THE BOOK--Playing The Percentages In Baseball

This is the good fight please keep fighting it.

A better framework, in my opinion, is to simply compute Wins Above Average (WAA). Calculate the value of an average ML at a given position and adjust using your WAA and win value. Equivalent, but I find it simpler to avoid dicking around with replacement level.

#2, I neither agree nor disagree. It’s just as good a baseline as any other. The main problem is that WAA equates a pitcher who throws 100 innings with league-average run prevention to a pitcher who throws 200 innings with league-average run prevention. The second guy is worth demonstrably more, and teams pay like it.

So then you have to add a playing time component, or at least mention it (like you quote player’s stats as 0.2 WAA / 200 IP against 0.3 WAA / 100 IP) but then you might as well convert the two numbers into a wins above baseline, and you’re back to the start again.

That’s fine, if that’s what you want to do.

You have to understand the implications, which means a pitcher with a 4.00 ERA in a league of 4.00 is going to always rank higher than a pitcher with a 4.50 ERA, even if the first pitcher has one start and the second one has thirty starts.

Is that what you really want to argue for?

Or, are you going to first put limitations?

Or, are you going to put in some sort of “regression” that makes that pitcher with a 4.00 ERA and 1 start equivalent to a pitcher with a 5.00 ERA and 30 starts?

Or???

But whatever it is, have a consistent approach, and stick to the results.  And be prepared to argue for your approach.

But also if you get too many nonsensical results, then admit approach is wrong.

The replacement-level thing really messes with people more than it should. Generally people like comparing to average because it’s easy to comprehend, but like CJ said, a baseline is a baseline, right?

With WAR you’re just going .32 instead of .5 and you don’t have the playing time issues.

The WAR as a framework should really be discussed more often. I find when I’m talking with non-saber baseball fans they use words like ‘good’ and ‘great’ but I have to pry a series of references out of them first to see how they define ‘good’ and ‘great’ and so on. If he/she says ‘2 WAR player’ or ‘5 WAR player’ then we’re in business right away. It doesn’t matter if they don’t like linear weights or UZR, but at least we are speaking the same language.

Just curious as to who or what we’re arguing against here. What intellectual rhubarb have I missed?

And yes, that means once you’ve established that implementation that you must follow it blindly.  Now, it’s not blindly from the outset, but blindly after everything has been established.  You are a slave to your own process, your own methodology.  It prevents you from just tossing aside results just because you don’t like the results.

Obviously you must allow for refinement of the implementation, right?  The scientific method doesn’t say “once you open your mouth, you’re stuck forever with what came out”, does it?  If you find that your implementation leads to a result that looks wrong, you should investigate, and see if your model is lacking.

Of course, if you twiddle with your model to arrive at your predetermined preferred outcome, you would be abusing the scientific method in a different way…

Greg, right.  Everything is based on assumptions.  You can change your assumptions.  Ideally, you’d have a BASIS for having assumptions and changing assumptions other than wanting Strasburg to finish ahead of Zimmerman and Gio.

The important part is to create that model first.

Someone told me a story of how this person couldn’t believe how high Reuschel ended up in career WAR, that that proved WAR was no good.  Of course, he was then challenged to create his own comprehensive system with the express purpose of trying to make Reuschel look much worse (not a good reason obviously), and he couldn’t do it. 

Basically, he was stuck with the idea that Reuschel was better than Jack Morris, because his model told him so.  To then still argue for Morris over Reuschel in face of your own consistent model means that you prefer having an ad-hoc model, one that is not consistent, and will lead to whatever pre-determined results you want.

The basic story is to at least create some basis for discussion, one that you can apply in a CONSISTENT manner across a larger universe.

If a is better than b and b is better than c, then the model must give you that a is better than c.

And the key point is that EACH PERSON can create his own personal comprehensive implementation.