THE BOOK--Playing The Percentages In Baseball

Wait, someone still reads her stuff?

Once a month BP publishes something so outrageous that I wish I still had a subscription so I could cancel it.

Seriously, when you work for Baseball Prospectus - are one of the masthead people, really - and you want to write that, who the hell are you bagging on besides yourself? Kahrl hasn’t done anything resembling sabermetric analysis in years that I’m aware of. Hand to God, as far as I can tell the last time Kahrl actually published “sabermetric” research was in 1999:

http://baseballprospectus.com/article.php?articleid=228

The hell.

Aha, I started to read that article the other day and after two paragraphs or so her prose just killed all of my original curiosity, saving me from that last excerpt.

The Book’s lineup analysis was good, but I have yet to see a study that quantifies the benefit from alternating batter handedness. There was a post on this blog a while ago in which MGL went after the Phillies manager for batting three lefties in a row in the middle of the order. If I remember right, the argument was basically: Giving the other team the chance to shut down your 3-4-5 batters with a LOOGY in a high-leverage situation is a bad move unless that lineup scores a lot more runs than any other order. Studies that ignore the platoon advantage indicate that there’s not much difference between lineup orders. Therefore, the three-lefties-in-a-row order is not ideal.

I buy the argument as far as it goes, but really it’s an educated guess as to how the numbers would come out if someone did a study. My point is, if you give me two proposed lineups for a team, I would not be able to say how much better one is than the other, because the factors that the current best analyses don’t take into account are just as important as the factors they do take into account.

I suppose this isn’t that big a deal, because working with the guidelines in The Book and using common sense (which would rule out the Phillies lineup) probably gets you 95% of the way to the best order. But we don’t know that for sure.

Say a team’s three best hitters are lefties, and they are batting 1-2-4. Someone says, back-to-back lefties is bad, they should be batting 1-3-5 or 2-4-6. Obviously the right answer will depend on the specific situation, but do we have any way at all of making that comparison right now?

The old MGL post is here:

http://www.insidethebook.com/ee/index.php/site/comments/does_charlie_manuel_know_what_he_is_doing/

There’s also this thread:

http://www.insidethebook.com/ee/index.php/site/comments/optimizing_the_batting_order/

Looking back at those, it seems like a lot of what I said is a rehash of the discussion on this BTF thread:

http://www.baseballthinkfactory.org/files/newsstand/discussion/daily_news_kram/

Also, I’d like to rephrase. “Studies that ignore the platoon advantage indicate that there’s not much difference between lineup orders.” should maybe be “Studies that ignore the platoon advantage indicate that for every lineup order with back-to-back lefties, there’s an almost equivalent order in which the lefties are split up.”

With the second phrasing, it’s not really “studies,” it’s MGL’s work in post #20 of the BTF thread. But even that was inconclusive, other than the easy result that there’s no excuse to bat three lefties in a row. The issue is, what does “almost” mean.

Bottom line, I think the question of how strictly to enforce the rule of alternating handedness is very much open.

Moving the pitcher from 8th to 4th costs you 0.10 runs per game (16 runs a season).

That is, making a single completely disasterous decision, that no one would ever support in their right mind, costs 16 runs.

Swapping the #2 and #3 hitter, two roles that are fairly distinct, will cause 2 or 3 runs of difference.

This is what we are up against.  No matter what you think we have not covered, how much do you think it can possibly affect something?  The pitcher thing gives you an upper boundary as to the difference between being blind and being thorough.

Other than the leadoff hitter, pretty much any decision you make is a couple of runs, AT MOST, here or there.  Usually, we are talking about one run here or there.  And this is based on 700 PA in each slot.

Even the three lefties in a row, which is foolish on its face, will only come back to bite you every now and then.  It simply makes so little difference, relative to how much energy is spent talking about it.

Study the issue, present the research, and then you will see.  Christina chose to present her conclusions first, and worse, acknowledge and ignore the research.

Give us an honest effort here.  That’s all I’m asking.  Present the evidence, and let the evidence lead to the conclusion, not your summary dismissal.

Honestly (as long as everyone is being honest in this thread), I have no idea what she is talking about.

She’s saying, I think, that sabermetrics should reflect real-life data and decisions, and not invent the way things should work based on some mathematical gymnastics.

And yet, most of sabermetrics is grounded in real-life data.  The irony of it is that she ignored the research that she says should be the linchpin of sabermetrics.

Even all my WAR analysis on trades and signings reflects the reality that most of them are “fair”.  They simply describe the reality that certain classes of players get discounts. 

And, they do point to some biases, like arb-infielders being undervalued and RBI-arb players being overvalued.  If Christina wants to argue that there is no bias, then she is dead wrong.

If she wants to argue that the #2 hitter should be a below-average contact hitter, she is dead wrong.

Our models reflect reality and they point to some suboptimization and biases.  We point those out in a logical and consistent manner.

We do not start with “I believe that...”, which is what all Ministers of Disinformation start with.  Her statement is in fact similar to saying “I believe that...” since she provided no basis for her statements.

That is what I thought she was saying.  I have no use for her kind of prose to be honest.  If I have to take 20 minutes out of my day to figure out what somewhat means, I have little use for it.

Anyway, since sabermetrics is simply using the scientific method to answer questions about baseball, these kinds of discussions are ridiculous.

If sabermetrics makes a “mistake” it is because “sabermetrics” is what fallible human beings make of it, not because there is something inherently wrong with the discipline.

It seems to be popular and chic to bash sabermetrics these days.

It is ironic when it comes from BP because I barely read them anymore.  Their research and writing is SO bad, from a sabermetric (the “truth") perspective.

Case in point (one of hundreds these days):

An article by Joe Sheehan on Tuesday about trading Teixeira:

Mike Jacobs makes an out 71 percent of the time. Mark Teixeira makes an out 61 percent of the time. Do you see?… Where I am going?… With this? Throw in defense, and Teixeira could well be worth 30-35 runs over Jacobs in just two short months. That would be a three-win improvement for the Marlins, who are kind of the anti-Braves:

Did he say that Teixera is worth 30-35 runs more than Jacobs in 2 months?

How can you read a web site whose most prolific writer just said something that is so patently ridiculous that it is not even worth discussing?

I’ve been an apologist for Joe for a while now.  I no longer am.

I must admit that none of the texts analysing lineup construction claims that it doesn’t matter, just that the effect is relativly small.

It’s also worth mentioning that even “small” inefficiencies are well worth fixing. Runs are a scarce and expensive resources, even if improving your batting order only might improve your offensive output 2-3 runs per season the alternative cost for getting those 2-3 runs can be close to a million dollars.

And as the saying goes, a million here and a milion there, after a while it adds up to real money.

If we’re going to figure out how much it can hurt to bat lefties back-to-back, shouldn’t it be done using win expectancy and not run expectancy, since lefty specialists will tend to be used in high-leverage spots?

Bjorn, touche (even a little gain is worth doing as they all add up to a large gain), but…

One of the problems with finding small differences is that if the model is not great, those small differences may not even be significant.

I am not a big fan of lineup analysis using Markov models other than as a starting point.

There are too many (admittedly) small factors that the models do not include, most notably baserunning.

For example, lately I have found that when I plug the pitcher into the 8th slot in my sim, in most cases it makes no difference. Thus, I am skeptical of the 2-3 run gain we get from Markov models, as my sim takes many more things into consideration.

Think of all the little things that might make a difference in real life, that none of the models takes into consideration:  Certain batters being able to hit to the opp field with runners on second.  Hit and runs or sac bunts if the manager is going to use them anyway.  Knowing what to do if a basestealer is on base ahead of you.  Certain batters having a little bit of “clutch” in them with runners on base.  Certain batters being “comfortable” in one lineup slot or another.

All of these things affect batting order but are not included in the models.

So if I am advising a manager, I am using the model results as a starting point and then letting my manager go from there.

My point is that as the resources used in production becomes more expensive it gets more and more worthwile to optimize the process.

Or in other terms, every run saved can buy a whole lot of analysis. Just to look at this example a smart team could probably justify (financially) hiring one or two highly qulified people full time to work exclusivly with analising batting orders.

Increasing production in the brute force way of adding more raw material (talent) is simply that expensive that not squezing every last drop of output you can get from the imput you put in should be unacceptable.

Correcting the things that even the crude models that exist say are clearly wrong (such as batting a fast but low OBP guy first) is the obvious first step but the equally obvious next step should be to spend more resources into building a better model.