FREE hit counter and Internet traffic statistics from

Friday, April 27, 2007

Chat Today!

Just a reminder that I'm chatting today on Baseball Prospectus - 1pm Eastern. You can also submit questions early.

Update: The transcript has been posted. Thanks again to everyone who participated.


Max from Wisconsin said...

Follow up to the question you answered in the chat today. You say in Tony LaRussa and the Search for Significance, "As a result, a measure like slugging percentage or OPS would probably be a better candidate for this kind of study, although it would require using a different kind of model." I agree and then some. If LaRussa's theory is correct that specific hitter vs. pitcher match-ups are so significant that they're worth planning the game around, wouldn't that be most likely to show up in DIPS stats such as HR or K, or perhaps even ISO or SLG, where either the hitter or the pitcher shows domination over the other, rather than in BA, which is highly dependent upon the defense and not as dependent on the hitter or pitcher? Has anyone ever studied this and published the results? Thanks, and keep up the great work.

Dan Agonistes said...

Absolutely. What I had envisioned is that you create some kind of graphical representation of a hitter's output against a pitcher when you less than say 20 plate appearances. The representation would be based on hit types, strike out, and walk rates and extra base hits. Then you would not only see how successful they were but whether a DP might be more likely or a sac fly etc.

Max from Wisconsin said...

It seems to me that conceptually you could get really far down the road with a lot less effort than what you're talking about (although maybe for you going guy by guy wouldn't be that much effort; it sure would be for me!).

I'm thinking of the chapter in "Baseball Between the Numbers" where they said there was a .22 correlation for regular season and play-off stats of pitchers and a .00 correlation for hitters. And I thought, "Of course: the pitchers are facing a relatively similar level of talent as they do in the regular season, but the hitters are only facing the very best pitchers." I've spent the last several months trying to find someone else to whom this has occurred, and that is my frame of reference.

Here's how I would go about it. First, answer the question, how good are play-off pitchers? It seems like this could be done fairly easily by someone. I would say off the cuff that play-off pitching is done almost exclusively by pitchers in the top 50% of the major leagues in VORP or WXRL. Then I would look at how hitters on play-off teams hit against the top 50% of pitchers in the majors and see how THAT correlates with play-off success. I'll bet it's closer to .22 than it is to .00 and if it is, that would back up the theory that hitting against play-off level hitting is not exactly the same skill as hitting against 162-game aggregate major league pitching.

What I just talked about was "tier-based performance," that being that hitters run up against a talent level of pitcher above which they experience a severe drop-off. What you and LaRussa are talking about is "match-up-based performance" where an individual can have another individual's number, so to speak. I think one is macro and the other is micro, one can help predict how well a team will do in the play-offs and the other can help predict how well a specific hitter will do vs. a specific pitcher. In both cases, I think we agree that BA is a weak stat to use here at best.

Another thought, and sorry to hog your blog; the thing that we're watching for here in both instances isn't an average, it's a range. If there is a player who has a .400 SLG against everyone and another player who mashes .700 SLG against major league pitching in the lower half of the talent pool but only .100 against talent in the upper half of the talent pool, which one do you want on the 40-man roster for the play-offs? Which do you want in the line-up when you're facing the Royals (or, this month, the Yankees)?

So I think you have a good approach, but maybe the first step is to more generally establish that tiers of talent do exist by looking at SLG rates and K rates first and see if there is anything there first before trying to take out the noise by going to match-up by match-up, stat-by-stat data. The study as you outlined it would be fascinating, however.

Max from Wisconsin said...

I would say off the cuff that play-off pitching is done almost exclusively by pitchers in the top 50% of the major leagues in in *EqERA* and *PERA*, is what I meant, not VORP or WXRL.

Dan Agonistes said...

I think though that based on my reading of the relevant study in The Book, it's not the case that there are tiers of performance as you suggest. The decline in performance is proportional to the quality of the hitter or pitcher. So in one sene you're correct that a good hitter suffers a larger in performance in terms of number of points in OPS or whatever measure you're looking at but that doesn't mean there are tiers.

Max from Wisconsin said...

Thank you for replying. I comprehend your point on tiers. I think classifying by both type of pitcher and skill level would yield both adequate sample sizes and interesting results, but let me spend some time studying The Book and perhaps I'll have a thought on that later in the season. From a match-up standpoint, I think re-doing your article on LaRussa but using OPS would yield more precise and easily-deciphered results.

One thought I had on judging the wisdom of LaRussa, or of any manager, is that there's a difference between their frame of reference and ours. He's forced to make a calls based on small sample sizes on every at bat whether he likes it or not, whereas analysts/fans can totally call their shots and just choose to write/read a different article. Unless you reject the idea of individual match-ups mattering entirely, a manager could reasonably think that, while several plate appearances do not necessarily establish a pattern, there is a chance that they are showing a pattern trying to manifest itself, and that if a hitter does in fact "own" a pitcher, it would show up as soon as just a few plate appearances. For example, in a pinch hitter situation, if a guy on the bench is 6-for-12 with 2 HR versus a particular pitcher, the manager doesn't necessarily have to decide that the guy DOES own the pitcher, just that the chances are better than anyone else on the bench that he MIGHT own the pitcher. He must almost always make this choice between hitters who have a smaller sample size against said pitcher than analysts are comfortable with. Hence, he may be making an intelligent managerial decision based on small sample sizes because he has little else to go on (assuming he's choosing between hitters who are fairly similar in ability overall). He can then use his powers of observation to augment that decision and may thus be making a rational decision that can perhaps never be quantifed... or perhaps they can. Never is a long time.

Keep up the great work.

Guy said...

Quick question on the log5 method you referenced: it seems to me that, at least within the ranges we usually work in, the simple formula PAvg * HAvg / LgAvg gives you virtually the same result as Log5. Are there situations where the log5 method significantly improves accuracy?

Tom said...

I completely accept that:
a: matchups matter
b: managers "see" things that others don't

What I reject is when managers form this opinion by looking at the numbers we look at. Just once I'd like to see a hitte be 3-28 against a pitcher, but a manager say: "This hitter is ideally suited to matchup against this pitcher, and that he's 3-28 is really just small sample size".

What we always get instead is "he's 11-23 against him, and he looks good against him, so I put him in".

A good scout doesn't need numbers.

That said, focusing only on say BB and K, which are the two most reliable measures (for hitters and pitchers) is a great idea.