Jake Peavy and the Third Time Through the Order
When Game Six of the World Series kicks of tonight, Jake Peavy will be on the mound for the Giants. Perhaps the biggest question of the night, however, will be how long he stays out there. Because if you’ve read FanGraphs for any length of time, you’ve probably heard us harp on the times-through-the-order penalty. By the time a line-up rolls over a few times against a starting pitcher, there are almost always more effective relief options than letting that starting pitcher remain on the mound.
More than any other strategy suggestion, the go-to-your-bullpen-early theory is probably the biggest area where the numbers and the traditional way of managing differ. Teams generally ride their starting pitchers until they get in trouble, removing them for a reliever after a rally has started. The data suggests that managers would do better to remove starting pitchers before the rally ever starts, though this would require managers to replace pitchers who haven’t yet failed. And for the most part, they don’t yet seem willing to do that.
To see this, we only have to look back to Game Two, Jake Peavy’s first start of this series. After a poor start, Peavy settled down at the end of the third inning, and by the end of the fifth inning, he had retired 10 straight batters. Heading into the sixth inning in a 2-2 tie, Bruce Bochy chose to stick with Peavy, despite the fact that he would be facing the Royals 3-4-5 hitters for the third time in the evening. A single and a walk later, and Peavy was removed from the game; both runners would score, as would three more, as the bullpen imploded upon being asked to get out of Peavy’s jam.
Peavy struggling against batters the third time through the order tonight is nothing new. In fact, in 2014, no pitcher in baseball was worse against hitters the third time than the Giants #2 starter. From Baseball-Reference’s split finder, we see that no one was even particularly close to Peavy in terms of struggling against hitters in their plate appearance against him; his .933 OPS allowed in those situations was 63 points higher than the next worst performer.
This data even made it into the Fox broadcast, with Tom Verducci repeatedly referencing the times-through-the-order penalty in the fifth and sixth innings, and noting how this is perhaps the most dangerous part of the game for both managers. While conventional wisdom remains in the camp of sticking with a starter until he gets in trouble, there’s no question that Peavy’s personal struggles with this split will come up in the broadcast tonight, assuming he pitches well enough to face hitters a third time.
And during the live blog of the game that Jeff and I will be hosting, I’m sure I’ll be right there alongside everyone else calling for Bruce Bochy to get his relievers into the game after no more than 18 batters faced by his starting pitcher. But I can’t help but shake the feeling that the push towards this good process is being fueled an improper use of data.
Because while Peavy’s 2014 numbers against hitters the third time through the order are terrible, we’re dealing with a sample of just 255 plate appearances, hardly enough to draw conclusions about a specific pitcher’s abilities in a given situation. And those 255 PAs don’t go along with Peavy’s career track record at all. Here are Peavy’s opponents OPS and the average OPS+ for that split against him in that situation for each of the last five years:
2010: .674, 76 OPS+
2011: .735, 92 OPS+
2012: .672, 74 OPS+
2013: .729, 92 OPS+
2014: .933, 148 OPS+
Prior to this season, Peavy had been better than the average pitcher against third-time batters in every season of his career with the exception of 2004, when he allowed a 103 OPS+ in those situations. This has actually been a long-time strength of Peavy’s, and one of the reasons why he’s been a quality starting pitcher in the big leagues. Using only the 2014 numbers to suggest that Peavy has some special deficiency at pitching in these situations is actually not a good use of the numbers, as his career numbers refute that suggestion to a large degree.
While the times-through-the-order penalty deserves the greater attention it has gotten of late, and while I entirely agree with the idea that Peavy should be removed from tonight’s game before he gets to face a hitter for the third time, we shouldn’t be teaching people that single season split data of this type is why it’s a good idea. It would be a good plan to remove Peavy after 18 batters faced even if his splits this year weren’t absurdly large; the fact that he’s been terrible in this split for 250 plate appearances doesn’t validate the concept.
But, for Giants fans, you probably want Bochy to put too much emphasis on the 2014 numbers, because then he’s more likely to go to the bullpen early than if he looked at his career numbers. It’s a case where the wrong process may very well lead to the right result. Let’s just make sure we’re not promoting the poor process of using single season times-through-the-order splits as the reason for why pulling Peavy early is a good move.
Dave is the Managing Editor of FanGraphs.
Nice work.
Here’s a related split, comparing overall OPS to 3rd time through the order OPS. Peavy is even worse in this light:
http://bbref.com/pi/shareit/V9T1u
This is like the third time through the times through the order theme.
Therefore, it’s production value should be interpreted in light of the well-established “third time through the third time through the order penalty penalty,” from here on referred to as TT-(TTOP)-P
Isn’t this the third time [to the power of 3 cubed] thru this subject?
If it’s such a big deal [and I’m inclined to think it’s GENERALLY pertinent, with, to lift from Alan Greenspan, notable exceptions), you’d think there’d be some consideration for what mass of Peavy’s season on this measure is distributed to Boston and which to SF. Yes, the sample size gets even smaller [100 or so events for SF], but it might lead to some sort of explanation or at least a reasonable hypothesis for this season’s deviation from his historical record.
Has anyone done a study of the TTO penalty which controls for pitch count? That seems like a potentially confounding factor.
yes, see two thirds down here: http://www.baseballprospectus.com/article.php?articleid=22156
MGL wrote a pretty conclusive 2 or 3 part analysis last week on TTOP. Required reading to get up to speed on the topic.
To sum the article: 1) there is a third time penalty. 2) Peavy suffered from it in 2014, but SSS must be considered. 3) Although Peavy had just retired 10 in a row the manager should have pulled him and gone to the bullpen because he put 2 men on and the pen allowed 5 to score.
Not very convincing. Baseball has known forever that starters tend to tire in the course of a game. Today we call that time through the order penalty. The managers job is to determine where the flex point is between the performance of a tired starter and fresh (lesser) pen pitcher.
I remember a story about Gibson where Schoendiest came out to take Gibson out of the game and Gibby waived toward the bull pen and asked (paraphrased) “do you really think any of those guys can get this out better than I can?” – and Schoendiest went back to the dugout and left Gibson in the game.
Time through the order is not new. Yes, it affects everyone. Starters, in general, are much better than middle relievers. It’s the managers job to determine when is the best time. Stats are too general to determine when to take a starter out – the decision is never a general starter vs a general reliever.
But continue to drill down – if you can divine general trends of good/bad decisions on the part of managers, you may still find oil.
If you’ve ever played baseball at a competitive level, you know how much easier it is to hit a guy who has shown you all of his pitches in two previous plate appearances that day. That’s why pitchers will often go through the first few innings trying to avoid throwing offspeed stuff — to combat this tendency.
That Schoendist didn’t have any relievers better than third-time-through Gibson says nothing about the deficiency. We have better relievers today.
This.
In my yout’, I was among those kids who played a lot of APBA & Strat-O-Matic. For all the years I ordered full season player cards — 1963 thru 1968 — there were no [zero] bullpens that had the significance they’ve had not just this season but in the last decade or so. A good team would have maybe one guy that filled something approximating the LOOGY-ROOGY role that most teams today have filled by at least two, closers were such a rare concept they weren’t glamorized as much as long relievers [Dick Radatz was a bullpen god those years], and it really wasn’t until Bruce Sutter’s 4th season of putting up large save numbers that other teams started thinking something useful was going on with it. And that was in 1979, way after Gibson was retired. 1968, when Gibson had that 1.12 ERA, he threw 28 — TWENTY EIGHT — complete games. And the Cards closer had all of 17 saves, a typical number pre-Sutter.
How long does the effect last? Are the Royals “still familiar” with Peavy from a week ago?
I think it’s important to be cautious W/R/T bias whenever you start a sentence with “if you’ve ever played baseball…” That is pretty much the same (il)logical inferential structure that has led to the current Know-Nothings running the game.
Fatigue and the number of times a pitcher faces the order are two entirely different and independent things. I have no reason to doubt that fatigue compounds the effects of a pitcher facing the order a third or fourth time. But they are not, and have never been construed as, the same thing.
Yes, it’s true that unlike relievers, starters have to conserve their energy. But starters also have to contend with the knowledge (and hope) that they are going to face the lineup several times. Therefore, they have to be mindful of avoiding (1) pitching into patterns that the hitters will discover and use to their advantage in subsequent at bats, and (2) revealing their entire hand the first time through the order. But these things have nothing to do with fatigue or conserving energy.
Hitters process information each time they face a pitcher. They process the speed of the pitches, the release point, the spin or movement, and the way the pitcher is trying to get them out. And hitters aren’t limited to information they process when they are at bat. They pay attention to how the pitcher is trying to get their teammates out as well, and hitters all talk to one another. And then, of course, hitters have access to scouting reports and the history they have had against the individual pitcher. All of this information is to the benefit of the hitters, not to the pitcher. (Granted, the pitcher and catcher also process information on the hitters they are facing and can adjust their approach accordingly – e.g., are hitters swinging at the first pitch more than anticipated, are hitters trying to hit to the opposite field more than anticipated, etc. And obviously, the pitcher and catcher are aware of the scouting reports and each hitter’s tendencies.) Therefore, it is more difficult for a pitcher to surprise or fool a hitter the third or fourth time through the order. And this is a fact that is independent of whether or not the pitcher is fatigued. This is the TTO or FTO effect in play.
Just watch Major Leaguers take batting practice. Typically, the first couple of swings are not as successful as subsequent swings, because it takes even the Gods of Hitting a bit to adjust their timing to the pitched ball. And that’s just with batting practice where the pitcher isn’t trying to get the hitter out, as opposed to facing an opposing pitcher in a game situation.
Hitting a baseball involves a complex neurological process. We know that the practice effect applies to taking exams and playing musical instruments. Why wouldn’t it apply to hitting a baseball? TTO and FTO data support what should be intuitive.
How is it possible for fatigue and the number of times a pitcher faces the order to be entirely independent? Starting pitchers who are fatigued never get to face the order for the first time (assuming normal usage patterns).
Occurring together does not mean dependent. A pitcher could be tired for a number of reasons when he starts the game, we just might not know it. Maybe those are the games he only gets through the order once or twice? On some days or between different pitchers, fatigue may happen at different rates or because of different reasons than just pitch count. Maybe its cold, maybe its hot, maybe you had a longer inning than usual, maybe the easiest pitch for you to throw isn’t working that day making you throw other, more taxing pitches? What ever.
But TTO is always the same and happens in parallel with fatigue, but neither TTO or fatigue cause each other. Some times a pitcher is going through the 3rd time through the order at around less than 60 pitches, sometimes its 90. Certainly fatigue will be grossly different in those two cases, but TTO will be the same.
But isn’t it impossible to prove that they are completely independent because you can’t measure fatigue? I don’t think pinch counts are good enough.
Proving independence isn’t the standard though, rather you need to to attempt to prove dependence. And if you can’t, then while that doesn’t mean they are 100% independent, it does mean you can’t show significant dependence.
Its the difference between proving a negative and proving a positive.
But if they were dependent, it would be impossible to prove because you can’t measure or control for fatigue. So since you can’t prove either dependence or independence, isn’t it presumptuous to say that they are independent?
While pitch count isn’t a perfect measure of fatigue, it should be good enough. People have also looked at “high stress” pitches/innings to measure fatigue. In this very thread MGL has mentioned some of his work attempting to do such things.
I’d say its presumptuous to claim we can’t measure fatigue….
Good comment. Bob Gibson anecdotes are always the perfect response to any kind of new-fangled thinking. Or anything else, for that matter.
Anecdotes relating to current or former Cards are always the best. Period. I think there have been studies done on it, and it’s a proven fact.
Severe mistakes you’ve made in this outburst:
1. The TTO penalty isn’t primarily about fatigue. It’s at least as much about batters’ familiarity with the pitcher as it is about fatigue. Even if a pitcher isn’t particularly tired after facing 18 hitters, he’s still going to be worse having faced the hitters twice previously than if he hadn’t.
2. This is related to #1. Starters are not better than relievers, at least not starters facing hitters the 3rd time. It’s not close. Starters facing hitters for the 3rd time through this season — in 16,515 PA — allowed hitters to hit .271/.331/.429. Relievers facing hitters for the 1st time allowed hitters to hit .241/.313/.362 in 29,730 PA. And that assumes average relief pitchers and there aren’t going to be any average relief pitchers pitching in the 5th and 6th innings of a close game 6.
The notion that we can’t generalize here is preposterous based on the amount of data involved, your story about 1 single incident with one of the greatest starting pitchers in the history of the game notwithstanding. Even so, Jake Peavy, circa 2014, isn’t Bob Gibson — not by a damned sight — and I’d bet my next paycheck that whoever Bochy might turn to is far better than whoever Schoendienst might have turned to in the late 1960’s.
1. Wouldn’t it be MUCH more correct to say, “We think TTO penalty is because of fatigue and familiarity, but we don’t know enough to tease them apart”?
2. starters are better than relievers, taken as a whole, in terms of general talent. A pitcher throwing 6 innings will likely be significantly worse than a reliever going only 1 IP, even if they have the same true talent level. But as a blanket statement, starters are better true talent pitchers than RP.
It’s important to get the details right when discussing this stuff.
I think the question is also does Bochy pull out all the stops to win game 6 or be a little conservative and save some pitchers for game seven just in case? If he gets 4-5 good innigs from Peavy and the Giants have a lead, he may just lean toward early removal to win today.
When you split the 2014 data between Red Sox and Giants, does OPS partition like ERA?
Did I miss something? If Peavy has been historically good third time through, why should he be pulled after 18 batters faced tonight? Are the relief options clearly better? Do even Peavy’s career numbers represent too small a sample to override the generic third-time-through penalty? I’d like some justification for the “…Peavy should be removed…” claim.
Lex, according to MGL a pitcher’s career numbers are irrelevant because the TTO penalty applies equally to all pitchers. And MGL is never wrong. Just ask him.
I really, really, REALLY wish Dave had cited my research on TTO and stated that there is no evidence that a pitcher’s individual TTO numbers has ANY predictive value.
One of my many pet peeves is saber oriented writers not explaining how the predictability of a sample of performance or a sample “stat” depends on TWO things and not just sample size. It also depends on the “inherent” predictability of the statistic you are measuring which is another less elegant way of saying the “spread in true talent in the population.”
In these types of discussions we need to disabuse ourselves of the notion that small sample means no predictability and large sample means some predictability. That is simply not true. We need to know what constitutes small and large samples for that particular statistic.
For example, a pitcher’s own BAPIP requires thousands of BIP before we attach much predictability to it. Pitcher K rate? 100 pOr so PA. Pitcher performance on odd and even dates? Virtually no sample size is meaningful.
So how do you/we know that 255 TBF is a small sample size for predicting Peavy’s true TTOP? WE don’t without additional information and analysis which attempts to quantify the spread in “talent” among pitchers with respect to the TTOP.
Unfortunately Dave did not give you that information. He coulda have and should have so people wouldn’t have to argue about whether we should use one season or multiple seasons or what is a small or large sample for Peavy’s own TTO data.
TTOP for pitchers is like BAPIP. There is little to no true differences among pitchers and thus no reasonable sample size means anything. We MUST assume that all pitchers have the same TTOP regardless of what they have shown in the past in 1 or 5 seasons.
Like BABIP we have found that pitchers with a greater repertoire of pitches have a slightly smaller TTOP than those with only 1 or 2 predominant pitches so if you want to look at how many pitches a starter throws and use that to tweak his TTOP that’s fine.
And yes, the evidence suggests that TTOP is largely independent of pitch count. Pitchers with low pitch counts still have around the same penalties as pitchers with high counts. It is also likely that fatigue from high pitch counts does not come into play until late in the third time or the fourth time through the order.
I’m pretty that’s mostly irrelevant because Dave’s real point was that Bochy is better off switching to relievers regardless, he’d just be doing it for the wrong reason if he was basing it on Peavy’s 2014.
Thanks for the additional information and clarification. I take it that adequate sample size is not a yes/no number where there is suddenly enough data to explain everything or predict future events.
I really, really, REALLY enjoying seeing Dave Cameron subjected to treatment under the Golden Rule.
In order for you to know whether or not I am applying the Golden Rule in responding to Mr. Cameron’s article, you would have to know what I would prefer had I written the same. Clearly you have no idea.
It also baffles me why an otherwise intelligent person would respond to a lengthy substantive comment with an ad hominem attack. Oh, wait…
I smell what you’re cooking, MGL, it’s clear that the ad hominem attack is against DC and not MGL, right? Dave is frequently a condescending prick to his critics, although he denies it. You don’t deny it (at least, you never have before) and are treating him in a similar manner. I approve of and enjoy this.
It’s interesting that you say TTOP for pitchers is like BABIP. When BABIP was in its infancy, fluctuations were attributed to luck. If you had a high BABIP, you were unlucky. Low BABIP? You were lucky, and that will turn around. Saber oriented writers acted as all pitchers had virtually no control over it, and they’ll eventually wind up right around league average BABIP.
Then R.A. Dickey developed a great knuckleball, and he’s consistently posted BABIPs under .280. Kershaw’s career BABIP is .271, and he hasn’t been over .280 since his rookie year. Jered Weaver’s career BABIP is .270, and he hasn’t been over .280 since 2008. Suddenly we realize that our initial impressions were wrong, and some pitchers do have the ability to post better-than-average BABIPs.
Point being – it’s nice that you’ve put in the time and effort for the TTOP research, and you’ve helped make some nice advances in this area of baseball analysis. But just because you haven’t figured out why some pitchers might have a lower TTOP penalty, that doesn’t mean that there are no individual pitchers (or types of pitchers) who can consistently get by with a lower penalty.
In a few years, a line like “We MUST assume that all pitchers have the same TTOP regardless of what they have shown in the past in 1 or 5 seasons” will look pretty silly.
Matt Cain is another one (.268 career BAbip). Of course the comeback on all of these would be that you’re still dealing with a fairly limited sample size and that if you continue to track these pitchers through the decline phase of their careers you’ll see their career BAbip normalize.
As for TTOP, as much as I’ve been reading about it this postseason I’m still not entirely sure how to apply it. For the most part what I’m hearing is that no starting pitcher should be allowed to face the opposite lineup a 3rd time though the order, however in practice there seems to be an exception for aces even though MGL’s research shows TTOP applying to the best SPs as well. So how is a manager supposed to know exactly where to draw the line between a Bumgarner and a Peavy or a Hudson? (Okay, not really the best example because that one probably seems obvious to everyone). But it’s not always that clear and I’ve not read anyone yet try to explain what types of pitchers should be exceptions, and when.
It is a tool for more accurately projecting how a pitcher is likely to perform at any given point in the game. Like all tools, when, whether and how to use them depends…
” Of course the comeback on all of these would be that you’re still dealing with a fairly limited sample size and that if you continue to track these pitchers through the decline phase of their careers you’ll see their career BAbip normalize.”
Wouldn’t it be just as likely that their BABIP is normalizing because, as they get older, their skills degrade and presumably this includes the skill of BABIP lowering?
A (simple and dumb) way to apply it:
Take a starting pitcher’s true talent level. He is 5-10% better the first time through, even money the second time through, then declines at 5-10% of his true talent each time through the order.
80% of Jason Vargas… is pretty bad.
80% of Kershaw, is pretty good.
For the third time through, if 90% of your SP > 100% of your RP, you can stick with your SP.
Just like career arcs, players who start from a loftier peak can better weather declines.
“For the most part what I’m hearing is that no starting pitcher should be allowed to face the opposite lineup a 3rd time though the order.”
Too blunt. Sometimes the penalty is worth it: e.g maybe the situation is low-leverage so it’s okay if you leave a bad pitcher in, or maybe your bullpen is depleted so dealing with the TTOP is better than the alternative, or maybe your SP is so good that him + TTOP is still better than the reliever you’d otherwise use.
“So how is a manager supposed to know exactly where to draw the line between a Bumgarner and a Peavy or a Hudson?”
Judgment or more complicated math. It depends on lots of variables.
“I’ve not read anyone yet try to explain what types of pitchers should be exceptions, and when.”
Because it’s less about the type of pitchers and more about the situation. A bad SP in front of a deep bullpen should probably always be pulled before the third TTO. A great SP in front of a weak or thin bullpen probably shouldn’t. Situations in between are judgment calls that depend on the relative strengths of the pitchers, the importance of the situation, and the option value of delaying your moves.
So science should reject and discard every “known” theory and fact because someday that may prove to be wrong? Ok, I got it.
You also have little idea what you are talking about with regard to BABIP. You are clearly not a subject matter expert in that area and I hope the readers herein do not take your comments seriously.
The problem is that you discuss your research as if it’s fact. It’s not. We know much more about TTOP because of your research, but we don’t know nearly as much as we’ll know in a couple years. No need for you to make absolute proclamations as if you know the answers already. You don’t.
Nobody is saying to reject or discard it. But let’s not act as if every scientific theory is already a fact.
For this statement:
“We MUST assume that all pitchers have the same TTOP regardless of what they have shown in the past in 1 or 5 seasons”
The argument is not that outperforming the TTOP isn’t a measurable skill – it certainly may be. The argument is about the predictive value of past TTOP numbers. Because TTOP performance can significantly vary over pretty decently sized samples, unless you have a much larger sample (I think MGL comes down on eight seasons), you can’t with any reliability predict future performance. This is why we have sample size thresholds for almost every performance stat – it lets you know the line between ‘it may be a fluke’ and ‘it’s probably a skill’. In this case, that threshold is extremely high so we can’t assume that, given someone who hasn’t met that threshold, he will outperform/underperform the league average.
@vivaeljeter I didn’t take it that he was saying “this is fact forever”, more that he was saying “this is what the up to date research says, so if you want to believe something else that doesn’t have objective supporting evidence, feel free, but you’re acting like Joe Morgan.”
Our models will turn out to be wrong to one degree or another. That does not mean we should abandon them in favor of subjective opinion.
“All models are wrong. Some models are useful.”
viva, what in the heck is wrong with you? Basically ever sentence in that post is hugely misguided.
The problem is that you discuss your research as if it’s fact. It’s not.
Research results in facts. What do you think it is, opinion? Research is finding a series of facts that (hopefully) can be used to disprove some hypothesis and support another. Absolute proof of a theory and facts are two completely different things. The sky being blue (on a clear day) is a fact. That’s what the research is. You take X data set, do Y analysis and get Z result. In this example, Z is a FACT. The theory is the why its blue, or what would explain that Z fact. So are you arguing with MGL’s FACTS or the hypothesis they support?
We know much more about TTOP because of your research, but we don’t know nearly as much as we’ll know in a couple years.
That’s a very large assumption you’re making about knowing more, even a lot more, about something in a couple years. Progress in science is often very non-linear, coming bursts then leaving for potentially decades or centuries even. Just because progress in baseball research has seemed to be fast moving in the last 15 years or so, that doesn’t mean it will continue over the next 2. Especially in when you narrow down to a very specific subtopic like TTOP.
No need for you to make absolute proclamations as if you know the answers already. You don’t.
Absolute proclamations is very extreme way to describe what MGL is doing here. He’s telling you what his research has to tell us. And at the moment its the best available information we have. Are you saying we shouldn’t operate with the best available information because maybe one day we’ll modify the theory? (And yeah, modify it, most theories aren’t thrown in the garbage in totality, rather amended.)
Nobody is saying to reject or discard it. But let’s not act as if every scientific theory is already a fact.
If you are going to have this argument you need greater control of your language. Words in a research/scientific setting have very precise meanings. Don’t fall to level of creationists claiming evolution isn’t fact its a theory.
Walt, research results are not necessarily facts. As some would say, data interpretation is more art than science. What if you did research and found out that the average BABIP in the 90’s was .298, so you should regress every pitcher to that BABIP. While the .298 average BABIP may be a fact, your statement about regression is not a fact.
In the case of TTOP, it may be a fact that pitchers have a 10% penalty the third time through the order (I made up that number – I don’t know the actual penalty offhand). That does not make it a fact that every pitcher has a 10% penalty. And that’s really my point. I’m not saying his data was wrong, or that his research was irrelevant. Without his research, we wouldn’t know nearly as much as we know now. But (to me) there was an arrogance that came across – especially in the statement that I quoted above – as if his research is the final word, end of story.
And while it was an assumption on my part that we’ll know a lot more over the next few years, I would consider it a small assumption rather than a large assumption. As people have mentioned, it’s all over Fox in the playoffs this year, there have been several articles written about it, it’s coming up in chats. It’s a hot topic. A lot more people will be doing research, and we should know more about it.
I do like Costanza’s interpretation. And if it was written in that style, with a line like “this is what the up to date research says”, then I may have just given it a thumbs up and moved on – because I do like the work that MGL put in, and the topic is extremely interesting. But instead, he came across as “I did the research, so I have all the answers on this topic”.
To his credit, he did write “the evidence suggests” in regards to pitch counts. So that’s something.
Oh Viva…
Walt, research results are not necessarily facts. As some would say, data interpretation is more art than science.
Err, here you go again. Is data interpretation the same thing as data analysis or is data interpretation the formation of theory from the facts that come of data analysis?
And yes, research results themselves ARE. ALWAYS. FACTS. The average drug response in a population of patients is a fact. Expression level of genes in X samples is a fact. That’s what you measured its a fact. Then you go through some analysis, usually relatively straight forward statistical analysis that has one valid method of analysis for a given purpose (ie, you can’t do a Poisson test with over-dispersed data, because it fails the assumptions of a Poisson test.). These are facts. You get A data by doing X experiment and analyze by B process to lead to Y result. A, B, X and Y here are facts.
Now, the how the facts are applied to invalidate or support a hypothesis simply follows logical reasoning. Meaning we search for the best theory to explain the data or the FACTS of the research. But there maybe multiple hypotheses that can explain those facts, including hypotheses we haven’t yet thought of. It is here were you can argue with MGL. You can look at the facts and try to find something else to explain them.
That does not make it a fact that every pitcher has a 10% penalty. And that’s really my point.
Then you aren’t actually paying attention, because no one ever said all pitchers have the same TTOP. Pitchers with lots of pitches have lower TTOPs, MGL even brought that up. But what was said, was that (other than the previous exception), we can’t prove a displayed TTOP in 1-5 seasons is demonstrative of true skill. Thus, we must assume the league average TTOP when predicting future performance.
But (to me) there was an arrogance that came across – especially in the statement that I quoted above – as if his research is the final word, end of story.
I’m sorry you see it as arrogance. He’s clearly, albeit forcefully, stating the data and the logical implications of them. Get over it.
A lot more people will be doing research, and we should know more about it.
The war on cancer says hi. Lots of people (and lots of dollars) working on something hardly assures a breakthrough . A breakthrough can become more likely with more effort, but there is a lot more to it than hard work, unfortunately.
I do like Costanza’s interpretation. And if it was written in that style, with a line like “this is what the up to date research says”, then I may have just given it a thumbs up and moved on – because I do like the work that MGL put in, and the topic is extremely interesting. But instead, he came across as “I did the research, so I have all the answers on this topic”.
To his credit, he did write “the evidence suggests” in regards to pitch counts. So that’s something.
So you really don’t have anything to say beyond the fact that you didn’t like the all capitals “MUST”?
BABIP is not about luck. That’s how it is misinterpreted by so many. For the most part, a pitcher’s BABIP is about his defense making plays behind him. Fly ball pitchers often have the ability to produce lower BABIPs than other pitchers because the expected BABIP on a fly ball is considerably lower than that on a ground ball or line drive. Why? The defense makes more of those plays.
BABIP fluctuations are something like 2 parts luck and 1 part defense and environment (park, weather, umpire, etc.).
Really good stuff from Walter above. When conclusions are drawn from empirical data, we know almost nothing with 100% certainty. Now, our Bayesian prioress influence our certainty in addition to the certainty suggested from the empirical data.
When I say something like, “We must assume X,” that does not mean we are certain of X. That simply means that the research and ensuing results suggest that X is our best assumption given what we know at the present time.
If we just threw up our hands and asserted that we know very little about the way the world works because we know virtually nothing with absolute certainty, we would still be living in the dark ages.
A net, despite having many more holes than string, is still a net. Just as,k a fish.
Could someone explain the significance of TTO?
I know what it means, but I’m confused why it’s been talked about so much. It seems like an irrelevant stat to me.
See my article on baseball prospectus. Google it.
http://www.baseballprospectus.com/article.php?articleid=22156
I just came for the comments. Not that I understand a single @#%% one of them.
Third time?? Heck, First time got him!
“Never Mind”