I've done some cluster analysis professionally. Both on baseball stats and in biochem and I'm not convinced it can be done in such a way that tells you much of anything in baseball. Maybe the pitch f/x data allows you to do that, but I'm guessing there is so much gray area that the clusters would be fuzzy enough as to not tell you anything of value.

I was talking to my dad about this the other day, and wondered how Pitch F/X could change things. I like the idea of clustering, just comparing similar pitchers. I think you could also go a step further, though, and create something like a pitcher profile for individual batters. Instead of finding all the pitchers with similar stuff to the pitcher he's going to be facing, what about pulling out individual pitches that are similar? If a guy has a fastball that routinely sits in the 89-91 mph range, you could pull out the results from every 89-91 mph fastball thrown to him in the last year or two. Same thing with breaking balls, even release points.

In some ways, I think clustering pitchers as a whole, instead of individual pitch results, is a better idea, because it's closer to the context of how a pitch was thrown (I'm guessing a batter sees a 75 mph change a bit different when one guy has a 90 mph fastball, and the other 98). It just seems that there are so many more ways to look at what to expect in any given matchup now, which is pretty exciting.

"Because of these reasons, I don’t know that many Managers rely on the data..."

Really? I kind of figured the opposite was true. I guess "rely" isn't exactly the word, but at least "consider". He would know better than me, of course.

More than anything, the matchups provide a convenient reason, not necessarily a good reason, for a decision.

Wow. A bingo for DePo. It always seems to me (warning! complete speculation) that managers often use match-up stats to justify an off day for a star or at bats for a favorite veteran.

DePodesta:

More than anything, the matchups provide a convenient reason, not necessarily a good reason, for a decision.

I think it's probably closer to the truth to use "convenient explanation", rather than "convenient reason".

The MSM - and by extension, most of the fan base - tends to believe that baseball managers and GMs are telling us the real reason for the decisions they make when they are asked the "why" question. I'm not so sure that the baseball people always do that, especially when the full answer might reveal something that they don't want to have revealed - or that will put someone in a less-than-good light if they do reveal it. So they fall back on a convenient explanation that superficially makes sense, gets the press off their backs, and takes the pressure off - but which is nowhere close to the real reason for the decision.

-- MWE

Well, I am waiting for a convenient OR superficial or ANY explanation as to why Craig Counsell is the de facto starting third baseman in Milwaukee while Branyan/Hall rot on the bench.

You have a reasonable platoon without major defensive yips as opposed to a guy who cannot hit but flashes a bit more leather.

Well, I am waiting for a convenient OR superficial or ANY explanation as to why Craig Counsell is the de facto starting third baseman in Milwaukee while Branyan/Hall rot on the bench.

Here's one.

-- MWE

More than anything, the matchups provide a convenient reason, not necessarily a good reason, for a decision

Maybe not statistically, but this may make sense in a more scouting oriented way? Some hitters just don't see pitches from certain pitchers well. Wouldn't that be a good time to rest them? And isn't that leveraging hitter/pitcher matchup data?

AW Mike, that's YOST SPEAK. Does that really count??

Counsell has started the last six games, including the series opener against the Washington Nationals on Friday, with four of those starts coming at the corner. In the last two weeks, Counsell has seen his playing time increase. Even with a couple of two-hit games this week, however, he was hitting just 5 for 28 (.179) with one walk and zero runs scored in his last seven starts before Friday, when he went 0 for 3 with a walk.

“He’s been playing well,” manager Ned Yost said. “That always helps. He’s a good defender at third base and a left-handed bat. He’s been hitting the ball really, really well.[/i]

Italics begone!

I rule.

AW Mike, that's YOST SPEAK. Does that really count??

Well, that was sort of my point. See:

He’s a good defender at third base and a left-handed bat.

Branyan isn't the former, and Hall isn't the latter. While his real reason is probably closer to this:

He keeps putting Craig in the lineup, and the Brewers keep winning.

he's not going to say that publicly, because it puts Branyan and Hall in a bad light (as though they were personally responsible for the losing, which is absurd of course) So he temporizes - mention the two things that Counsell has that his other 3B options don't (so it's "clear" why he's playing), hint that he's really doing well even though the numbers aren't good (maybe just been a little unlucky with the bat), and move on to the next thing.

-- MWE

I think MWE's explanation that managers/GMs don't necessarily say what they really think when responding to media inquiries is right on the money. If they were always candid, they wouldn't properly be doing their jobs.

More Italics. Italics Rule

Instead of finding all the pitchers with similar stuff to the pitcher he's going to be facing, what about pulling out individual pitches that are similar?

The problem is that this type of analysis can be really misleading. Every pitch a pitcher throws relys in some way or another on his other pitches. It's not necessarily the velocity or movement on a pitch that makes it effective. It's the difference in velocity and movement from the other pitches he is throwing.

So I don't think that type of analysis would tell you very much...

Really? I kind of figured the opposite was true. I guess "rely" isn't exactly the word, but at least "consider". He would know better than me, of course.

I thought the same thing. Ron Gardenhire, for one, seems to use them a lot, usually with the smallest of samples. 4-for-6 career against a guy will get you the start almost all the time with Gardy.

Ron Gardenhire, for one, seems to use them a lot, usually with the smallest of samples. 4-for-6 career against a guy will get you the start almost all the time with Gardy.

The operative word here is "seems". Gardenhire may not - probably is not IMO - basing his decision on the guy's stats, but to how the guy has looked in his ABs against the pitcher. Does he see the ball well? Does he get good swings? Does the hitter have trouble with a slop guy, but handle a power guy (or vice versa)? Now, Gardenhire may USE the stats when he talks to the media - because the stats confirm the things he sees - so rather than having to launch into a description of what he sees he can simply say "well, the guy has hit him well, he's 4-6 in his career" to make the same point.

-- MWE

Also the media are much more likely to accept and not second guess the "well, the guy has hit him well, he's 4-6 in his career" line. Especially if the game is/was lost...

However, there’s something to be said for the possible utility of batted vs. class of pitcher matchups. With the advent of Pitch F/X, it’s now very possible to group pitchers together who have common arsenals. A lefty sinker/slider guy requires a different approach from a right-handed fireballer with a devastating change, and with cluster analysis, it’s just a matter of plugging in the right input to figure out which pitchers might be considered in the same class.

I have been WAITING for someone to trumpet this. I mean, platooning L vs R is one obvious cluster analysis. We have hte data that shows flyball/groundball pitchers are another. Ancecdotally, some guys cannot get around on a 95+ MPH heater, so I'd expect that may yield some results. Guys do do or do not throw big curves may be another. Get with it, guys! If some of the MLB clubs were not already making use of this info (discretely), I'd be both surprised an disappointed.

*** platooning batters who have marked flyball/groundabll tendencies against BB/FB (reverse) pitchers has been shown in multiple studies over the last 12 years to be a good idea. This has not made its way into the mainstream yet tho ***

platooning batters who have marked flyball/groundabll tendencies against BB/FB (reverse) pitchers has been shown in multiple studies over the last 12 years to be a good idea. This has not made its way into the mainstream yet tho

help me out here--this was one of the findings of some ur-sabermetician way back in the 60s--not Earnshaw Cook, the other one (can't think of his name)

the quote was something like "the vector product of a flyball and a groundball is a line drive"

Elias trumpeted the idea of flyball/groundball platooning in their Bill James ripoff series in the '80s.

-- MWE

Most splits don't really tell you anything, much less ones by pitcher. But in most cases where there's even a question about "who starts today" there's probably no harm in basing your decision on them. For example, your backup C is going to have to start at least once this week anyway. Several things might go into that decision (day game after night game, personal C, tiredness of the regular, etc.) but there's no harm done adding "hey, he's 4-6 against tomorrow's starter" to the mix. It probably doesn't mean anything but, who knows, maybe this is one of those rare cases where it does.

And what I mean by most splits don't really tell you anything is that batters don't vary much in terms of L/R and H/R splits (i.e. every RH batter hits lefties better and this season's differences in L/R splits have very little predictive power for next year's splits ... or, to put it another way, splits regress very heavily towards the mean).

Yes, all else "equal," whenever possible you want an RHB vs. a LHP and maybe you want a GBB vs. a GBP.

I dunno, maybe finer groupings of batters and pitchers would reveal some other things (fastball vs. breaking ball hitters wouldn't be a surprise).

But in the end, you're not going to be able to do a lot with it. You've only got 5 bench players, one of them is usually a crappy-hitting C and another is a crappy-hitting SS. The gap between your starters and your bench is generally fairly large. In total, you might platoon at a corner OF and maybe one more position. And then, between injuries, arbitration and FA and everything else that goes into roster management, are you always going to be able to have a good combination of L/R, GB/FB, fast/breaking, offense/defense players around?

Every small bit helps of course but we're still probably talking about getting your 4th OF 5 extra hits a year at most.

In other words, it's still the fog question -- you're trying to detect small effects. That makes it difficult statistically to detect them and, even if you detect them, it's going to be of fairly small benefit. Cluster analysis is the sort of tool that's useful for distinguishing fairly large effects (as Sean basically states in #1).

Take that guy's example. We can already reasonably accurately class pitchers into 6 categories by L/R and GB/neutral/FB. Cluster analysis might allow us to make another cut or two by, for example, secondary pitch type -- breaking/change/splitter. So now we've got 18 types of pitchers. (Heck, I haven't even got K/contact pitchers yet.) Well, it's still gonna be 3 seasons before a full-time batter would average 100 PA against each type which is still too small a sample to tell us anything useful. And, as alluded above, you mainly want this information about your bench/platoon players so the samples will be miniscule.

None of that explains starting Counsell though. :-)

Elias trumpeted the idea of flyball/groundball platooning in their Bill James ripoff series in the '80s.

I remember that--but there WAS someone before that (well before that)

can't for the life of me remember his name, though

I remember Bouton talking about this guy and his research findings and mocking the idea that the neanderthals of baseball would ever embrace statistical analysis

no harm done adding "hey, he's 4-6 against tomorrow's starter" to the mix

I don't think any manager (except possibly LaRussa) actually does it this way. More likely, as I said above, the manager relies on what he *sees* in that handful of PAs. And I would argue that what the manager sees would likely be validated statistically if we DID have a big enough sample.

When Elias was publishing the Baseball Analysts, they did a breakdown of the players who hit the most groundballs and the most flyballs over a period of years (minimum 1000 ABs). When they did it, they discovered that the player who hit the most fly balls was Gary Redus. One of the Hirdt brothers went up to Pete Rose and asked him if they knew which player had hit the most fly balls. Rose asked them "How many ABs", he was told 1000, and he answered immediately, "OK. Gary Redus." The questioner was stunned that Rose knew instinctively what they had just spent a significant amount of labor to discover - but I think that most managers really do know, or at least intuitively sense, a lot of things that outsiders like ourselves only find out when we do the statistical analysis.

-- MWE

I'd say that Pete Rose was a lot smarter about that type of stuff than many managers.

Ask Jerry Manuel which reliever is best to use against LHPs, and he won't have a guess.

I thought the same thing. Ron Gardenhire, for one, seems to use them a lot, usually with the smallest of samples. 4-for-6 career against a guy will get you the start almost all the time with Gardy.

Dusty Baker did this sort of thing a lot when he was with the Cubs. There were so many seemingly random lineups that got marched out there, and during the pre-game or post-game shows he'd say that Neifi hit pitcher X well, or Macias was 5-for-14 against pitcher Y, and that's why they got the starts. It used to infuriate me when you'd have Zambrano or Prior (good Prior) starting, backed up by a lineup featuring 3 or 4 bench guys playing.

Was that the real reason? With Dusty, I might think that it was. It happened enough that it becomes hard to write it off as a possibility.