9 walks, 0 runs!). A grounder scoots through with the bases loaded, scoring two runs. It was 6 inches from a double play. Looking back at the line score, we get the impression that the pitcher got lit up when 6 inches could've meant a quality start. On the flip side, sometimes a guy gets hit around pretty hard, but the hits are distributed as such to spread those hits out in such a way that not many runs scored. Generally speaking, I don't think that type of spread tends reflects a specfic clutch skill of the pitcher. He just got lucky that while pitching poorly, he didn't give up many runs.

The idea is to get a better sense how well the guy actually pitched (or has pitched over a small sample of starts/appearances), and thus how well he's likely to pitch in the future. Looking at what happened in part due to his pitching is looking through a fog.

Would it be possible to create a game score like metric that incorporated BB, HR, LD, FB, and GB, using average run values for each non-out event? The final score is a tally of "average runs" allowed. If you wanted, you could then divide that run total by the number of outs or innings the pitcher accumulated for a rate stat.

This acknowledges that pitchers can affect batted ball types and tries to give them more credit for what they do versus what happens after a more complex series of events not as within their control. If a guy doesn't strike many out but still gets out, his R/O score would still be higher because batted balls tend to lead to more runs than strikeouts. Even a perfect game would have a positive run score, unless he struck out all 27 batters. And if a guy gives up 4 hits in 7 innings and they just so happened

I realize that after doing this, it might become clear that pitchers do have more of an influence on timing that I think they do. That at bat outcomes are not actually distributed normally and that the skew makes this type of analysis meaningless (evidence of this would be runs scored totals that over the long haul come in higher or lower than the "average runs" total). But we won't know until we try.

So, am I crazy? Is somebody already do this? Are there significant methodological problems -- poor assumptions, etc? Thoughts?

ochre
04-04-2008, 03:35 PM
Sounds like you want to define "clutch" for pitchers :)

bucksfan2
04-04-2008, 04:01 PM
Sounds to me like you want to quantify the intangibles. I think you are entering an area where the value/number would be too subjective to take serious. I think there will be a problem differentiating between human error and outside forces. Lets say a ground ball is just outside the reach of an infielder. Why was that just outside of his reach? Was it because he got a poor jump on the ball? Was it because it hit a hard spot in the grass/dirt? Could it have hit the collar where the grass ends and the dirt begins causing the ball to shoot? Is the fielder fatigued? Another thing you have to consider is whether the fielder would have been able to field the ball cleanly and then make a corresponding throw to get an out.

RMR what you are trying to quantify is a very important part of baseball. I think this is done by the scouts on a grading scale. I think a good scout can look at a player and say he was unlucky in a game or he pitched better than the numbers say. I also think that you learn more about a pitcher by what he does when he isn't on his top game. He may be hit hard but not be hit hard when it counts. Look at Cueto from yesterday, he was very very impressive and on the top of his game. The thing that will be important for me is how he pitches when he is slightly off. The best pitchers limit the damage while the aveage pitcher will get lit up.

Far East
04-04-2008, 04:10 PM
Within a given outing, the run scoring variability due to defense and random timing of events can be huge (9 walks, 0 runs!).

The "random events" comment reminds me of Reds' Jim Maloney's (8/19/65) no-hitter despite giving up 10 walks.

Just as improbable -- or more so -- was the St. Louis Brown's Bobo Holloman's gem on May 6, 1953. Improbable, because he only managed to win 2 other big league games to finish his one-year (rookie, at age 30) career at 3 wins,7 losses.

“Some people would call him (Bobo Holloman) a screwball I guess, but I’m mighty happy that he pestered me into giving him his chance to start that game (the no-hitter he threw on May 6, 1953). He proved to me that he’s just about as good as he thinks he is.” - St. Louis Browns Manager Marty Marion

RedsManRick
04-04-2008, 04:23 PM
Sounds to me like you want to quantify the intangibles. I think you are entering an area where the value/number would be too subjective to take serious. I think there will be a problem differentiating between human error and outside forces. Lets say a ground ball is just outside the reach of an infielder. Why was that just outside of his reach? Was it because he got a poor jump on the ball? Was it because it hit a hard spot in the grass/dirt? Could it have hit the collar where the grass ends and the dirt begins causing the ball to shoot? Is the fielder fatigued? Another thing you have to consider is whether the fielder would have been able to field the ball cleanly and then make a corresponding throw to get an out.

RMR what you are trying to quantify is a very important part of baseball. I think this is done by the scouts on a grading scale. I think a good scout can look at a player and say he was unlucky in a game or he pitched better than the numbers say. I also think that you learn more about a pitcher by what he does when he isn't on his top game. He may be hit hard but not be hit hard when it counts. Look at Cueto from yesterday, he was very very impressive and on the top of his game. The thing that will be important for me is how he pitches when he is slightly off. The best pitchers limit the damage while the aveage pitcher will get lit up.

I guess I communicated it backwards. I'm trying to measure how well a guy actually pitched. That is, as a pitcher, did he do the things that lead to the fewest runs being scored.

The impact of other things like defense, event timing -- both random variation and that based on pitcher clutch, etc. are simply the difference between what actually happened and what the pitcher did, as measured by the average effect of those bat events.
The entire subject of pitcher clutch, limiting damage, or caving to pressure, etc. is something that might come out of a comparison between my stat and runs allowed -- it is not the point of my stat. I'm trying to measure pitcher performance -- the pitcher, himself, in isolation.

When you look at how many hits a pitcher gave up, for example, the first paragraph comes in to play. Who knows the hundreds of variables in play that affect whether or not a specfic ball was a hit or not? I want to remove your entire first paragraph from the conversation.

For example,

Let's assume that at bat events have the follow run weights. (a linear weights model could come up with these pretty quickly, I presume). SO are omitted since the have a 0.0 run value.

BB: 0.1 runs
HR: 1.5 runs
GB: 0.03 runs
FB: 0.04 runs
LD: 0.2 runs

The we apply this to pitcher performance. On 4/3, Johnny Cueto did the following (count of batted ball types are made up for this example).
0 BB: 0 runs
1 HR: 1.5 runs
6 GB: .18 runs
7 FB: .28 runs
4 LD: 0.8 runs
Total: 2.76 "average runs"

So, based on the at bat events resulting from Cueto's pitching, we would have expected an average of 2.76 runs. Now, as it so happened last night, batted balls found a lot of gloves, a credit to luck and/or the Reds defense. There might also be nights where he gives up 6 ER even though his "average runs" score is 4.71. The "average runs" or perhaps "fair runs" is an alternative figure to Earned Runs or Runs Allowed. You could easily make a rate stat out of it.

Thus, we get a better sense of how much effect the timing of the events impacted the final outcome. If we are correct in assuming that at event timing is more or less random, we would want to control for it. This would allow us to get a much better sense of that. That is the end point of what I'm aiming for.

Now, if over time, we see that certain pitchers consistently beat their "average runs" figure, we might say that they can positively control at bat outcomes. That is to say they are good at minimizing damage. And on the other hand, if certain pitchers consistently lose to their "average runs" figure, we might say that they are "prone to the big inning". This would be a very interesting finding in my book.

One thing that FIP really fails to do is account for batted ball type. It also is an adjustment off a mean based on every ball being hit in to play, rather than as absolute value. In my mind, we should start from the assumption that no ball is hit in to play and add runs every time it is (or a player is put on base via walk). This is essentially a modified FIP, using batted ball types, starting from 0 instead of 3.2 ,and applied on an outing by outing basis instead of just in aggregate.

Does that make sense?

IslandRed
04-04-2008, 05:12 PM
Sounds almost like you're trying to develop a "derivates" (borrowing the term from Moneyball) system for pitchers.

First possible questionable assumption would be the randomness of events, which leads to assigning static weights to non-strikeout events. You said:

The entire subject of pitcher clutch, limiting damage, or caving to pressure, etc. is something that might come out of a comparison between my stat and runs allowed -- it is not the point of my stat. I'm trying to measure pitcher performance -- the pitcher, himself, in isolation.

Except that a knack for limiting damage etc. is a measure of pitcher performance if it is detectable, repeatable and based on sound thinking. Take Tom Glavine in his prime. His walk rate wasn't as absurdly low as Maddux's, but it was still low, to be sure; but when he issued a walk was very much under his control, and tended to be a matter of not giving in to good hitters in key situations, not random acts of poor command. To not account for that is to not measure Glavine's performance accurately. Similarly, it's rather an old-school concept to be more willing to challenge hitters with the bases empty. The difference between a good and bad pitcher isn't just that Good Pitcher gives up ten fewer home runs; often, it's the three-run jacks in key situations that are missing from his ledger, not the solo shots when leading 4-0.

But for most pitchers it's probably not enough of a factor to worry about. The next possible glitch, if you're going for a true evaluation of pitchers, is that to an extent you're still measuring outcomes (what did the hitter do with the pitch) rather than process (how good was the pitch), and while GB/FB/LD is a better breakdown than simple BABIP, it's still pretty coarse for getting at the true answer. I'm really anxious to see what happens with Pitch/fX data down the road.

Still, the idea has some potential, and I'd be interested to see if it does a better job of predicting and correlating with real-world results than does something like xFIP.

RedsManRick
04-04-2008, 05:17 PM
I don't think this necessarily requires pitcher clutch be non-existent to have value. Even assuming pitcher clutch does exist (defined as the repeatable skill of a pitcher to suppress run scoring below expectation), this metric would give us a starting point from which we can measure the effect of pitcher clutch. Defense would be a part of it as well. But then you can pull in some defensive metric to control out defense and you have left the timing aspect.

I think it would be very interesting to see the spread of the difference, once defense is factored out. How much does pitcher clutch really matter? What's the scale? .1 of a run per 9 IP? .5 runs?

Maybe I'll spend some time on it on Sunday if I can.

bucksfan2
04-04-2008, 05:29 PM
RMR I get your point and it is intersting. I think as IslandRed allued to you would even further have to wait situational aspects of a given game. Do you wait an intentional BB differently from a regular BB or an un-intentional intentional BB differently? Also do you weight the different types of balls put into play depending on their pitch count? For example if a pitcher gives up a line drive or HR on a 0-2 pitch that is obviously a mistake but if that is given up on a 3-0 or 3-1 count then do you weight it differently?

Also as mentioned above I think there is a big difference between a solo shot and a multiple run shot. If you have a multipe run lead 3-0, 4-0 giving up a solo HR doesn't do too much damage. If you give up a few hits and enable a quasi rally then you are doing more damage than one solo shot.

RMR you may be up to something but I think the big problem is finding a way to quantify all the different variables into the formula.

M2
04-04-2008, 06:21 PM
Let's assume that at bat events have the follow run weights. (a linear weights model could come up with these pretty quickly, I presume). SO are omitted since the have a 0.0 run value.

BB: 0.1 runs
HR: 1.5 runs
GB: 0.03 runs
FB: 0.04 runs
LD: 0.2 runs

The we apply this to pitcher performance. On 4/3, Johnny Cueto did the following (count of batted ball types are made up for this example).
0 BB: 0 runs
1 HR: 1.5 runs
6 GB: .18 runs
7 FB: .28 runs
4 LD: 0.8 runs
Total: 2.76 "average runs"

Nifty idea. I'm assuming hits would be counted by batted ball type (GB, FB, LD), otherwise those would have no run value.

Really, the next step would be to determine the actual run values of those incidents and then plug in a season's worth of games to see what it nets you.

I think it's an intriguing idea in that it at least attempts to get at bat on balls effects in pitching (where I suspect differences are logarithmic rather than linear).

RedlegJake
04-04-2008, 06:41 PM
Let's assume that at bat events have the follow run weights. (a linear weights model could come up with these pretty quickly, I presume). SO are omitted since the have a 0.0 run value.

BB: 0.1 runs
HR: 1.5 runs
GB: 0.03 runs
FB: 0.04 runs
LD: 0.2 runs

The we apply this to pitcher performance. On 4/3, Johnny Cueto did the following (count of batted ball types are made up for this example).
0 BB: 0 runs
1 HR: 1.5 runs
6 GB: .18 runs
7 FB: .28 runs
4 LD: 0.8 runs
Total: 2.76 "average runs"

So, based on the at bat events resulting from Cueto's pitching, we would have expected an average of 2.76 runs. Now, as it so happened last night, batted balls found a lot of gloves, a credit to luck and/or the Reds defense. There might also be nights where he gives up 6 ER even though his "average runs" score is 4.71. The "average runs" or perhaps "fair runs" is an alternative figure to Earned Runs or Runs Allowed. You could easily make a rate stat out of it.

I like the idea. Figure like an ERA, multiply by 9 divide by innings or perhaps, on a Batter faced basis, x27 divided by batters faced (Not sure of the math on that one)for efficiency and what you can expect for every batter faced.

RedsManRick
04-07-2008, 02:49 AM
Ok, preliminary results are in. Because runs are scored due to the combined pitching efforts of starters and relievers, the analysis only makes sense using either complete pitcher innings (ideal), or whole teams. Actually, the best way to do this would to be each inning itself be the observation, with it's associated set of observed values (Runs, Walks, Ground Balls, etc.). However, because I'm lazy, I went with the first reasonable idea that came to mind.

Since I couldn't find batted ball data by team, I took the individual players who comprised each team to do the preliminary math. Of course, some players changed teams, so I would have to account for them somehow. So, since I couldn't get their team splits on batted ball data, I took only the pitchers with at least 10 IP who pitched for only 1 team during the 2007 season. Yes, I know that doesn't actually get rid of those times when a pitcher was relieved by a player who pitched elsewhere. But it did answer the question about where those guys go, since I didn't have batted ball data by each team they played for.

Once I calculated the number of LD, GB, and FB (off the percentages times a calculated number of balls in play), I then had to subtract out the HR (since the percentages include them). Throw in BB, and this gave me the mutually exclusive 5 AB outcomes I discussed earlier.

Because it's late and I'm tired, let me skip to the results. Run values:

LD: .324
GB: .019
FB: .005
BB: .506
HR: 1.744

Applying these to teams and players shows the following. All you'll see displayed here is Team/Player, Runs/9, Fair ERA (as I'm calling it), and the difference. The difference column can be read as a measurement of the impact of external factors (park, defense, luck)on the pitcher's performance. Negatives means they were helped and positives mean they were hurt.

Team R/9 FERA Diff
Indians 4.22 4.07 -0.15
Dodgers 4.16 4.32 0.16
Angels 4.51 4.34 -0.18
Red Sox 4.07 4.35 0.28
Blue Jays 4.21 4.43 0.22
Athletics 4.73 4.45 -0.28
Rockies 4.46 4.57 0.11
Brewers 4.75 4.62 -0.13
Twins 4.45 4.64 0.19
Giants 4.30 4.64 0.34
Mariners 5.02 4.64 -0.38
Braves 4.26 4.73 0.46
Cardinals 5.01 4.73 -0.29
Tigers 4.76 4.76 0.01
Royals 4.65 4.77 0.12
Yankees 4.84 4.77 -0.06
Pirates 5.11 4.78 -0.33
White Sox 5.17 4.79 -0.39
Cubs 4.23 4.79 0.56
Mets 4.53 4.82 0.29
Diamondbacks 4.40 4.89 0.49
Reds 5.28 4.99 -0.30
Astros 4.94 5.02 0.08
Orioles 5.31 5.06 -0.25
Devil Rays 5.76 5.08 -0.68
Nationals 4.87 5.15 0.29
Rangers 5.46 5.23 -0.23
Marlins 5.38 5.24 -0.14
Phillies 5.00 5.28 0.28
Average 4.72 4.72 0.00

Well, I did something right, because I see the sorts of numbers I expect. The Devil Rays and Reds pitchers were particularly screwed by external influences. Nats, Mets, and Giants were all helped. But I don't know what to make with the Diamondbacks -- it that the effect of intelligent bullpen leveraging? What about the Cubs? Are the Padres really neutral? Defense offsetting park perhaps?

Player R/9 FERA Diff
Gary Majewski 8.61 3.91 -4.70
David Weathers 3.82 4.06 0.23
Aaron Harang 3.88 4.08 0.19
Marcus McBeth 5.95 4.12 -1.83
Jared Burton 3.14 4.19 1.05
Bill Bray 6.28 4.40 -1.88
Bobby Livingston 5.59 4.63 -0.96
Mike Stanton 6.09 4.68 -1.40
Bronson Arroyo 4.66 4.87 0.21
Matt Belisle 5.62 4.96 -0.66
Eric Milton 6.03 5.17 -0.86
Jon Coutlangus 4.83 5.27 0.44
Homer Bailey 6.35 5.36 -0.99
Kirk Saarloos 7.59 6.12 -1.48
Todd Coffey 6.35 6.35 0.00
Thomas Shearn 4.96 6.61 1.65
Mike Gosling 6.00 7.64 1.64
Elizardo Ramirez 7.71 8.27 0.55
Phil Dumatrait 15.00 9.50 -5.50
Total 5.28 4.95 -0.33

FERA clearly agrees with it's cousin FIP, Majewski got screwed by outside influences, as did McBeth, and even Mike Stanton. Todd Coffey and his crazy HR/FB shows up too. Meanwhile, Tom Shearn's luck is shining through. Burton got a healthy bump too.

Ok. There are all kinds of problems with this analysis, particularly at the player level, because I'm not clear on how runs are attributed (as opposed to earned runs). Also, I think the HR inclusion aspect is problematic. The real point of this metric is to isolate pitcher from external influences. HR are affected by park, altitude etc. And because they are more or less a function of LD & FB, we might better get at the question by excluding them, even if it makes the model "fit" worse -- fit being the stats term for the formula adding up to the thing it's supposed to -- runs in this case.

However, I'm not sure if we should care about that. In fact, it's what we'd expect. Park effects are making pitchers look more different that they actually are. They make Jake Peavy look better than he is and Aaron Harang worse than he is. I run a quick and dirty version ignoring HR (simply treating them as another LD or FB) and it did some interesting things.

Now that I have something on paper, I'm going to try and revisit this and work on tightening up some assumptions and see if I can't get it closer to something I'm comfortable with. Part of the problem when you're trying to measure "reality" is that, by nature of the challenge, you can't compare your estimates against the real thing. This is the same fundamental challenge with fielding metrics. If we had the "real" numbers, it would render the whole exercise moot. But since we don't, we can only compare our results against others. So, once I get it tightened up a little, I'll compare it to some other metrics out there and see how it does.

I must admit, I'm a bit surprised we hear about the batted ball data being used to evaluate fielders, but not pitchers or fielders directly. Why are relying on backing in to metrics for hitters and pitchers through summary data (event outcome run values) instead of simply event run values? Maybe it's a sample size issue. You have to aggregate at the outcome level because you don't have enough GB to zone 7 events, but I do wonder if somebody has tried...

IslandRed
04-07-2008, 11:14 AM
Also, I think the HR inclusion aspect is problematic. The real point of this metric is to isolate pitcher from external influences. HR are affected by park, altitude etc. And because they are more or less a function of LD & FB, we might better get at the question by excluding them, even if it makes the model "fit" worse -- fit being the stats term for the formula adding up to the thing it's supposed to -- runs in this case.

That flies in the face of convention (home runs being one of those things that's assumed to be within the pitcher's control in measures like DIPS or xFIP), but it could work as long as you adjust the LD/FB run values to account for their share of what previously went into the HR space. It requires an assumption that pitchers have no control over the HR/FB rate, which is not always the case for individuals, but it will work fine in the aggregates.

RedsManRick
04-07-2008, 11:33 AM
That flies in the face of convention (home runs being one of those things that's assumed to be within the pitcher's control in measures like DIPS or xFIP), but it could work as long as you adjust the LD/FB run values to account for their share of what previously went into the HR space. It requires an assumption that pitchers have no control over the HR/FB rate, which is not always the case for individuals, but it will work fine in the aggregates.

Well, I do wonder. Controlling for park, is there are HR prevention ability, as captured in HR/FB and HR/LD rates that are different from merely allowing those batted ball types in the first place. I've not seen evidence suggesting that.

Rather, due to the correlation between the run value of LD and FB with HR, I think FIP uses HR essentially as their proxy. This doesn't sit well with me. The art of pitching is essentially putting the ball in the strike zone while minimizing the ability of the hitter to make solid contact. When HR are included, the run value of FBs nearly disappear. This just doesn't pass the smell test to me. HR happen on well hit FB and well hit LD. They are subset of those events when a batter gets his bat on the ball a certain way. Treating them as something else entirely inappropriate creates a category of batted ball events called non-HR FB.

Using HR suggests pitchers can control the portion of FB (and LD) which become HR. But if that's truly a skill, shouldn't we also see the other side of that, pop-ups? Maybe we should include those too. What about bloopers? What about worm burners?

It seems that looking at homers explicitly is the first stepping in moving in to a, perhaps appropriate, further subdivision of batted ball types. To me, it's a slippery slope down to raw hit F/x data, using speed and trajectory instead of hit type classifications. It also introduces park effects, which as well all know have a significant affect. How would Chris Young have fared in Cincinnati? How about Boston where liners become doubles instead of HR. Do we credit the pitcher for the positioning of the walls?

Anyways, this was really rough and I'm not sure how much better I can get it with the data I have, but I'll at least try. One way I think we can test it is by looking at it's year-to-year consistency for a given player. Presumably, the ability of pitchers is more constant than the external affects. We'll see.

IslandRed
04-07-2008, 12:06 PM
To me, it's a slippery slope down to raw hit F/x data, using speed and trajectory instead of hit type classifications.

It does point out the difficulties of trying to shoehorn all batted balls into just a few classifications. When Adam Dunn hits one of his shots that is simultaneously high and on a rope, coming back to earth 450 feet later, is that a line drive or is that a fly ball?

I think that's why the value for non-HR FBs was so low in your initial figuring -- if a ball is hit in the air, doesn't go out of the park and yet is hit hard enough that it can't be reached by an outfielder before it falls in for a hit, it probably goes in the books as a line drive at least 95% of the time.

Cedric
04-07-2008, 12:42 PM
Pretty soon there isn't going to be anything to discuss about a player. I like leaving something open to the personal opinion of a fan that has watched different pitchers from different eras. Not everything needs to be put in a tidy number, IMO.

RedsManRick
04-07-2008, 12:44 PM
Pretty soon there isn't going to be anything to discuss about a player. I like leaving something open to the personal opinion of a fan that has watched different pitchers from different eras. Not everything needs to be put in a tidy number, IMO.

I don't think we're remotely close to getting to that point, nor will we ever. I've never understood the idea that analysis somehow runs counter to the qualitative side of the game. Player evaluation is not a zero sum game.

bucksfan2
04-07-2008, 12:47 PM
I have two questions for you.
1. It seems that your metric will favor strikeout pitchers rather than non strike out pitchers. I understand that anytime you put a ball in play you are dangerous but there are pitchers who make a living off keeping hitters unballanced at the plate. I wonder if there is a way to differentiate between weak and hard hit ground balls as well as soft and hard liners?

2. Is their a way to weight hits/runs due to the run margin in a particular game. A 2 run HR with a 6-7 run lead isn't nearly as damning as a 2 run HR with a 2 or even 4 run lead. It seems that pitchers with a big lead will pitch differently and not sweat a run here or run there but if the game were closer those runs would be differently. On the same hand if you are down 2 runs and you give up a two run inning that is much more damaging than being down 6 runs and giving up a two run inning.

RedsManRick
04-07-2008, 01:20 PM
I have two questions for you.
1. It seems that your metric will favor strikeout pitchers rather than non strike out pitchers. I understand that anytime you put a ball in play you are dangerous but there are pitchers who make a living off keeping hitters unbalanced at the plate. I wonder if there is a way to differentiate between weak and hard hit ground balls as well as soft and hard liners?

Yes, it does -- and it should. This is the reality of baseball. While certain pitchers can pitch such that they allow weaker balls in play that others (see Mariano Rivera), the spread of this skill within the major league ranks is quite small -- almost to the point of being negligible. And I would argue that the ability to do this strongly correlates with strikeout ability. After all, what's the difference between a poorly hit ball and a strikeout? There's a spectrum: miss - weakly hit ball - strongly hit ball - crushed.

Take the case of Greg Maddux. In his prime, he missed a fair amount of bats. However, as he lost some velocity and movement with age, balls got hit in to play more often and a little bit harder. His BABIP during his prime in the mid 90's tended to be in the .270 range -- pretty much the floor of what's ever been done before. He's now in the .300 range regularly. His HR/FB rate has shot up as well. He's not just striking out fewer, he's getting hit harder. I would argue that those those things are directly related to the same underlying ability/performance.

Sure, Maddux still keeps hitters off balance with his pitch selection, control, etc. And he's famous for pitching to his defense. So, to your second point...

2. Is their a way to weight hits/runs due to the run margin in a particular game. A 2 run HR with a 6-7 run lead isn't nearly as damning as a 2 run HR with a 2 or even 4 run lead. It seems that pitchers with a big lead will pitch differently and not sweat a run here or run there but if the game were closer those runs would be differently. On the same hand if you are down 2 runs and you give up a two run inning that is much more damaging than being down 6 runs and giving up a two run inning.

I would posit that pitchers never try to give up hits. Sure, they might pitch to contact a bit more in a blow-out, but I think that effect is somewhat minimal. Even then, why would we want to adjust for it? To say that because a pitcher was trying to give up runs, that we shouldn't count for it or should somehow adjust doesn't make sense to me. Should we adjust a player's OPS upwards if he purposefully sacrifice bunts? That's certainly the argument people use for not counting sacrifices at at bats. But I think that kind of thinking confuses the point. Are we measuring what happened or trying to measure an ability?

In this case, we're trying to model the "reality" of how the guy pitched - no more or less. If he chose to pitch to contact, I'm not sure we'd want to account for that choice by somehow giving him a break. I guess if you had the data, you could filter out any at bat in which the run differential or leverage was a certain amount. But just because you're up or down 6 runs doesn't mean it makes sense to allow better hit balls or not to try and strike guys out. Sure, you want to avoid walks and so you might be around the zone a bit more -- but you should be avoiding walks anyways. While there is a real world logic to easing up to save bullets, I'm skeptical about giving guys a break for stinking in low leverage situations. Maybe the guy is a good pitcher who's simply easing up. Or maybe he's just asked to pitch in that situation in the first place because he stinks. I'm very hesitant to assume the former.

SteelSD
04-08-2008, 02:22 AM
Um...Rick, there were shorter answers to bucksfan2's questions. ;)

1. Baseball itself favors high K-rate pitchers. Always has. Always will.

2. Like Hitters, Pitchers can't choose the game state in which they pitch. There's no real information to be gleaned by incorporating game state.