THE BOOK--Playing The Percentages In Baseball

I don’t have any quarrel with the hypothesis that certain pitchers are more suited to starting or relief and that the 1 run difference or so in ERA we see between when a pitcher starts and relieves may be different for different pitchers or different types of pitchers, but I see nothing in this article that suggests anything whatsoever.

First of all, he tells us that Tavarez has had many starts and many relief appearances and that the difference in his ERA in both roles is like 1.30.  So why are we even suggesting that he might not suffer the typical penalty when starting?  He seems to fit the typical profile of the pitcher who pitches 1.30 runs per 9 worse as a starter than as a reliever.

The fact that he pitched better in a starting role this year for 30 innings is so inconsequential, he should be ashamed to even mention it.  Simply look at his entire career as a starter and as a reliever.  If there is little difference and we have large sample sizes for both starting and relieving, then maybe we can talk.  I really don’t understand why he chooses Tavarez as the poster boy for this theory. Shouldn’t he at least choose a pitcher that does NOT have a large career differential between starting and relieving?

Is that 1.30 run differential including this year?  Even if not, how much would this year’s 30 IP change his career differential?  Is there any suggestion that he is a completely different pitcher or starter this year than he was in the past?  I don’t think there is any mention of that in the article.

OK, he looked at the “comps” and what did he find?  I’m really not sure.  Maybe I didn’t read the article closely enough, but did he find that among the comps that their starter/reliever differential was substantially less than the average pitcher (1.0 runs per 9 or so)?  I don’t think so.  Isn’t that the first thing you want to do when testing the hypothesis that a certain type of pitcher does not have the typical “reliever penalty” - get a large sample of those types of pitchers who have pitched both in relief and as a starter and look at their differential?

He found that of the 21 pitchers 9 were starters or something like that?  That does not suggest anything to me.  Even if that suggests that these types of pitchers are more suited to starter than reliever, that has NOTHING to do with how much of a penalty they would have going from relief to starter.  Any pitcher who is above average (in pure skills) or part of a group that is above average will tend to be a starter.  Starters are much better pitchers than relievers.

And I have no idea what he means by this:

The starters and the relievers both had ERAs 1% better than average (just about Tavarez’s career average, by the way, which he equaled in 2006), lending more credence to the idea that Tavarez might experience less of a penalty moving into the starting rotation (since these two groups of pitchers are supposed to be similar in ability).

Absolutely no idea.

I don’t get the point of this article at all.  Maybe I am missing something.  DSG is usually brilliant and sensible, but this article, I don’t know.  Maybe he is starting to suffer from “Joe Sheehan syndrome....”

I think MGL is being too harsh.  David was very clear that this was only a speculative look (by a hopeful Sox fan?), and that it was more likely than not that Tavarez would turn out to have a traditional starter/reliever split. And he couldn’t be more clear that he understands that 30 innings doesn’t prove anything—hell, that’s WHY he does the rest of the analysis.  No need for “shame” here (unless you want to accept some, for needlessly piling on).

I thought looking at the pitch ratings was the more interesting part.  (Treder’s low-HR pitchers actually had a typical starter ERA penalty overall).  What he’s arguing is that this pool of 21 pitchers with this specific skill set, and who presumably have equal underlying skills, performed equally well as starters and relievers.  If true, that would be very interesting.  (You couldn’t possibly calculate a reliever penalty for such a small group, unless by chance all had pitched signif innings in both roles.)

Now, I don’t know if this is valid or not, and I don’t think David does either.  You’d have to see if pitchers with similar pitch ratings really do seem to perform at very similar levels (i.e. have equal overall pitching skill).  And, we’d need to be sure that the pitch ratings are not themselves influenced by the role—i.e. maybe a 5 FB pitcher in relief is only a 4 as starter.  So in the end there may not be any important information.  But I credit David with trying to explore this data to see what we can learn from it.

I agree that looking at the scouting information is critical.  We won’t know if a guy is better suited for one role or another unless we start off with a huge sample size in both roles, which of course we won’t have for almost all cases.  So, hitting the scouting information is a necessity.

I also agree; I thought David’s analysis was interesting.

The advantage for relievers is theorized to derive from being able to throw harder and not to need to mix in their worse pitches as much because they don’t face the same batter twice in the game very often.
It seems reasonable that a pitcher without a plus fastball and who throws a few pitches of fairly equal quality might not be able to take as much advantage in the relief role. I agree that the use of pitch ratings was intriguing; although subjective, it’s a direct look at a relevant factor in success as a reliever vs success as a starter.

Part of the gap between Tavarez’ career numbers as a starter and as a reliever is skewed by terrible starts at the very beginning of his career in 1993-94. He had 4 more (terrible starts in 1996) and then had some regular starting opportunities in 2000-2002. Looking at 2000-2006 only, the difference in Tavarez’ ERA as a starter and reliever is about .70

I don’t think I was being too harsh.  I see nothing of value in the article other than, “Gee I wonder if Tavarez or other pitchers like him (or perhaps certain types of pitchers in general) might not suffer the typical “reliever/starter penalty?  But either I can’t or won’t try and figure it out.” End of article.

There are many silly things in the article as well.  For one thing, I can’t stand the, “Well, I know that 30 IP means nothing, but...” David is above that.  As I have said many times, he is a brilliant young analyst and will someday dethrone all of us.  This is not one of his finer pieces to say the least. If some of you guys think it is “interesting,” good for you…

I had the same thoughts as MGL while reading the article this morning. The underlying idea is certainly worthwhile, but the selection of Tavarez as the focal point, and the meager study involved, did not impress me at all.

OTOH, D Gassko tries to be somewhat prolific in the frequency of articles he contributes. So, given that role or responsibility, we should be forgiving of an occasional cluncker.

Mickey, it depends on what you want from an article like this. If it’s any sort of iron conclusion, you’re not going to get it because there isn’t enough data to support anything but the null hypothesis. My question was: Is there data to suggest the null hypothesis is wrong? That question might bore you, which is fine, but it happens to interest me. But I will respond to your comments, because I think you’re mis-understanding some of the article.

Is that 1.30 run differential including this year?  Even if not, how much would this year’s 30 IP change his career differential?  Is there any suggestion that he is a completely different pitcher or starter this year than he was in the past?  I don’t think there is any mention of that in the article.

It does not include those 30 innings. This was actually a poor job of research by me—Eric Van looked into it further and found that if you ignore (a) Tavarez’s rookie season, (b) Some spot starts in ‘94 and ‘96, and (c) 5 starts sandwiched around a DL stint in 2002, all of which are reasonable if arguable decisions, Tavarez has been much better as a starter, about 7% better than average (versus about average as a reliever). And in the past five years, he has been significantly better each year starting than relieving all starts included. Again, those numbers are not definitive and you could argue them to death, but they are suggesting something.

OK, he looked at the “comps” and what did he find?  I’m really not sure.  Maybe I didn’t read the article closely enough, but did he find that among the comps that their starter/reliever differential was substantially less than the average pitcher (1.0 runs per 9 or so)?  I don’t think so.  Isn’t that the first thing you want to do when testing the hypothesis that a certain type of pitcher does not have the typical “reliever penalty” - get a large sample of those types of pitchers who have pitched both in relief and as a starter and look at their differential?

There’s not a large enough sample to run that kind of comparison. Even if I were to increase the sample a lot, there probably wouldn’t be enough pitchers, and I’d be including guys that weren’t really comparable to Tavarez. Luckily, since the pitchers are all comparable, we can assume that they have equal abilities and then simply compare their performances as starters to their performances as relievers. And what we find is that their performances are equal.

He found that of the 21 pitchers 9 were starters or something like that?  That does not suggest anything to me.  Even if that suggests that these types of pitchers are more suited to starter than reliever, that has NOTHING to do with how much of a penalty they would have going from relief to starter.  Any pitcher who is above average (in pure skills) or part of a group that is above average will tend to be a starter.  Starters are much better pitchers than relievers.

See above. These startes *shouldn’t* be better pitchers, according to the scouts.

And I have no idea what he means by this:

Again, see above.

I don’t get the point of this article at all.

The point is to ask, “Is there any evidence that suggests (SUGGESTS) that Tavarez might make for a good starter?” And the answer is yes. The statistical answer is that his ERA will be .80 runs or so worse as a starter than it would be as a reliever. But there are some reasons to believe that he will not incur such a harsh penalty, and maybe no penalty at all.

Whether that evidence piles up to a hill of beans, I don’t know. It’s up the reader to decide that. If you don’t believe that it does, that’s fine. Frankly, I just don’t know the answer, but I thought it would be worthwhile to throw all the information I had out there.

"Luckily, since the pitchers are all comparable, we can assume that they have equal abilities and then simply compare their performances as starters to their performances as relievers.”

Maybe.  I think this would be more convincing if you did a larger study showing that similar scout ratings = similar performance.  Also, could it be that the scout ratings are for Tavarez as a reliever?  If he has a “6” FB as a reliever, can he maintain that as a starter or does it become a 4 or 5?  Maybe a “6” FB from a starter is not comparable, but instead signifies a better pitcher.

Maybe.  I think this would be more convincing if you did a larger study showing that similar scout ratings = similar performance.  Also, could it be that the scout ratings are for Tavarez as a reliever?  If he has a “6” FB as a reliever, can he maintain that as a starter or does it become a 4 or 5?  Maybe a “6” FB from a starter is not comparable, but instead signifies a better pitcher.

***

It’s certainly an issue. It’s an assumption and not a fact.

And by the way, not that I think this is a “clunker,” nor do I mean to say that I should be let off the hook should I write one, but please do remember that I’m currently writing about five-to-ten thousand words a week of baseball analysis. I give my best in all of it, but not everything can be groundbreaking.

"I’m currently writing about five-to-ten thousand words a week of baseball analysis.”

Where else besides THT?

Well, I just finished a half-summer-long stint at Heater (http://heatermagazine.com/). Not sure if what I do there is that interesting to people here, but the publisher (John Burnson—he also writes or has written for Baseball HQ) has kindly consented to me putting up my articles from the past year on my blog, so you’ll get to decide, I guess.

I’m also working on the THT Annual, which comes out in November, with articles due a month-and-half before then. I’ll have three or four pieces there.

And then there are two projects I can’t tell you about yet. Both very cool, and you’ll hear about them soon.

I undestand about the quality v. quantity issue.  That does not make you immune to severe criticism of course!  That is how you and we are able to discern the eventual truth.

As far as Tavarez, maybe I misread the article, but all I saw was that his lifetime (you did not say whether that inlcuded the 30 IP, but that wouldn’t change the lifetime numbers much I wouldn’t think) differential was 1.30, MORE than the typical penalty.  That did not jive AT ALL with the main crux of the article which was whether perhaps Tavarez and/or similar pitchers might be an exception to the typical rule.  That made absolutely no sense at all.  That would be like saying, “I have a hypothesis that Troy Glaus may be a true clutch player.  Here are his splits - .800 OPS in clutch and .850 OPS in non-clutch.”

O.K., now you say that, well that 1.30 was including some funky samples.  Shouldn’t you have said something about that in the article so the whole thing didn’t appear bizzare (that you are suggesting a hypothesis on a player whose numbers indicate the exact opposite)??  And even if you did, I am highly suspect of cherry picking samples, even if we have DL involved and recency versus a while ago, etc.  Not opposed to it, just highly suspect.  Sample sizes get in our way so much of the time, taking even smaller samples and then trying to justify them is a dangerous thing to do, as I am sure you understand and appreciate…

I agree with that completely. Of course, I’m also going to use your own arguments against you now...that is, you’re the biggest proponent of not using statistics to back-up things when the statistic itself is not significant. The 1.30 vs. 1.00 difference is not—so it doesn’t really tell us anything. So the null hypothesis (Tavarez should be a run worse as a starter) stands, and that’s where my article begins, essentially. By the end of the article, my point is that the null hypothesis still stands, but is weakened by the evidence.

You sound like Ford defending the Pinto!  Just kidding…