THE BOOK--Playing The Percentages In Baseball

Didn’t someone examine that reliever’s runs allowed was the same as a starter’s first time through a team’s lineup?

"If it’s that small, then the relievers are either getting more movement on their pitches, or they are locating them much better. “

Or batters hit the same pitches worse when they come from a reliever.

Jeff: that’s not true.

JB: you are suggesting that a batter hits the SAME pitcher worse, based on whether that pitcher starts the game or not over and above whatever “stuff” he is showing?

For that to make sense, and it could, this has to happen:
1. The batter and starter BOTH wear down during the game.
2. The starter wears down FAR more than the batter
3. When the reliever enters the game, he sees a depleted batter at bat

In order to test that theory, we should look to see how relievers perform when they come in early in the game and when those same relievers come in late in the game.

I like the theory.  Now, we just have to test it.

"Jeff: that’s not true.”

Certainly part of the 80-85% difference is the “times through the order” thing. And then there is the weather issue (it is a little bit colder later in the game, which is part of “times through the order” of course).

Tango, how many runs per game is the difference between first time through the order and the average PA of the starter?

Then again, are we sure that it is 1 run?  I have always suspected that it is quite a bit less than that and that the high difference we see is due a lot to selective sampling.  How to adjust for that, I am not sure…

By the way, over the last 5 years, for pitchers with at least 100 TBF as starters and relievers, the average difference, using the delta method weighted by the min of the two TBF, per 500 TBF is:

s=1.26
d=3.0
t=.35
hr=4.32
bb= 1.36
K= -17.85

A large chunk has to come from managers optimizing opposing batters for relievers.

Couldn’t it simply be that their fastballs are only getting marginally better, and they’re throwing less of their worst pitches? If a guy has 4 pitches, and only one or two are really effective (hence the reason he’s been moved to the bullpen), couldn’t he improve by a pretty significant amount by marginally improving the top 2 and scrapping the bottom two?

"For that to make sense, and it could, this has to happen:
1. The batter and starter BOTH wear down during the game.
2. The starter wears down FAR more than the batter
3. When the reliever enters the game, he sees a depleted batter at bat “

This doesn’t have to be the case.  Maybe batters prepare much better for starters.  Maybe relievers face more favorable sets of batters (not just L/R platoon necessarily, teams might be exploiting batter tendencies in ways we aren’t aware of).  Maybe there’s some other phenomenon at work.

I just think that if increased velocity only explains 15-20% of the improvement in reliever ERA, then it’s likely that a reliever will get better results throwing the same pitch a starter throws.  Maybe someone wants to check by accounting for movement/control too.

I have relievers being about .16 runs per game better than starters through their first 9 BFP.  That would be right in line with Jeremy’s estimate of .7 MPH faster fast ball and Tango’s estimate of .7 MPH explaining .15 to .20 runs per game difference.  The question then becomes where does Tango’s 1 run per game better come from, and have I made a mistake in my calculations of .16 runs per game?

"I have always suspected that it is quite a bit less than that and that the high difference we see is due a lot to selective sampling.”

Why would dual-use pitchers tend to exaggerate the average reliever advantage?  If we assume there is a range of advantages (from zero to Mariano), we would expect regular reliever to be those who have a very large spread, so they can’t succeed as starters (which is always the preferred role). So most relievers probably have, if anything, an ever larger split.  FT starters, on the other hand, will either have a small starter/reliever spread or just be very good.  Guys who move between the roles would, I think, tend to be average (or even have a smaller than average split).  Am I missing something?

I just now realized FanGraphs has this data going back to 2002. For all pitchers with at least ten innings pitched as both a starter and reliever in the same year (690 guys), the average fastball velocity difference was 0.34 MPH. 2003 Chris Reitsma was 93.4 (70%) relieving, 89.5 (61%) starting. 2006 Chuck James was 85.6 (74%) relieving, 88.1 (64%) starting.

I also think the velocity advantage for relievers is understated here, because in many (most) cases the starts come earlier in the 3 year period.  Pitchers fail as starters and then become relievers, not the other way around.  So the starter innings come at a younger age.  For example, Dessens (one of his outliers) has 5 starts at age 36, and most of his relief innings at age 38. And velocity delines every year after about age 22.  In addition to normal aging, some of these guys probably had injuries that forced them into the bullpen role.  So, although this poses real sample size problems, you really need to compare same-season performance. 

Another issue is length of appearance.  The one-run shorthand applies to short relievers, I think.  To the extent some of these guys are pitching in long relief, the advantage will be smaller.

Tango, how many runs per game is the difference between first time through the order and the average PA of the starter?

Each time through the order is about 8 or 9 wOBA points, which is about 0.30 runs.  So, if the starter and reliever both start off at 4 RPG, then the 2nd time through the order makes it 4.3 RPG, and the 3rd time through the order makes it 4.60 RPG.  That means that the starter is an average of 4.30 while the reliever is 4.00.  The times through the order therefore explains about 0.30 runs.

The speed of the pitch seems to explain another 0.20 runs.

So, that gives us 0.50 runs.

It’s possible that the fresh reliever against the tired batter adds a bit more.  (We’ll know as soon as someone tests it.)

Add a bit more for possibly
- better location and
- better movement, and
- favorable matchups and
- something else

and if you give each of those a 0.10 run explanation per game, then voila: one run, explained.

***

MGL, you might like this:

http://www.insidethebook.com/ee/index.php/site/comments/starter_v_relief_1953_2008/

Rally already compiled all that information for us.  So, try to slice and dice it to remove the selection bias if you can.

***

The question then becomes where does Tango’s 1 run per game better come from, and have I made a mistake in my calculations of .16 runs per game

Peter:
1. Check the link above that accesses Rally’s data.

2. Are you talking about the same pitcher as a starter and reliever, or are you talking about two different pools of pitchers?  If it’s the latter, then I can certainly believe it.  You got crappier pitchers as relievers, but who have the benefits that starters don’t get.  It’s like they have a lower tax rate, but they also have a lower income level.

The fact that some guys are literally faster as starters tells us we have a sample problem, not that some pitchers have a reverse effect.  It means that they got hurt as starters and moved to the pen.  Or they (more rarely) are a young guy who improved their velocity and then got moved into the rotation.  It has to be extremely rare that a pitcher couldn’t learn to throw as hard in relief as he does as a starter.

Eric had a series of articles a few months ago:

http://www.baseballprospectus.com/article.php?articleid=9885

Did I link to this?  I’ll be disappointed if I didn’t. 

Anyway, he shows the same 0.7mph difference, but he also shows a huge 1.8 inch movement, side-to-side.  The curveball and changeup also had much more movement.

Sliders barely changed, so a pitcher who relies heavily on the slider likely won’t find himself getting much better in the bullpen.

Great article by Eric.

Sh!t.  Just checked the archives, and I missed his entire three-part set.  I’m really sorry I missed his series the first go-round, seeing how much interest I have in it.  I can only guess that at the time I only had Unfiltered in my RSS Reader and not the main site (as I do now).  My bad.

Because the fact is that the difference in runs allowed in the two roles for the same pitchers is enormous (1 run per game), and 0.7mph is going to explain about 15-20% of that.

I have interpreted this to mean that the typical starter who moves to relief sheds a run off of his RA average but *not* to mean that he becomes 1 RA more effective.

Part of this 1 RA difference (I thought about 0.3 RA but I’m not sure where I got that from) is just due to how runs are assigned when computing RA or ERA.  The pitcher who comes in for the later half of the inning is less likely to be charged with a run than the guy who starts the inning and is then relieved (if they pitch equally well).  Releivers tend to pitch the later half of split innings whereas starters are always pitching the first part.

So, basically, I thought starter--> relief means ~1 run lower ERA but only about ~0.7 ER/9 more effective.

Am I misinterpreting this?

A few thoughts/questions.

1. Tango, where are you getting the 0.7 mph = .15-.20 runs?  Is that from my “rule of thumb” that I posted here a couple weeks ago in the Strasburg/Seaver thread?  That rule of thumb was 1 run per 4 mph change for starters and 1 run per 2 mph change for relievers.  I’ve since done a comprehensive study (soon to be published at THT) and confirmed/refined those numbers a little bit, but they are basically correct.  However, I’m not sure whether the starter or reliever number would apply here to someone transitioning between the roles.

2. I’m skeptical about putting any credence in the pitch movement changes from Eric’s article.  I’d sooner believe they were from some combination of sample size/random chance and park/measurement effects than a real skill change.  But I don’t have proof one way or the other.  I’m also skeptical of assigning a blanket run value to changes in movement on a given pitch type.

3.  My studies have also found around a 1-mph difference for pitchers moving between starting and relieving.  However, this sample is fraught with selective sampling issues because pitchers are always moving from one to another for a reason and that reason is often performance related.

My article for the 2009 Hardball Times Annual was supposed to be on pitchers moving between relief and starting roles and the causes for the performances changes.  However, I found it was far too complex a subject.  There are so many things going on that could change performance that one either ended up with far too small a sample size or one dealt with data that was incredibly polluted.  Maybe this is a place for multi-variate regression or some such statistical tool that’s beyond my expertise.

I do think that it’s very easy to overlook important variables, such as temperature, for instance.  Temperature has an effect on fastball speed.  It also affects trajectories of balls in play.  I don’t have a good feel for how much temperature changes during a typical night game.

Yup, misinterpreting.

I’m looking at it strictly from the component viewpoint (wOBA).  The gap in wOBA for the same pitcher as starter and relief was 27 points in The Book (1999-2002 data) and 29 points using 1952-2008 data (from Rally).  More or less.  Don’t remember the numbers exactly.

And if you go from a .330 wOBA to a .360 wOBA, your runs per 9IP goes from 4.3 RPG to 5.3 RPG.

Yup, misinterpreting.

ok, thanks, I might have to our adjustments for guys moving between starting and relief. 

Temperature has an effect on fastball speed.  It also affects trajectories of balls in play.  I don’t have a good feel for how much temperature changes during a typical night game.

Mike, temperature is an interesting idea.  The temperature might fall by about 4 degress F from 7 pm to 9 pm.  Greg (hittracker) has HR park factors increasing by 0.26 for every degree of temperature so a 4 degree drop would only be a 1% decrease in HR’s, I think.

Guy, the selective sampling I meant was that when a pitcher fails as a starter, he tends to get sent to the pen, and when some pitchers excel as relievers, they tend to be moved into the rotation.  So even if starters and relievers were exactly the same true talent, we would see a difference in their starter and reliever stats.  How much, I don’t know. I think that is an important question, though.

I agree that any analysis has to control for age.

As far as velocity plus times through the order equaling .5 runs, I would not be surprised if that is the correct total number (the average difference for all pitchers between when they relieve and when they start) after you adjust for the selective sampling problem I stated above.

It may be true that the average FB increase in a conversion only 0.7 mph, but I have seen as much as 3-4 mph with certain individual pitchers, and I suspect the uncertainty may be greater than the observed difference.

The numbers could be off if the the sample includes SP’ers who due to age related decline or injury get demoted to the BP, where his velocity will not increase as much when looking at before-after numbers.  Or young relievers moved to the starting rotation who seek to impress in the beginning (and struggle to make it to 5IP).  There is also the noise from the speed data having as much as 1 mph difference from park to park.

My observation with SP’ers show that the first inning the velocity is lower than in subsequent innings.  It is not always the case where a SP’er loses velocity as the game progresses.  Before this happens, fatigue tends to affect the pitchers command, and it is this loss of command that gets him removed from the game.

As for relievers effectiveness relative to starters, some of it is obviously that they can throw harder having to pitch only 1 IP.  Also in part due to the fact the hitter faces the RP’er so few times during the course of the season (maybe 1-3 times), not to mention the first time in the game.  Hitters also spend more time watching video of the starter, at least in the regular season.  The Angels for example always had tough (bad) AB against Papelbon, but may have prepared more for him in the playoffs last year.

The depleted batter theory is interesting.  I suppose players in certain positions and who were on base earlier in the game could be fatigued enough to effect performance at the plate in later AB’s.  OTOH, some hitters seem to struggle earlier in the game in their first AB, and we know that pinch hitters rarely hit as well as when the same players start the game (of course, they may be hitting against an ace reliever).

From B-Ref, AL 2009

Hitters OPS

SP-1st AB 733
SP-2nd AB 780
SP-3rd AB 806
RP-1st AB 733
RP-2nd AB 869
RP-3rd AB 897

Getting 2-3 AB against RP’ers in a game suggest lower quality RP’ers are being used.

I am a skeptic on the depleted batter hypothesis, but do not rule it out for some hitters.  David Ortiz seemd to fatigue last year after his first AB.

SP-1st AB 1015
SP-2nd AB 698
SP-3rd AB 801
RP-1st AB 726
RP-2nd AB 333 (SSS)

OK, MGL, I see the selective sampling issue.  At the same time, we’re mainly trying to assess the advantage gained by relievers, most of whom don’t start at all.  As a group, they must have a significantly larger split than the dual use pitchers being looked at here—or else many would become starters!  So you can’t just regress the swing guys and think you’ve measured the advantage enjoyed by other relievers. 

Plus, you have the serious injury and aging issues in the swing sample that I raised above. These too will understate the reliever advantage.

"So you can’t just regress the swing guys and think you’ve measured the advantage enjoyed by other relievers. “

Sure, I wholeheartedly agree with that, if I get what you mean.  It is the same issue we have with many things, such as looking at the UZR difference when players switch positions and assuming that that same difference would apply to players who have not switched positions.  Or even taking MLE’s and assuming that the same adjustment/conversions would apply to players who have not played in the majors yet. 

I agree that it must be assumed that the differences for relievers must be pretty large or else more of them would be used as starters.  Although you have the issue of stamina or pitch types.  Relievers may be more valuable as a starter IF they could throw more than an inning or two and/or IF they had more than 2 pitches.\

BTW, I have thought about the notion that as a general rule starters need more pitches (obviously that is not true for all pitchers as many successful starters only have 2 pitches, more or less).  I think it goes like this.  For the first time through the order, the optimal thing is to use your one or two best pitches, more or less.  Hence relievers often do that.

Now, if relievers had to pitch the 2nd, third, and 4th times through the order as well, they would need to have more pitches, otherwise their performance would drop dramatically (which someone verified above).

So it might seem that the optimal thing for starters to do would be to use only one ore two pitches early and then add other pitches later.  But, they need to practice those pitches early in order to get a feel for them.  That is why they throw all of their pitches even the first time through the order, although they throw only one or two pitches more often early than late, I think.

And, that is another reason for the difference between starters and relievers.  Starters are throwing other pitches early in the game which are not their best pitches, just so they can warm them up for later use.

So, if we actually compared a starter his first time through the order to when he relieves (first time through the order), I think we would find this when he relieves, all contributing to a better performance as a reliever:

1) A little higher velocity.
2) A little colder temp.
3) Different pitch selection, favoring his better pitches.
4) Possible fatigue by the batters, although I doubt this, because I think it is more likely that the batters are “cold” the first time through the order and warm up later.

Whether he gets more movement or not, I don’t know.  It is certainly possible, as movement is related to velocity.  In other words, if I know I am only going to pitch for one inning, I can not only throw harder, but I can create more movement on my pitches (by putting out more effort/energy).  For example, it takes a lot of “work"/energy to snap off a good curve ball or slider.

I just found another possibility with increase in RA/9.  Relievers, like left-handed Chris Sale, faced 45% LHH hitters. If he starts, the number will be closer to 25%. He has a FIP of 2.10 vs LHH and 3.73 vs RHH. Just by facing more RHH, he will see his ERA increase.

Sh!t.  I wish I would have thought of that!  Great idea.

Ok, let’s work through it.  The platoon split is around 20 wOBA points.  That’s about a .6 or .7 run gap, per 9IP.

Now, a LH starting pitcher will face 25% LHH and 75% RHH.  So, that’s a net 50% of hitters giving us a .3 run gap.

Sale faced 45% LHH, 55% RHH, so that’s a net 10% of hitters, giving us a .06 or .07 run gap.

So, that means the changing in LHH distribution he’ll face as a starter will push his RA9 up by almost .25 runs.

We are definitely getting closer.

I will run query for the %’s for all relief and starters tomorrow.

I always thought we controlled for handedness of opponents.  Maybe that is why I get a diff of .82 rpg and Tango gets 1 or more.

A LH starter will face somewhat inferior bitters I think. The only way a team gets 75% RH bats is by starting some bench players. Not a huge factor, but it will somewhat reduce the impact of losing the platoon edge.

Thinking some more on this, the change in hitter quality could be a pretty big factor.  Sale is shifting 20% of his PA from LHH to RHH.  Who are these hitters?  The LHH are good enough to be allowed to face Sale in (often) high-leverage situations, despite having the platoon disadvantage.  The RHH are guys who sit on the bench except against good LH starters—mainly 4th OF types, a few backup MI.  I suspect that nets out to a pretty significant drop in average talent.

I just ran the numbers from 2001 to 2010:

LHPs
RHH faced at reliever: 59.3%
RHH faced as starter: 78.2%

RHPs
RHH faced at reliever: 56.7%
RHH faced as starter: 50.6%

The next thing to check Jeff would be the quality of those hitters, as Guy is suggesting.

Since the gap occurs against LHP, that’s where the focus should be.  For now, you can look at the career totals of the hitters. 

If you are interested, you would do this:
RHH v LHP, career against SP
RHH v RHP, career against SP

LHH v LHP, career against SP
LHH v RHP, career against SP

That establishes the quality of each hitter.

Your pool of hitters is those guys who faced LHP, weighted by how often they faced that LHP.

Hopefully that makes sense…