Intersting...

One problem with comparing a pitcher´s BABIP to his team´s is that, as noted in Tom Tippett´s critique of DIPS, is what happens when a pitching staff generally has a skill for low BABIPs. In this case, it becomes increasingly difficult (if not impossible) to statistically disentangle pitching from defense.

1. Sac bunts should fall in category 3, not category 2.
2. Parks can impact walk rates, by way of the lighting conditions, hitter's backdrop, visibility, size of foul territory, etc.
3. This analysis appears to neglect the (observeable) fact that some pitchers do not pitch well from the stretch. A pitcher who struggles from the stretch will (consistently) put up a worse ERA than his components would suggest.

To commenter #2:

What evidence do you have that pitchers have any ability to prevent a team from sacrifice bunting consistently? It seems to me that if a team wants to drop down a sac bunt, there is litte the pitcher can do about it. Sac bunts are more or less beyond a pitcher's control...if you face an aggressive team like the As who don't bunt, you're not going to give up sac bunts.

I suppose it's possible parks could influence walks a little, but I am highly skeptical that that effect can be measured accurately. It seems to me that walk rates are more heavily effected by the quality of the pitcher than any park effect...and if there is an impact, I'm of the beleif that it's negligible for 99.999999% of pitchers.

And you are correct about the stretch. That's something I hadn't considered, and will need to look into.

poster #1...I agree there is the potential for some error in teams with a clsutering of good pitchers (mid-90s Braves for example) but I still believe this is closer than starting with RA and adjusting for defense...as the article notes...research is ongoing and I'm to suggestions...though you're right that it would be nearly impossible to tell if the clustering effect were occurring.

Dan Szymborski's 3 year weighted park factors show a range of BB park factors of .79 to 1.27. This suggests that the effect of the ballpark on walk rates is non-negligible.

Or perhaps that BB park effects are completely unreliable as a general rule. The way park effects in general are calculated these days concerns me. I don't believe that the effect of a park should be measured in terms of a ratio of events to league events...I believe it should be measured in terms of events per opportunity added (which is why I did my HR park factor in that form, BTW) and when it comes to estimating the effect of a park on walks, the pitchers who happen to appear at that park will have an ENORMOUS influence on the walk rates within that park, and cause all kind of serious problems with any attempt to quantify such. No matter how much 3-year weighting you do, you need to find some way to determine how wild the pitchers are who throw in that park before you go around calculating park factors assuming a roughly average pitching wildness.

Let's think logically about this for a brief moment.

Do we really anticipate that small effects like lighting and the batter's eye backdrop around going to cause the walk rate to change in a park MORE than the RUN SCORING RATE changes (Coors Field excluded of course)? What you're effectively saying if you assume that those park factors have any meaning at all is that the lighting in a park is more important than its' dimensions, contours and grass, which is beyond bizarre to be claiming.

Seriously, how can we honestly defend the reasonableness of a statistic which claims that the park can have a 25% pull on walk rates? It just defies common sense.

Just scanning the results but it looks like according to this metric 8 of the 10 best pitchers since 1957 played the majority of their career in the nineties. There has to be some kind of bias there.

Road to a successful thread:

1.) Have your article linked
2.) Post yourself in the thread at a 50% clip
3.) Challenge the rationality of the other posters

Seriously, how can we honestly defend the reasonableness of a statistic which claims that the park can have a 25% pull on walk rates? It just defies common sense.

Perhaps it is reasonable to think that a 25% pull on walk rates is too high. But it is not reasonable to decide to ignore park factors for walks entirely on that basis. Quite the contrary - it shows they need to be considered, and considered carefully. If you wish to come up with your own park factors for walks, then go right ahead. If you wish to (try to) demonstrate that parks have a negligible affect on walk rates, then go right ahead. But throwing away park factors for walks is not acceptable, when prima facie parks do affect walk rates.

As for sac bunts - the number of times the batting team attempts to bunt is largely beyond the pitcher's control. But the batter has no guarantee of executing the bunt successfully; and it is up to a combination of the fielders and the pitcher to stop him. When the batting team calls for a sac bunt, there are many possible outcomes, including:

Sac Bunt
Fielder's Choice
Double Play
Error; both runners safe
etc

Road to a successful thread:

1.) Have your article linked
2.) Post yourself in the thread at a 50% clip
3.) Challenge the rationality of the other posters

Well that's more than a little unfair. Would it be preferable that he didn't respond at all?

I'm here because I was informed that my article was linked by BTF...I can't tell whether your comment was a complaint Fridas Boss, but if there is to be discussion of the DNRA article, I fail to see why my participating in said discussion is unfavorable to BBTF and I why I should not respond to the concerns of others. And on the issue of park factors, YES I challenge the rationality of others because theway park factors are calculated today is completely irrational.

To poster #8, no I wouldn't conclude that just because there has been a recent surge of great pitchers means there is a bias. Are you saying that Pedro Martinez, Greg Maddux, Roger Clemens, Randy Johnson, Mike Mussina John Smoltz, Kevin Brown and Curt Schilling aren't legitimately great pitchers?

I think, if you're sensing a bias, the only real bias there is at the moment is that most of the pitchers have not yet completed their careers, and therefore have not yet had the requistite "downside" chipping away at their DNRA, whereas if you go back to the pitchers in the 50s and 60s, they had the downside recorded, but had the early part of their career left off because it was before the advent of PBP.

How appropriate that you use Tom Glavine as an example. He will serve excellently to demonstrate part of the problem with this type of approach.

Here are glavine's splits since 1987 with bases empty and runners on

Situation       IP     H     BB     K     HR
Bases Empty   2051.1   2367     628   1475     223
Runners On    1673.0   1606     760    975      96

What should be readily apparent is that Tom Glavine is not the same pitcher when runners are on base that he is when the bases are empty. When runners are on base, he nibbles more, gives up more walks, but refuses to leave the ball over the plate. Any model that uses a linear weights average for the HR weight will totally misvalue him.

To poster #8, no I wouldn't conclude that just because there has been a recent surge of great pitchers means there is a bias. Are you saying that Pedro Martinez, Greg Maddux, Roger Clemens, Randy Johnson, Mike Mussina John Smoltz, Kevin Brown and Curt Schilling aren't legitimately great pitchers?

I think the bias isn't really specific to your method but rather in all methods that compare pitchers to the average. It is much easier to be better than average in high scoring eras. Sure all those guys were great pitchers but there were pitchers just as good in the sixties, seventies, and eighties.

Do we really anticipate that small effects like lighting and the batter's eye backdrop around going to cause the walk rate to change in a park MORE than the RUN SCORING RATE changes (Coors Field excluded of course)?

I'm not sabremetrician, but I don't see this as absurd. One percent in one component category is not worth 1% in another. The value and frequency of the event are the important factors for determining whether or not the spread of results is too great.

Assuming an otherwise neutral ballpark, what would the BB PF have to be to increase the Run scoring PF by 1%? Well, according to linear weights, a walk is worth 1/3 of a run. Assume a 700 run / 162 G environment with a 3 BB/9IP league average. That's 486 walks a year per team. So, 243 home walks. To increase home run scoring 1% (3.5 runs), there would have to be 10.6 more home walks. So that's a 4% increase in walks to create a 1% increase in run scoring.

Given how few walks are necessary to make a big difference in the park factor, it may be that the data are too noisy to measure the effect of parks on walk rate. That would be hard to judge without seeing the year to year park factor data. At the same time, it also means that very small differences in ballparks could very well account for big differences in their home walk rate.

Posting for GGC (who can't due to connection issues):

What should be readily apparent is that Tom Glavine is not the same pitcher when runners are on base that he is when the bases are empty. When runners are on base, he nibbles more, gives up more walks, but refuses to leave the ball over the plate. Any model that uses a linear weights average for the HR weight will totally misvalue him.

Amen. For all it's warts, regular old ERA still has some value. Component ERA, DIPS ERA, et al fail to capture how a pitcher changes his approach.

#11 and 12: Surely there's some middle ground between not participating at all and a 50% post rate, no? Must everything be so black and white?

That being said, welcome, SABR Matt. I hope the discussion is fruitful.

I notice that the team with the 1.27 walk factor is a short season team. Even over a 3 year period, thats a small sample, only 200 games or so.

Could this just be random variance? If you've got 200 or so teams, and some of them play short schedules, even assuming there is no real ballpark walk factor could you by random luck alone find one team with a 1.27?

Its possible that even if there is a real effect here, the effects would be less in the majors, where something like lighting or the makeup of the mound is more uniform.

Anyone know if players walk more in day or night games? Its possible the makeup of the schedule is at work here in the minor league numbers.

For what its worth, I ignore walk park factors in my MLE's and projections.

Here are glavine's splits since 1987 with bases empty and runners on

Situation IP H BB K HR
Bases Empty 2051.1 2367 628 1475 223
Runners On 1673.0 1606 760 975 96



What should be readily apparent is that Tom Glavine is not the same pitcher when runners are on base that he is when the bases are empty. When runners are on base, he nibbles more, gives up more walks, but refuses to leave the ball over the plate. Any model that uses a linear weights average for the HR weight will totally misvalue him.

FWIW in his last few years, Al Leiter compensated fopr his declining stuff by doing just that, nibbling at the plate, walking more batters, but refusing to leave the ball over the plate. His pitch counts became extreme- even when pitching well he couldn't get out of teh 6th inning- but it was an effetive approach for awhile- but the decline in his stuff became so extreme that any pitch he threw near the strike zone was hittable- and that was that.

Glavine on the other hand- has done this for years...

I noticed that about Glavine sometime around 1992. Obviously its a sign of decline and you can't pitch like that forever. Glavine probably doesn't have more than 3-4 years left of this.

Interesting split there regarding Glavine.

You can point out issues and I can recognize that is one of them...however, what would you suggest be done about that? For pitchers who perform different in the stretch and in the wind-up...how would you have me rate them? It seems to me that if you're a different pitcher in the stretch and wind-up...that's a part of who you are and what makes you valuable (or not)...are you suggesting that Glavine is prone to sudden collapses? Perhaps a bifrocated markov chain approach...two different sets of values for two different situations? When bases are empty one set...once someone reaches another? That would be extremely difficult to implement and I'm not sure it would result in a different net pitcher value when you add it up.

On the issue of park factors, Tango did me a favor and pointed out that for real major league teams with significant playing time, the BB factor range is .90 to 1.08 which is a much different statement...and the vast majority of teams were between .96 and 1.04.

I still hate the concept of ratio park factors...it makes absolutely no sense to claim that a park will impact the rate of a specific event occurring more the more that event occurs. Parks impact you the longer you play there...not the more you do in any one compnent. I am nonetheless pursuaded that I need to look at park BB effects and park K effects though I'm guessing that will once again be a VERY minor change in the results.

This is good though...picking up a lot of ideas on how to refine my approach...that's why I went public with the methodology.

Matt, I agree it's difficult to account for this. Don't you agree though that applying the average linear weight value to Glavine's "should have been" homers allowed seems wrong, given the above numbers? He allows roughly half as many homers with runners on as he does when the bases are empty. Thus, the average HR value will be lower for him than for other pitchers. Also, I'm not sure how fair it is to attribute more home runs to him (if you indeed do - i'm not sure) based on his being DIPS lucky and pitching to more batters, because (still in the "would have been" world) he would have pitched differently had those guys been on base.

Any methodology that smooths BABIP will have more predictive utility than ERA. Without looking at really granular data, though, I simply don't have confidence that such a metric is giving the right credit to the right people. IMHO, many pitchers who have really low BABIP seasons, in addition to having help from their defense, probably also experienced high BABIP talent that season for whatever reason. Their stuff was just a little better, or something like that. I'm not sure how to quantify it. But I've been to plenty of games, and I've seen Mike Hampton throw an 8 inning gem with very few strikeouts and lots of outs in play; and damn it, the defense wasn't playing all that well, hitters were just not making solid contact. These types of methods won't be able to acknowledge that.

Also, to finish a thought, there are other methods that have higher predictive utility than ERA (such as looking at strikeout rate, GB rate, walk rate, HR/OF, park, etc), and I don't know that any DIPS based stat will be better than using all of that data for projecting forward.

It seems like you're interested in assigning retrospective credit, which is a noble goal but very difficult to do properly when constrained to the retrosheet dataset.

Remember, batted balls in play become hits at different rates when the bases or empty or when there is someone on base. BABIP last season was .290 with no one on, .309 with runners on base. If you're going to start comparing a pitcher's singles allowed rate to his team's and league's, it may be worthwhile to break out his performance in those two situations and compare them separately.

You find that in 2001 Glavine allowed 2.3 more singles than his team's rate. With no men on, the Braves allowed .222 singles per BIP, with men on base .243. Glavine allowed .252 1B/BIP with no one on base (12 more than his team), but .200 with men on base (12 less than his team). He ended up even, if you account for the situation -- and, more importantly, the distribution of Glavine's singles allowed minimized their run impact.

Now, I really have no idea if that's repeatable or not. Of course, whether or not you want DNRA to be prescriptive or descriptive is another question -- if you just want it to be descriptive, it doesn't really matter if it's a repeatable skill.

I still hate the concept of ratio park factors...it makes absolutely no sense to claim that a park will impact the rate of a specific event occurring more the more that event occurs.

I don't know, that means a hitter who hits a lot of deep fly balls will take more advantage of a homerun park than a slap hitter. A guy who works deep counts will take more advantage of a park conducive to walks than Jeff Francoeur.

I'm open to looking at other methods of measuring park effects, but the standard one seems pretty intuitive to me.

Just to correct an apparent misconception...DNRA doesn't "smooth BABIP" aside from attempting to screen out the team's input on each pitcher's BABIP. I've preserved all of the differences between team rate and pitcher rate.

Now I'm open to this idea about breaking all pitcher data into men-on-base and bases empty, computing two separate LW groupings for those events and comparing pitcher performance to his team's in each of those groupings, because men on base definitely changes the defense's ability to prevent hits and also raises the stakes. That's a good point to start from.

Now I'm open to this idea about breaking all pitcher data into men-on-base and bases empty, computing two separate LW groupings for those events and comparing pitcher performance to his team's in each of those groupings, because men on base definitely changes the defense's ability to prevent hits and also raises the stakes. That's a good point to start from.

If you are going to start making this type of situational adjustments you might as well go all the way and just use the change in run expectation for each event. Since you are not trying to predict future performance, but only measure past performance, the change in run expectation is the only measure that weighs performance in its proper context. I guess to get an accurate run environment adjusted number you would have to adjust the run expectation values from the league average to values that reflect the run environment of the pitcher's stats.

The problem there MrD is that you lose contact with the point of this statistic (that being to effectively screen out the influence of a pitcher's team defense). Pitching from the stretch is a skill that can be examined without going all the way down to situational analysis, which loses the main objective of this study. The point of this metric is to get as precise as possible a reading of exactly how the pitcher would fare with neutral luck, neutral defense, and neutral strength of schedule (which BTW is another thing I'll be adjusting for as soon as I am able...I need to properly rate hitters in context-neutral setting to rate how skilled the hitters each pitcher faced were).

Matt - I agree with the neutral defense and neutral strenght of schedule parts of your objective but I don't think you should adjust for neutral luck. At least not at this stage of your project. What people call luck is caused by either small sample error OR factors that haven't been accounted for yet (like Glavine's strategy of pitching differently with men on base). If you attempt to filter out luck you may actually be filtering out differences that have real causes and hence be losing an opportunity to make some real insights.

As far as making a change in run expectation stat defense neutral, you can still calculate how much better the Atlanta defense was than the league average as a percentage of runs saved and apply that to Glavine's change in expected run totals AFTER they are calculated, instead of trying to correct for Glavine's expected events with a neutral defense and then using those to calculate his run totals. That keeps in any information about real differences that Glavine may of had about how he pitched differently in different BaseOut situations, but still has a factor for how much he was helped by his defense.

That strikes me as the same kind of choice that leads to RA-based metrics (such as RSAA)...the central assumption there is that the pitcher is an entity that is effected by an outside satellite (his defense). I see the fielders and pitcher as far more symbiotic than that...I believe you should start with RA or situation-dependent things like Run-Expectency Added and then adjust for defense becasue I think defense permeates every aspect of pitching (aside from the DIPS elements). Do you suppose Glavine would put the ball in play as much with men on base if he didn't have confidence in his fielders to make the plays? He learned that style back in the early 90s when the Braves were among the better defenses in baseball year after year, and it stuck..it betrays him today and now he's giving up more hits on balls in play because of it.