THE BOOK--Playing The Percentages In Baseball

All runs scoring after an error with two outs in an inning counting as unearned.  Nothing better than watching a guy get two outs, have someone reach on an error, then walk the next couple of batters and give up a grand slam, while all the runs are counted as unearned.  I don’t care how hard anyone tries, but there’s just no way to form a logical argument that a home run should ever produce zero earned runs.  You gave up a home run!

"The batter shall keep at least one foot in the batter’s box throughout the batter’s time at bat, unless one of the following exceptions applies, in which case the batter may leave the batter’s box but not the dirt area surrounding home plate”

This isn’t so much a bad rule, just one that is never enforced. Then it becomes a bad rule because having them around and not enforcing them causes trouble, if umpires can just enforce them whenever they feel like it.

David I have heard that complaint many times and of course when I compile pitching data I count everything, but I have no problem with that rule.

In the long run, it will even out.  The idea is to put all pitchers on the same plane with respect to their defenses (at least as far as errors are concerned).

The way they do it is prefect!  Just assume that at error was not made and then reconstruct the inning, which is exactly what they do.  That way every pitcher is treated the same - as if all plays that SHOULD have been made by their fielders WERE made.  No problem there. It is a logical and perfectly fair rule and method of evaluating pitchers after factoring out “error defense.” There are other ways to do that of course, which are “fairer” in the short run, but in the long run, they way they do it is perfectly fair for all pitchers (assuming that official scorers are reasonably consistent in their scoring of errors - which they are not of course, but that is not the issue we are discussing).

I am sorry, but I think it is a common fallacy that the way they compute ER for pitchers is not fair. It is. In the long run.  There are many things in the short run that are not fair of course.  So what?

As far as the batter’s box rule, while it is rarely enforced, I don’t see it as a wrong or unfair or poorly written rule…

I find any rule that isn’t always enforced unfair because it allows those responsible for enforcing it leeway to choose when they want to do so.

But yeah, there’s nothing specifically unfair about the rule I chose.  Except it’s the only one I know of that isn’t enforced.

A couple of decades ago, I read that when a team protests the game, and wins anyway, the protest is cancelled.  But if they lose, and the protest is upheld, the game is replayed from the point of the protest.

That’s unfair to the other team.  It means that when an umpire screws up and team A protests, team B has to win TWICE to win the game.  They have to win the original game, and then they have to win the rescheduled game when the protest is upheld.

Ways to change the rule:

1.  Team B can agree with Team A that the rule was improperly enforced.  In that case, the situation is immediately fixed and the game continues.

2.  Only the affected inning is replayed if the protest is upheld.  At least this gives team B a better chance.

3.  If the protest is upheld, the game is replayed from the point of the mistake REGARDLESS of whether team A won.  (So a protest could retrospectively be seen to have backfired.)

I agree with David.  If you are going to reconstruct an inning that has an error, then why not reconstruct the NEXT inning too.

The guy has 2 guys on base, there’s an error for the 3rd out, so we still have 2 outs, the next batter hits a HR.

Why can’t you reconstruct it so that the HR is the leadoff hit of the next inning?  As it stands, the reconstruction of the inning makes it so that the guy with the HR gets no PA for ER purposes.  It’s as if he never batted!

The entire “ER” thing is a joke.  It does not even out at all.  At all.  You look at Johan Santana and Brandon Webb, and tell me it evens out.

Tango: it DOES even out, in the expected value sense.  A pitcher with a crappy defense should expect the same ERA as a pitcher with a good defense.  It may not even out in the short run, because of luck, but at least the expectations are the same.

An alternative would be to count the ROE run as earned, but give the pitcher an extra .1 innings pitched.  This is still slightly unfair because a four-out inning is worth (in run expectation) more than 4/3 of a three-out inning.  Plus, the totals won’t match.

I’d be OK eliminating ER completely, and just using R.  If you charge the pitcher when your old center fielder can’t get to an easy fly ball fast enough, why not also charge him when he drops it?  It’s just that the error seems obviously unfair, but not the lack of speed.  But if you think about it, both should be equal.

Nope.

ER are biased against GB/FB pitchers.  Most errors are from IF, and that means more unearned runs for GB pitchers.  That’s why I mentioned Webb (GB pitcher) and Santana (FB pitcher).  Look up the GB and FB pitchers, and you’ll see that’s the case…

Right.  But the SAME pitcher with a crappy defense should have an equal ERA to the SAME pitcher with a good defense.  As soon as you try to charge the pitcher with the HR that wouldn’t have happened if not for the error, you screw that up, and the high-error-defense pitcher will have a higher expected ERA.

The fact that errors are correlated with GB pitchers is another reason to just eliminate ER and go with runs.  Errors happen, and if a pitcher’s style means more of them happen, his stats should reflect that.

And right, if Ozzie makes a spectacular play, why can’t we reconstruct the inning such that a hit was allowed instead?

The ER/R designation is beyond silly.  It’s an attempt to take recorded dispassionate data, and try to make sense of it by trying to make it fair. 

The job of the data recorder is to record data as best he can.  It’s the job of the data analyst to aggregate and adjust the data. 

***

Webb has: 442 ER and 75 UER (3.20 ERA)
Santana: 516 ER and 40 ER (3.19 ERA)

Sure, a guy with a worse defense will allow more runs. That’s a given.

But, more errors does not imply worse defense.  Santana pitching in front of the same defense as another Twins pitcher gives up fewer UER per run allowed simply because he’s a FB pitcher.

You may as well remove any HR that clears the fence at 301 feet, since it’s not a real HR.

It’s all a joke.  Record what you saw.  Let the analysts account for the biases and adjustments.

I agree with Tango on this, but I just wanted to phrase something a little bit differently than others have.

I think that the whole concept of the earned run is part of the reason why the close relationship between fielding and pitching is ignored or at least overlooked by many.  The ER by its very nature pushes a notion that fielding is a given, and that it is only noteworthy when it fails.

It assumes perfection when we know that there is no such thing.  I’ve never understood why I should assume that there will be no errors, and then adjust a pitcher’s record accordingly.  We know there will be errors (and as Tango has pointed out, we know that certain types of pitchers will have more committed behind them).

The attempt to seperate fielding from pitching is to be applauded, but it is terribly incomplete.  Not only is it incomplete, but worse yet it only recognizes the negative contributions of fielders.  If you had an anti-error for a brilliant play that should NOT have been made with ordinary effort, and reconstructed the inning accordingly, you would still have all of the problems that Tango has mentioned, but at least the rule would give a balanced impression of the effect of fielding.

I think the ER rule is one of the root causes of people’s failure to see how intertwined pitching and fielding are.  If every time there was an error made, the pitcher’s RA went up, people might notice that fielding affects the pitcher’s stats, and take a longer look at how that is true for fielding events other than errors.

You may not like it and it may be fair, and yes, GB pitchers are worse pitchers because infielders make errors, etc.

But, IF you are going to try and make pitcher stats a LITTLE fairer by “factoring out” or adjusting for “error defense” (and remember, for 99% of the people in the world who play or follow baseball, “errors” ARE defense), ERA is as good a way as any to do it.

If you are going to assume that all errors were caught (and again, I think there is nothing wrong with that), obviously from a sabermetric perspective, you would just subtract .5 runs per error from a pitcher’s RA at the end of the game of season, but from a practical perspective, that is ridiculous of course.

There is absolutely nothing wrong with ER or ERA and the intent is noble (not to punish pitchers whose defense makes a lot of errors).

If you really wanted to make sure that ground ball (and low K) pitchers who are going to be more affected, you simply adjust ERA for that.  One way is to “give” all pitchers a league average number of errors per GB and per FB and then adjust for number of errors above or below that.  That is ridiculous too but it solves the “problem” of GB and FB pitchers.

And yes, 301 home runs are “unfair” too and if you really wanted to get closer to a pitcher’s run prevention talent, you would adjust for that too!  You just made my point Tango.  You brought up other examples that would make a pitcher’s stats more fair!

So you want to throw out one method of making a pitcher’s stats fairer because they don’t use other methods?

So are you seriously saying that ERA is NOT a better indication of pitching talent than RA, IN THE LONG RUN? If that is true, then I will concede all of my arguments, since my ONLY argument is that ERA is a better indicator of pitching talent than RA.  That was also the ONLY intention of the inventors of the ERA rule, more or less - to try and make a pitcher’s main stat more “fair” given his defense, in terms of errors made.

So THAT is the only question which should be addressed.  Is ERA or RA, IN A LARGE SAMPLE, more predictive of a pitcher’s future ERA or RA if he were to switch teams to a random team?  If the answer is ERA, then all the arguments are moot.  If that is the case (that ERA is better), and I think it is, obviously ERA does not do a PERFECT job, and there are plenty of other things that could be done from a sabermetric perspective, but are not practical.

And, the fact that what happens after an error “changes the batting order,” or some such thing, is a ridiculously silly argument, since the change of batting order has no net effect one way or another, again, in the long run.

Bottom line is that IF YOU WANT TO ADJUST FOR DEFENSIVE ERRORS, so that everyone is on the same playing field, assuming that an error never took place and reconstructing the inning (or the game if you could - although that will NOT yield a different result in the long run) is a “clean” and perfect way to do it.  If you don’t like the idea of adjusting for error defense, for whatever reason, then OBVIOUSLY you won’t like ERA! 

The rule is a perfect way to do what the rule writers wanted to do, which was “adjust” for errors.  Again, obviously errors are part of the game and since they are, they hurt GB contact pitchers.  And if you want to, you can certainly say that that is part of what makes a good pitcher - the ability to avoid ground balls because infielders make a lot of errors.  But, the rule writers wanted to see what a pitcher’s performance would look like if no one made any “mistakes” on the field, at least obvious ones, and they accomplished that perfectly with the ERA rule!

Great discussion.  I agree with some other commenters that throwing everything that happened after the error that would have been the 3rd out is bogus.

Say there are two pitchers who:  walk the bases loaded, get 2 popouts, and then my 3B drops a popup (and, say, 2 runs scored).

Pitcher A strikes the next batter out.
Pitcher B gives up 25 home runs in a row.

How can you possibly argue that the two should be “evaluated” to be equal in terms of runs given up?

Sure, the runs that score due to the dropped pop fly should not count against me.  And the runners on bases shouldn’t either, because they’d be gone.  But I fully agree with Tango that we should “start the next inning.”

That would give pitcher A 2 runs, both unearned.
Pitcher B would have 29 runs - 25 earned and 4 unearned.

Jon, yes, that would be unfair in the short run.  So are lots of things.

That won’t happen in the long run.

If you want to adjust a pitcher’s performance for the errors behind him, you MUST do something.

In the short run, just adding or subtracting a half run or so for every extra or fewer error is the right way - that is the same thing we do with component ERA.

In the long run, re-constructing the inning is going to work out EXACTLY the same as adding or subtracting .5 runs per extra error made or not made.  Exactly.

Just like component ERA and regular ERA will match up exactly in the long run but not in the short run.  Just like RA and ERA will match up exactly in the long run (assuming an average defense, and adjusting for GB/FB ratio and contact rate).

So your “dilemma” with using RA or ERA is whether you want to adjust for errors at the same time risking the situation you describe, at least in the short run.

But you can NOT say that ERA does not represent a pitcher’s true talent minus his team’s error defense. It does.  And you certainly can’t criticize the writers/inventors of the ERA rule, who simply wanted to make sure that pitchers did not get penalized for bad error defense behind them. They know nothing about “sample size” issues, or the fact that an error costs around .5 runs in the long run.  What did you expect them to do?

I’m fine with the idea behind “you can’t assume the DP” in that, if a SS makes a bad throw to the 2B, it should be a one-out error and not a two-out error. The SS can’t control what the 2B & 1B do on the second leg of the DP attempt, so you can’t assume that those two players would have completed the second out had the SS done his job. Fine.

But the last example you give...about the 1B dropping the ball...bugs me more than any other play in baseball. Once the first out is recorded, there is no more “assuming” involved. If the pivot man makes a bad throw or the 1B drops the second throw, it’s an error. The first leg of the play is done and over with; the second throw is a separate play and deserves an error if it’s botched. It astounds me that official scorers continue to screw up something so obvious.

I’m fine with the idea behind “you can’t assume the DP” in that, if a SS makes a bad throw to the 2B, it should be a one-out error and not a two-out error.

Of course.  That is never the case, and I am certainly not arguing for that.

It astounds me that official scorers continue to screw up something so obvious.

Official scorers don’t screw that up.  They are not allowed to give an error on a DP attempt when one out is made, no matter how bad a player screws up.  There is NO justification for that.  None whatsoever.

Update:  I am surprised that no one caught this, but apparently you CAN (and should) score an error on the back end of the DP when a fielder muffs a good throw.  You just can’t award an error to the fielder who makes a bad throw after an out has been recorded.  I guess the justification is that a fielder turning the pivot often makes a bad throw because he is rushed or is slid into by the base runner.  That is not an unreasonable point of view, although I suppose you can make an argument for an error when the throw SHOULD have easily been made and it is not.

I may have to take back what I said about this rule.  It is not so bad.

Here is the rule:

No error shall be charged against any fielder when he makes a wild throw in attempting to complete a double play or triple play, unless such wild throw enables any runner to advance beyond the base he would have reached had the throw not been wild. NOTE: When a fielder muffs a thrown ball which, if held, would have completed a double play or triple play, charge an error to the fielder who drops the ball and credit an assist to the fielder who made the throw.

ER and ERA are not the same.  Arguing that how earned runs are decided makes sense because ERA has some predictive ability after 20,000 innings is… weird. 

Accounting for all runs as unearned when a pitcher gives up a home run is stupid.  The predictive nature of ERA has nothing to do with that statement.

while I have nothing to add to the ER/A discussion in particular, it does remind me of what I’ve long felt about baseball:  there should be a clearer understanding of the difference between what I’ll call rule-based stats and interpretive stats. 

rule-based stats are simple:  someone hits the ball out of the ballpark, someone crosses home plate safely, etc:  stats record these things.  but many of the statistical rules (wins, saves, holds, errors, un/earned runs and its effect on rbi) tend toward the arcane and the arbitrary.  they interpret:  they retroactively impose a narrative upon the game; they make us view the game through a distorting filter. 

obviously everybody here knows this, but it seems to me that the distinction between two types of numbers ought to be made explicitly more often.

The rule that bugs me the most is that teams can play 4.5 innings of a game, then if the game is rained out, they have to start over.  Why in the world wouldn’t they start from where the game was stopped???

#17: I never knew that was in the rulebook. Thanks for posting it.

#20, that is a good one!

#18, while the logic of the earned run rule makes sense to the average fan, you, and the others who don’t like earned runs, have a good point.

I would still like to know which one (RA or ERA) better represents a pitcher’s talent after one season or so.  I am not sure it is RA.  And, I say “predictive ability or nature” as a proxy for true talent, since that is what we are discussing - which approach best reflects a pitcher’s true talent in a neutral (for error defense) context.

MGL - I’m not sure that there’s a significant difference. I did a weighted correlation, using the harmonic mean of IP in Y1 and Y2, and all of them were right around .15 - I did ERA to ERA, ERA to RA, RA to ERA and RA to RA. I did this for pitchers from ‘55 and on who pitched for different teams in consecutive seasons. I’ll publish the results later if anyone’s interested. (Maybe I’ll throw in FIP and BaseRuns as well.)

What I really dislike about ERA versus RA is the sheer inconvenience of it. Let’s say you have a pitcher with a 4.50 ERA on a team that scores 4.5 R/G. Let’s assume for convenience’s sake that the team’s bullpen is also 4.50 ERA. What’s the expected win percentage when the pitcher starts? Not .500! It’s counterintuitive and confusing.

ERA to ERA:  0.155510938
RA to ERA:  0.147564118
ERA to RA:  0.151724278
RA to RA:  0.147117113

I don’t know if a .005 difference is worth getting worked up over. Neither of them seem to correlate especially well.

What about the rule that withholds an RBI on a GDP?

I think that if nothing else the correlations suggest that it is not “ridiculous” to use ERA because occasionally one pitcher allows 10 runs after a 2 out error and another pitcher allows none.  In fact, they suggest that ERA is better at recording a pitcher’s true talent even for one year (although the differences in the correlations are probably not significant).

And that it obviously does not take 20,000 IP for those things to even out.

I don’t have much of a problem with no RBI on a GDP.  The reason is sound.  We want to penalize a guy when he makes so many outs in an AB.  So we essentially dock him an RBI.  I have no problem with that at all.

----"We want to penalize a guy when he makes so many outs in an AB.”

Well, he is already being ‘penalized’ because he gets a GDP. And if you still want to only award an RBI when the batter ‘deserves’ it, then don’t give an RBI on single outs (like SF).

I don’t have much of a problem with no RBI on a GDP.  The reason is sound.  We want to penalize a guy when he makes so many outs in an AB.  So we essentially dock him an RBI.  I have no problem with that at all.

I don’t like all these weird discretionary aggregation techniques.  A batter reached base on error, but is considered “out” for data recording purposes, a SF where a batter makes an out is considered a non-event for batting average and slugging purposes, but is considered an out for on base percentage?  A pickoff with the runner not actually being pickedoff (the implication of a pickoff is that it’s a pickoff-out, not a pickoff-safe!), or a caught stealing where the runner is still safe ("I went to court, was caught stealing, but got off on a technicality because I didn’t get my Miranda rights read to me!").

Please.

I agree with Dave C about who cares about the correlation when it comes time for the data aggregator to aggregate data.  The Webb/Santana data shows clear bias, a bias that is found in a small subset of players (the extreme GB, FB pitchers).  Of course we’re not going to find much difference if we look at all pitchers, since most pitchers are not extreme GB, FB pitchers. 

We’re not going to find much correlation in players changing parks either, I would gather, since most park factors are under 5% (at the run level… even smaller at the component level) to begin with.  That doesn’t mean we shouldn’t have park splits.

What I really want is for a “stat” - a data point, really - to tell me something about winning or losing ball games. If two pitchers have different true-talent ERAs but the same true-talent RA, over a long enough period of time we’d expect them to be equally valuable pitchers. If we could know a player’s true talent level, we’d want to know RA.

It’s the same with saves, the difference between wild pitches and passed balls - none of it tells us anything useful. Which I guess I don’t mind in the abstract, but in reality all of that stuff is crowding out useful information.

The other thing I don’t like about ERA is that it doesn’t mean what it seems to mean. It looks like runs per game, but isn’t. This can cause problems when you treat it like runs per game.

Take two pitchers, one with a 3.00 ERA and a 4.75 ERA. (Assume “perfect” knowlege for a second - equal parks, equal defenses, equal luck, no sampling issues.) How many more runs will Pitcher 2 allow in 200 innings?

42 runs - not 39 runs. Remember, a player’s ability to give up unearned runs is very strongly associated with his ability to give up runs in general. This means that ERA tends to (very slightly) underrate good pitchers and overrate bad pitchers in relation to each other.

Okay, so it’s a sublte and very pedantic thing, but what’s the benefit of introducing errors of that sort? Or of routinely overrating extreme ground ball pitchers?

Regarding Harvey’s 4.5 inning rule, the reasoning is obvious.  There are more beer sales if you replay the whole game, rather than half.  That’s not a good “baseball” reason, of course.

I think suspended games are messy in terms of rosters, stats, etc.  Player X was on the roster when the game started, but not when it resumed?  Really?  That is more outrageous than ignoring 4 innings of wet baseball from time to time.  That is probably the baseball answer....having more suspended games is just messy, so why do it (when we can make more money by not doing it)?

Colin, I’m with you.

I don’t have a problem with the WP/PB distinction, AS LONG AS both numbers are reported for the pitcher and the catcher.  Clearly, Charlie Hough’s catchers are going to have more than their fair share of PB.  I don’t want to data recorder to try to be fair and make some of those PB as WP.  Call them PB, tell me who threw it and who was the catcher.  That’s all I need to know.  The data analyst will take it from there.

And yes, RA is our final target that we want to correlate against, just was WPA/LI is our final target (or possibly even WPA).

Re: runs vs. earned runs:

It seems to me that over a season, errors are much more random than pitcher tendencies to allow them.  So for a single-season, ERA is a better indicator of ability.

But over a career, the randomness in errors cancels out, and any differences are much more likely to be intrinsic to the tendencies of the pitcher.  So, for a long career, RA is a better indicator of ability.

Phil, you are ignoring the bias in GB/FB.  It doesn’t matter if the results are mostly random, if someone like Brandon Webb can get away with giving up 28 (!!) unearned runs one year and make his 3.59 ERA look not bad. Or Derek Lowe also giving up 28 UER also in 2004?

How about Tim Wakefield giving up 30 unearned runs!

Those are the top 3 pitchers since 1993 in giving up UER.  Derek Lowe repeats in 6th.

Sorry.  When I see a stat like UER be so heavily biased toward a certain group of pitchers, me no like.  I especially don’t like the “reconstruct this inning, but won’t reconstruct the next inning”.

Maybe someone can look at all the PA that were thrown out for UER purposes.  That is, if Webb gave up alot of hits and walks after a 2-out error, I certainly wouldn’t then snap them out of existence as UER purports to do.

Who then is the leader in “phanton UER performances”?

To highlight: people here are seriously suggesting that Webb and other pitchers performances after a 2-out error be not considered, or only half considered, as if it has LESS meaning than any other PA?

Really?  It’s one thing to suggest giving him an extra out, so that he has a 4-out inning (which you can possibly do… if you were to also give someone a 2-out inning for a fantastic fielding gem by Ozzie).  But to suggest to simply ignore the entire performance altogether?

Clearly it won’t affect the correlation much, since, how many PAs could that be.  But, it’s just as clear that you certainly can’t count it as 0, or 0.5.  It deserves the full weight.

It looks to me that this is going to be a bloody fight today.  Let me get some towels and a bucket of water, and let’s take it to the next round.

Tango, I’m with you.  I don’t care about correlations in the long-run or whatever.  I’m speaking as a baseball fan.  I just can’t see how you can throw away a potentially horrible pitching performance, just because it occurred after a 2-out error.

It’s certainly reasonable that an ERA title could be decided over this.

I’m arguing for fairness, not for predictive value.

Change the rule that allows a reliever who blows a save to get the win.  It is not so much the reliever getting the win that bothers me, but that he is taking a win away from a starter or an effective reliever who would have gotten it.

Good call.  I’ll even remove the “5 inn min” rule.  I don’t see why a reliever can pitch 1/3rd of an inning but a starter can’t pitch 4 2/3 of an inning for the win.