THE BOOK--Playing The Percentages In Baseball

From the information we’re given and assuming the period of time is one season or less, they both pitched equally as well (I guess) and they both are likely to pitch the same in the future.

Unless this is a trick question, and one pitcher will be caught by Jason Kendall in the future, in which case that pitcher’s RA will be better by at least a run.

I want to say Pitcher A was/will be better, but I might need to infer some things about that terrible start. Throw it out as an outlier? That would seem to depend more on the number of starts they made.

If “stats” are outcome-related stats (ERA for example, or even xFIP), I would rather have the inconsistent guy for a number of reasons.  One being that he may have unluckily spaced out the hits he allowed.  Sort of out of his control, but a bunch of hits in close proximity make outcome-related stats look worse than they are.  A second reason being, from a win probability standpoint, you’re much better off giving up 0, 0, 0, 0, 0, 18 runs in your starts than you are 3, 3, 3, 3, 3, 3, 3, 3 runs in your starts.

If “stats” include every stat ever - GB rate, K rate, # of balks, IBBs, bases stolen off of, etc., then I think you’re no better off with one than the other.

I don’t see how this could possibly be quantified with your vague example!

Well Pitcher A was obviously more valuable, but that’s not your question really.  If we can conclude that consistency has some repeatability to it (and I don’t see why it would’t) than it’s likely that a little bit of the difference in consistency between pitcher A and B is skill.  Thus pitcher A pitched better, and is likely to pitch better in the future for the same reason.

Over the past period of time you would want Pitcher A since his team has a chance of winning in all games but one game while the team of Pitcher B has a more difficult chance of winning in all games. 

The rule of thumb is you want inconsistent pitching and consistent hitting.  (There’s an old Hardball Times article that shows this.)

Going forward both pitchers would project the same.  Objectively there’d be no difference.  Subjectively I would admit I’d choose Pitcher A.

Assuming both pitchers have a relatively large sample…

1. Pitcher A has a higher median level of performance, while they have the same mean. If he only had one terrible outing, then I think you have to throw it out and judge him based on the rest of his appearances. A single outlier shouldn’t alter results or conclusions.

2. Pitcher A should pitch better, because like I just said, a single outlier in an otherwise good sample shouldn’t throw off our entire view.

MGL I love these, even though i am usually wrong.

#1 I would say it would depend of the offensive ability of the team.  The pitcher’s job is to give its team a chance to win.  If the team is an offensive juggernaut like Toronto, the pitcher that is constantly bad may have them in one more game.  On the other hand, the a lower scoring team with the pitcher with one bad outing would just give up the one game and be in more games other wise.

#2 For the future data, there would be no way to tell which would be better unless there is some data to see if consistency is project-able from season to season

The question being asked really could jsut as well be this:

You have Pitcher X facing 500 batters, 25 at a time, and in one set of 25, he pitched horribly.

You have Pitcher Y facing the exact same 500 batters, getting the exact same results out of the 500 batters, and facing those batters 25 at a time, but in a different random sequence, and in no set of 25 did he pitch horribly.

MGL is asking if one is more predictive than the other.

I say no.

It’s not a trick question, and I don’t know the answer.  #4, 1) is wrong (sorry).  Why would you “throw out” any performance?

Perhaps certain pitchers do not “try as hard” in very low leverage situations.  Once an outing has become terrible, the fact that a pitcher continues to pitch poorly in it might have reduced predictive value (an example of pitching to the score).  Also, perhaps it is possible to identify/isolate and thus correct or at least mitigate factors that have led to one specific particular terrible outing, so as to lower the likelihood of such an outing in the future.  I know teams and coaches believe this to be true--whether it is true or not I have no idea.

But if Pitcher X’s horrible pitching is grouped together and not at random, doesn’t that leave open the possibility that the horrible pitching was not by chance? Maybe for those 25 batters Pitcher X was high on Vicodin. The other 475 he faced sober.

Pitcher Y, on the other hand was always sober and his horribleness was randomly distributed.

I think in that case, you would base any predictions on X’s 475 sober batters faced. Unless you knew that next year X would be high for 25 batters again.

This is just a slight variation of the conversation we had in New York, isn’t it?

I’d argue that inconsistent performance can be (is not always, but can be) a sign of potential for improvement.  I’d take my chances with a guy who alternates between good and bad over a guy who has never shown the ability to be more than mediocre. 

A good example might be Edwin Jackson.  A few years ago, you ran a whole bunch of posts about how awful he was, because his major league performance never matched his stuff.  But Edwin Jackson had unfulfilled potential, which was evident at times, just not always. 

Betting on Jackson a few years ago turned out to be a pretty good idea, even though he was bad overall.  His inconsistency was one of the reasons why scouts thought there was reason for optimism. 

Give me the inconsistent guy every time.

This is why we use harmonic means sometimes rather than basic means. The more inconsistent pitcher is the better pitcher, as the bad start is most likely an outlier.

Maybe, perhaps, maybe, and maybe.  And maybe I’m the long lost Earl of Cheshire…

I think Bill James wrote something about this 20 years ago.  Looking at game scores, I think he found that consistency was not a pitcher talent.

Pitcher A has been better.  Being less consistent lets the team leverage the bullpen in his outings better.  When Pitcher A sucks, you put in the mop up relief pitchers and call the game a loss.  When he’s good then you can put your best RP in to win the game.

I don’t know if you can tell which will be better in the future.  I’d guess if the player is young then inconsistent is good because its a sign that he has the stuff but needs to improve things like the repeatability of his delivery and release point.  If the player is old then consistency is probably better because you are worrying about aging patterns and wear and tear.

"Which pitcher is more likely to pitch better in the future, by how much (approx.), and why? “

I’ll go with Pitcher A as well. 

Even if I can’t point to a specific external factor for Pitcher A’s one horrible outing (and none was provided) I would not assign 100% probability that it is simply random sequencing simply because there are other possible explanations. 

The how much part depends on the probability assigned to Player A’s bad outing being chance and I won’t guess at it here without more info.

Sounds like pitcher A gave you a better chance to win more games in the past - guaranteed loss in his bad start, but higher win probability in all his other starts than B.

Going forward I don’t see an advantage for either one.

Echoing what Rally/#16 said. Given an average number of runs scored by each pitcher’s team, Pitcher A had a guaranteed loss in one game but gave his team a better chance to win in each other game. Going forward, however, I’d have no idea, so I’m inclined to say even.

pitcher A is VINCENTE PADILLA!

I put some numbers to the assumption.  Each pitcher gives up 100 runs in 20 starts.

A allowed 4 runs in 19 of them, and 24 in the other.  His team scores 4.5 runs per game, so using the pythagorean formula he goes 11-8 in those games, and loses the stinker for a 11-9 record.

Pitcher B gives up 5 runs every time, and his final record is 9-11.

As far as question #1:

“Which pitcher “pitched better” for that period of time?  I realize that that may be subjective. “

I don’t think there’s enough information.  The example never stated whether the pitchers were starters or relievers and never indicated how many outings were covered.

Based on the information given: 1 bad outing evening things out then a realistic example would be, perhaps, 25 IP and 13 ER as relievers (with Pitcher A giving up 7 of those in one outing and both Pitchers appearing in low-leverage situations since Pitcher B has been consistently bad) or 30 IP and 18 ER over 5 starts (with Pitcher A tossing a 2 IP 6 ER game).  Not a lot to go on there.

I’ll have to go with pitcher A, though reluctantly. Two reasons:

1) While we have no reason to assume that pitcher B is better than his overall stats, we have some reason to assume that pitcher A is better than his overall stats.

2) The actual variance in performance is helpful to my team, assuming that pitcher A keeps it up. To make an extreme but simple example, say that pitcher Alpha pitches twice and gives up 6 runs both times, while pitcher Beta pitches twice and gives up 1 run the first game and 11 runs the second. Which pitcher likely has the better winning percentage? That’s right, pitcher Alpha, since pitcher Beta likely lost both games while Alpha split.

Er, pitcher Beta has the better percentage since he split while Alpha tanked both of them.

Can we just cut Kyle Kendrick and Jamie Moyer and trade for Roy Halladay?

Ditto the information complaints. I think things have been covered pretty thoroughly. I assumed that these guys are starters. As Rally pointed out, pitcher A should give you a better chance at winning more games. Honestly, I’m reminded of Brett Myers, who puts up streaks of utter dominance mingled with terrible 3 game slumps. Then he goes on being decent. Not exactly the same, but that’s what my mind correlates to Pitcher A. Pitcher B sounds like Matt Belisle. Put him in the bullpen and hope he dominates for a season.

#23, actually I think that the general rule of thumb is that if your pitcher is worse than average, you want more variance and if your pitcher is better than average, you want him to be consistently good.

Now, for one really bad outing, of course, the value of that pitcher is better than his overall stats.  So I assume that pitcher A created more theoretical value than pitcher B. Whether you want to call that “pitching better” - well - that’s up to you. I don’t like words which have no precise meaning when the discussion is about quantitative things.  They don’t usually forward the discussion.

As far as going forward (which pitcher has the better projection), I don’t really know. I assume that both are about the same.  Is “consistency” a talent? I don’t know. I doubt it - to any degree that is (it might be tied to propensity to get injured).

Almost any trait is a “talent” (for example, BABIP for pitchers).  The question is what is the variance of that talent among players?  In most cases like this, it is small.  Too small to be of any practical significance.

It’s not a trick question, and I don’t know the answer.  #4, 1) is wrong (sorry).  Why would you “throw out” any performance?

I didn’t mention throwing out any performance, you must have been referring to #6.  David Gassko showed here that inconsistent pitchers will win more starts than consistent ones:

http://www.hardballtimes.com/main/article/same-old-same-old/

And Rally’s more relevant experiment in #21 also confirms that.  So for question 1, pitcher A was more valuable.  I guess that means he was better, but like you said it’s subjective. 

For question 2, I would think that like with most things in baseball, the guy was was inconsistent in the past is more like to be inconsistent in the future (if only slightly).  This could be because of things like erratic mechanics, or being a knuckleballer or whatever.

FWIW, BaseballHQ published a study this season showing that consistency IS a skill (ie, a skill with significant variance) for batters.

I take the guy with one horrific outing for next season. Maybe his back was tight that day.

"FWIW, BaseballHQ published a study this season showing that consistency IS a skill (ie, a skill with significant variance) for batters.”

As I said, almost everything like that IS a skill.  The pertinent question is how much of a certain “split” represents skill and how much represents random fluctuation.  In the case of batter consistency, I would think that the “skill” is slight, and thus a batter’s consistency in one or two seasons has almost no predictive value.

Do you have the link to that citation?

Here it is:

http://www.baseballhq.com/free/free100702.shtml

I presume that’s Wexler at post 30 replying to MGL?

It seems to me that HQ simply is looking at gross totals, and not comparing to a player’s overall score.

So Pujols looks “consistent” in that, he’s really good almost every week.  Well, duh, no? 

What is being asked is if you have two guys who ARE EQUALS, but one has more dominating/disastrous weeks than the other, than will the guy with more dom/dis weeks continue to have those dom/dis weeks in the out-of-sample data?

The HQ test does not do a “all other things equal”.

T-squared, I dont’ really understand your crticisms@#31. Can you kindly try and rephrase so that I might catch up to your thinking? “Gross totals” of what vs “overall score” of what?

Macey acknowledges that it’s unsurprising that better players would have more DOM weeks from year to year.

But, then he goes on to measure the standard deviation of their performances from week to week, and finds that players who were in the lowest SD bin in year one, also had a lower SD in year two. Presumably, there are players in each bin that are NOT EQUALS in terms of talent, yet they both display their own different talent level as consistently from week to week as the other guy. Does this not answer your main objection?

Right, I think he controls for the talent of the player in the second part of the article, by dividing the variance (or SD) of the performance by the mean, but I am not sure.

I’m not convinced by the BaseballHQ article. Even if consistency is not a skill, I would still expect to see the same type of results that Macey gets.

Because he defines Consistency Ratio as (SD of performance)/(mean of performance), better players will have better (lower) consistency ratios. It’s no surprise that Pujols is #1 in this stat. Of course a player’s Consistency Ratio in year 1 will be predictive of his CR in year 2, since it’s in part a measure of his overall talent.

This doesn’t say anything one way or the other about the question Tango asks in #31.

The study linked in #30 only included players who played 24 weeks each of the last 3 years.  This sample bias eliminates all players who were injured and on the DL, and others who were not very good, so their bad weeks get them send back to the minors.  So the study had already eliminated some of the more inconsistent players before it was run.