THE BOOK--Playing The Percentages In Baseball

MGL: May 1
http://www.insidethebook.com/ee/index.php/site/comments/you_wanted_crap_yap_from_a_premium_writer/

You followed that up with great research a few hours later:
http://www.insidethebook.com/ee/index.php/site/comments/what_do_good_and_bad_starts_by_pitchers_tell_us/

We had a good talk here:
http://www.insidethebook.com/ee/index.php/site/comments/here_is_an_interesting_question_having_to_do_with_one_of_the_other_threads/

Related link:
http://www.insidethebook.com/ee/index.php/site/comments/another_article_about_when_current_season_stats_become_real_or_something_li/

***

I mentioned on the mailbag that alot of people yap, and… well, let me just requote myself:

What it comes down to is we all quantify intangibles. I’m sure you are a great worker at work, and I’m sure you do alot of little things that simply aren’t being noticed. They somehow all add value or contributions to your company. Alot of intangible things that we simply can’t properly value, but we know is there. However, you are being paid right? Presumably, someone with the exact skillset as you, but without the intangibles would get paid a bit less than you? How much less? 1% less? 20% less? That’s your boss putting a value on your intangibles, and that is you putting value on your intangibles. I’ll bet anything that a person’s intangibles will make up less than 5% of that person’s total compensation.

So, everyone was yapping about Cliff Lee.  But, when I asked on May 19 to make the readers put their money where their mouth was:
http://www.insidethebook.com/ee/index.php/site/comments/is_cliff_lee_on_a_6_1_streak_or_a_0_1_slump/

I ended up with some 80 or 90 people voting on a 4/34MM contract for him, starting May 19.

When the first article came out by Rob Neyer, he was moderate and correct in calling Cliff Lee as a slightly below average pitcher before the season, and a slightly above one now.  MGL was quite forceful that Neyer was wrong, but at least he later corrected himself.  Some Primates took quite the opposite viewpoint.

So, Neyer was right all along.  And The Book Blog readers were right all along as well.  Basically, this was a phony outrage story.  Cliff Lee was now in the Westbrook / Lilly / Silva class of pitchers, a pitcher who (should) get a 4 yr and 30-40 MM deal.  That’s the true story, and everything other than that was simply a case of small sample sizitis.

As I wrote about CC the other day, here are his numbers with all 14 starts, and here are his numbers if you take away the abysmal two 9ER starts from very early, 4/11 and 4/16:

14 GS: 91.1 IP, 95 H, 28 BB, 87 K, 4.34 ERA, 1.35 WHIP

12 GS: 84.0 IP, 75 H, 21 BB, 82 K, 2.79 ERA, 1.14 WHIP

And, on top of that, he’s been even better lately.  From May 9 to June 10:

7 GS, 53 IP, 45 H, 10 BB, 50 K, 2.04 ERA, 1.04 WHIP

He had two bad starts in his first four, but they were so bad that they inflated the rest of his numbers to the point that some people will still see his ERA/WHIP and wonder “what’s wrong??”

Since my May 19 poll:
http://www.baseballmusings.com/cgi-bin/PitcherInfo.py?StartDate=05/19/2008&EndDate=06/14/2008&GameType=all&PlayedFor=0&PlayedVs=0&Park=0&PlayerID=1636

Cliff Lee has made 4 starts, allowing 14 runs, all earned, on 22.2 innings (5.56 ERA).  24K, 9 walks, 1 HR.  His FIP puts him at under a 3.0 ERA, and his BABIP is over .400.  So, a huge mish-mash here.

Yet CC is “posting higher numbers because of the pressures of the contract situation” and Lee “has found the magical production elixir.”

This was a great article from Joe.  It’s a double-standard.  I find the same thing applies to W-L records; a 7-16 pitcher (like Matt Cain last year) will generate negative opinions in the minds of a multitude of fans, yet a 16-7 pitcher will be considered by these same fans to be too good for luck to have been a significant factor.

At the same time, you can’t show throw out two starts, after the fact.  The decision-making has to happen before you look at the data.

You can, for example, decide that you will throw out the worst 10% performances of all pitchers, and make your analysis based on that.  If of course you can show that by doing so, you can better forecast players.  Otherwise, if someone like Pedro and Maddux never had those kind of games in their heydey, you may end up shortchanging them.

You can also reweight each game, such that the better games get more weight.  Again, you have to prove that this is correct, by seeing if this correlates better with future performance.

This is all what this is: how much does this particular sample tell me about the true talent level of the player?  Is everything simply a random sample (I don’t know)?  Do more recent performances tell you more (yes)?  Do better performances tell you more (I dunno)?  Do worse performance tell you more (I dunno)?  If he is performing injured, does that mean we should discount more (maybe)?  Does being injured mean he’s more likely to be injured (maybe)?

You have to establish the answers to these questions, before you can go ahead and pick and choose the samples and decide how to weight each one.

I see no reason to throw out any samples of performance for any pitchers whether they be 2 horrendous games, 4 bad ones, one inning where they give up 10 ER, or whatever.  No reason whatsoever.  How often do you hear (usually from the MSM), “He pitched a great game other than one inning where he allowed 3 runs.” I fail to see how pitching great for 6 innings and then allowing 3 runs in one inning is any better or worse than allowing 3 runs dispersed among 7 innings.

Unless, as Tango says, you can show that this is better for forecasting players, and I can’t think of any significant reasons why it would be.  Even if a horrendous performance by a pitcher suggests some severe injury or something else anomalous unique to that day, the key word is “suggest.” Since there will be plenty of times when there is nothing wrong with a pitcher on those days when he pitches horrendously, and some pitchers will have extreme good days/bad days by luck alone, the most we would want to do would be to slightly over or underweight certain performances.  Maybe. Throwing them out or heavily discounting them has to be wrong.

The thing about Cliff Lee is that he is probably a little bit of bad example of “small-sampleitis”, as it has been shown from the pitch f/x data that, if nothing else, he has somehow added a couple of miles per hour to his fastball, and perhaps more movement on all his pitches (the increased speed may be related to increased movement).  So with him, there probably has been a fundamental and significant change in true talent.

There are no shortage of other examples where pitchers who have likely not changed their true talent level to a significant degree (of course our estimate of that true talent changes) have pithed lights out or horrendously for at least a month, if not more, at a time.

I am not basing this on anything other than conjecture, but possible one reason why throwing out an inning might make sense is if the pitcher always get shelled at the end of his starts, because he is tired. This might point to either a crappy manager/pitching coach who don’t know when to pull him, or maybe the pitcher is on a team with a bad bullpen, so he is being left in since even though he has become less effective, he still gives the team the best chance to win.

The point that I get out of highlighting the two very abd starts is to show that CC regressed back to his mean a long time ago, but that fact was masked by those two starts which inflated his season stats. We browse the stat sections, looking at the year to date stats, and think that because he has a 5 ERA that he has been pitching poorly all season, which is not the case.

I agree with Andy that it is possible that a pitcher might get shelled at the end of his starts.  And, we know for a fact that starters pitch worse each time through the order.  So, at the very least, we should be comparing the pitcher to what an average pitcher would have, had he also had to pitch to 27 batters or throw 120 pitches.

It’s possible that the forecast for a pitcher’s ERA has an extra 0.25 runs if your manager is Dusty Baker.

Andy, my point is still the same.  You don’t want to throw out those stats.  You might want to discount them, or underweight them, but not throw them out.

As Tango says, the proper way is to normalize each inning or for pitch count.  But certainly not to throw them out.

Let’s say that a certain pitcher gets stretched out more than the average pitcher such that he pitched to 10 batters after 125 pitches.  And let’s say that he has an ERA of 10.00 for those 10 batters. You don’t want to “throw those stats away!” You simply want to normalize them to the average pitcher, which would mean maybe discounting that 10.00 ERA by maybe .25 runs.  That is not going to make that much difference.  As soon as you start throwing out horrendous spates of performance for almost any reason, you are going to greatly overrate someone’s talent or performance.

Brian, to me, whether someone has pitched 3 games of 0.00 ERA and 3 games of 10.00 ERA in any order, or they have pitches all 6 games at a 5.00 ERA each, makes no difference whatsoever (other than the slight extra predictive value of the most recent games).  They both pitched badly the entire year.  Period.  Unless you are just speaking to someone in a casual and literal sense, such as, “Yeah, CC pitched badly in his first few games, and has pitched well since.” But CC has pitches badly this season if his ERA so far is bad.  I don’t care that he pitches badly at the beginning and then pitched well.  Why should I?  He has pitched badly this season period.  In fact, my projection for him is way worse than it was before the season started, and he is no longer an elite pitcher like Beckett, Peavy, Halliday, Webb, etc.

Neyer:
http://insider.espn.go.com/espn/blog/index?entryID=3447722&name=Neyer_Rob

So was it likely, two months ago, that Lee had become a better pitcher than Sabathia? No. Of course not. But was the possibility interesting enough to mention? Yes. Absolutely. And when another Lee/Sabathia Question comes up again next year, we’re going to mention it again. It’s either that, or wait until June to write about statistics.

This is an interesting philosophical question.  I enjoy reading my Sunday sports section, and the blurbs from Graziano (MLB) and Chere (NHL) of events for the week.  And afterwards, everything they said goes away.  Virtually everything they say is, in effect, irrelevant and based on small sample sizitis.  But, for whatever reason, I enjoy reading them.  I guess I need my fix.

But, Rob posits that what happened with Lee/CC was interesting enough to note, and that we shouldn’t wait until June to talk about it.

Is he right though?  Highlighting it allowed us the opportunity to do a great deconstruction, and to show that the needle moved a bit on Lee (a new 4/34MM deal in numeric terms).

I think Rob is right to highlight it, as long as what he says, and what his readers say only leads to more questions.  It’s those people who refuse to learn and accept about the relationship between samples, random variation, and true talent that is simply infuriating and leads to the useless debates we see, and the irrelevant posts we see in our Sunday paper.

As long as we have more questions than answers, I’m happy.

One reason I could never do Rob Neyer’s job is that I couldn’t come up with something to write about every day.

In the early season, I’m interested in things like whether C.C. Sabathia and Cliff Lee are pitching at new performance levels, but it’s awfully hard to come to an answer that meets any scrutiny.

So far this season, I’ve looked into four players at some depth to see why their performance was poorer than expected.  I’ve published three of those studies, but in none of them could I say that I definitely knew something was wrong. 

Jack Cust wasn’t pulling the ball in April, and he did that last year in his slumps, but what did the mean?  There were some allegations of slow bat speed, but I had no way to measure that.

Jose Valverde and Eric Gagne were not locating their fastballs very well, and Gagne wasn’t using his breaking pitches very much, but again what did that mean?  Valverde turned out fine and Gagne was hurt, and there was no way for me to know which was going to be which ahead of time.

I also recently looked at Roy Oswalt, but I didn’t publish anything from that because, frankly, I couldn’t see much difference in what he is doing this year compared to what he was doing last year except that last year his four-seam fastballs and curveballs weren’t turning into home runs.  If I continued to look, I might find something yet, but I share this to say that, even with PITCHf/x data available, it’s awfully hard to determine the reasons for changes in player performance in most cases.  If we had access to scouting information, too, or HITf/x data when that becomes a reality, we’d have a much better chance.