THE BOOK--Playing The Percentages In Baseball

I posted something on his site about xFIP, which he might be interested in.  I know people like David and Graham have worked hard on more advanced pitching metrics, using batted ball info.  Personally, I think FIP and xFIP are 95% there, and I’m not sure you gain much by the added complexity.  Of course, it’s always fun to play with the numbers.

xFIP inherently includes the impact of a groundball pitcher, because it sets home runs equal to a standard rate per outfield fly allowed.  So if a pitcher is a groundball pitcher, his lower outfield fly rate will be captured.  Actually, that’s true for FIP too, though the home run rate can vary a lot from year to year.

You give .10*FB in place of HR, in the FIP equation, correct?

This presumes that the rest of the BIP are distributed based proportionately to GB, LD, Pops, using league numbers.  (It’s implied anyway.)

In my example above, I force the LD and Pops rate to be proportionate to the league numbers, and take the actual GB numbers.

I think mine is a bit better, if only because I use an extra piece of information.  However, that extra piece may not have much impact.

And of course, I would include HBP and exclude IBB.

But, generally speaking, I agree with studes.

By the way, in playing with this year’s data, I get a run value of .045 runs per GB and .026 runs per FB (leaving out infield pops and home runs).  I set my weights so that total weights = runs scored in the majors.

Actually, 11% times outfield flies, which is equivalent to 10% times all flies, but pitchers do have some consistency in the IF/F rate.

"I set my weights so that total weights = runs scored in the majors. “

I can’t tell you how wrong you are, but I’m sure we’ve had this discussion plenty of times.

Tom - Thanks for the comments and your time to post my question. 

Does anyone know how well fly ball rates correlate from one season to the next?

I really like how well GB correlate (.807) vice HR (.470) from one season to the next.

Google:
DIPS Revisited Lichtman
And post a link back here

Jeff might want to check out statcorner.com and take a look at tRA.  It sounds like he’s replicating some of what Graham has already done.

DIPS Revisited:

http://www.baseballthinkfactory.org/files/primate_studies/discussion/lichtman_2004-02-29_0/

If I were to have a list of the top 10 sabermetric articles one must read, that would be one.  It provides the basic principles for a series of related issues.

I’d really love to see how tRA compares to FIP in predicting future performance. I’m unaware of any study done on this issue - has there been one?

"I set my weights so that total weights = runs scored in the majors. “

“I can’t tell you how wrong you are, but I’m sure we’ve had this discussion plenty of times.”

Why is this wrong. If you trying to determine absolute runs, wouldnt you want your weights to equal runs scored for the league

Yes, I’m confused.  I thought it was okay to set the outs weight so that you get either runs above average or total runs.  In fact, that’s what you said in your “How Runs Are Really Created” article, I think.

I think that the reasoning is that you should reconcile the weights out to zero (that is, runs above average) and then convert those weights into absolute runs (something Phil said the other day has me thinking that adjusting the out value, rather than adding .12 to all PAs, is the correct way to do it, but I still need to think that over.)

As for what the difference is, I couldn’t tell you. I’m still a little puzzled myself.

Perhaps Tango is saying that Total Weights should = Runs Scored in the League, as opposed to Total Weights = Runs Scored in the Major Leagues.

Right, what Colin is saying.  First you adjust so that the sum equals zero runs.  The reconciliaton to absolute runs (which I’ve since amended from the Runs Created series) should be to add a fixed number to PA.
http://www.tangotiger.net/reconcile.html

What studes does is add a fixed number to the outs, thereby putting out of balance the run value between the events.

I can’t tell you how wrong you are, but I’m sure we’ve had this discussion plenty of times.

Obviously, we’ve never had this discussion before at all!  Oh well.  Go ahead, tell me how wrong I am!!! 

Okay, I think I figured out what’s been bugging me here, and I’m going to have to disagree with Tango on this one.

First, let’s call a spade a spade - FIP, xFIP, tRA, etc. are all properly understood as pitching linear weights. Just so’s we’re all on the same page here.

Okay, now as for Tango’s reconcile article. Here’s what he says:

The key to trying to understand this is that while you would properly assign +1 runs for a HR and -1 run for an out (in this league), hitters are assigned PAs and not outs. The reconciliation has to be worked out at the PA level.

That is true - for hitters. This is not true for pitchers or teams, however; in fact, it’s the reverse - pitchers are assigned outs, not plate appearances/batters faced.

So while I think Tango’s reconcile position is correct for hitter linear weights (this is not a strong conviction, however), I don’t think its appropriate for linear weights used to evaluate pitchers or teams. The right decisions to make for comparing and evaluating individual hitters are not correct and cannot be correct for run estimation. Individual hitters can avoid making outs, but at the team level the number of outs are fixed.

It is only correct if after you figure out the absolute runs, that you also figure out the adjusted IP.  That is, if studes wasts to give +.05 runs for a GB instead of -.12 (or whatever), he then has to figure out the outs made by a similar process, and then figure out the absolute runs figure per 27 “effective” outs.

I would bet, though I haven’t worked it out, that it would be extremely similar to simply doing what I do in terms of finding runs above average.  (Realistically, it would be better if I figured runs above average, and then converted that using a form of BaseRuns because a pitcher is his own team.)

In any case, if studes works out the absolute runs his way, he must also figure the adjusted IP.  I don’t know or remember if he does that.  If he does, then we’re both happy.

tRA does adjusted outs, I know. Don’t know about xFIP.

I use a modified BaseRuns equation myself:

http://statspeak.net/2008/08/creating-a-dynamic-fip-with-baseruns.html

I don’t estimate outs properly either. I should consider that.

And I don’t know why I bothered linking that, because I know I’ve talked to both of you before about it.

I am no linear weights expert (not even close) so I’m sure I’m doing some things wrong in my calculations.  Honestly, this is what intimidates me about linear weights, and why I don’t use it much.

But, for the Annual, I’m doing what I did last year.  I first enter generic linear weights for each event (using Tom Ruane’s tables), and then I adjust to the overall run environment by changing the out weight.  I now understand that that is very wrong.

In my original work three years ago, I used weights above/below average, but I saw that people clearly didn’t know how to interpret that info.  For instance, a blogger said that every groundball “gives up runs.” Lots of people don’t intuitively get the concept of above/below average on linear weights, and I’m the kind of guy who likes to connect with his audience.

So, two years ago, I switched to absolute weights, using the methodology described above.  I then ranked batters and pitchers by runs above/below average per plate appearance.  David and Tango pointed out to me that I should have used outs (which I actually kind of understand), so I changed it to outs last year.  Did the same thing this year.

I don’t know what “adjusted outs” are.

When you do runs above/below per PA, you have to use Linear Weights.  That is, the sum of all runs at the league level must be zero.

When you do total runs, such that the sum of all runs at the league level matches the total runs scored, then you must do total runs per out.

Which of the two most closely follows your process?

Studes,

I checked out tRA, which is the stat Tom and Colin keep referring to when they talk about adjusting the outs.

Here is my unqualified opinion on the subject. You only need to adjust outs, if you are adjusting the pitching stats to a neutral park. If your FIP does not neutralize the pitcher’s stats to a neutral park, then don’t worry about adjusting the outs.

On how you should calculate or your Linear Weights, you can get away with just adjusting the out value if the Linear Weights you are using are appropriate to your era. For instance if you wanted to calculate a set of LW for 2008, then I would just group Tom Ruane’s LW from the AL 1993-2004, and the NL 1993-2004 and then adjust the value of the out categories so that they sum to zero. To get “Runs Created” Linear Weights, you want to add the average RO (Runs/Out) to the LW values of the “out” categories. In Ruane’s case the out categories would be OOUT, SO, GIDP, SH, SF, and CS.

Here is what Tango and I mean by adjusting the outs, and it has nothing to do with LWTS and everything to do with DIPS/FIP.

A FIP/DIPS estimator says that a pitcher is not responsible for hits, only balls in play. (We can argue later whether or not this is Truth or a useful abstraction of truth.) Well, by the same token, a pitcher is not directly responsible for OUTS recorded on balls in play.

So, instead of dividing by outs, you need to divide by “FIP outs,” or the number of outs a pitcher would record based upon an average defense.

When you do total runs, such that the sum of all runs at the league level matches the total runs scored, then you must do total runs per out.

Which of the two most closely follows your process?

The latter.

Thanks for the clarification re: adjusted outs.  I’m talking about my batted balls methodology here, not any FIP-like calculations (sorry for the hijack, but it fit when I first brought it up).

Then I don’t have an issue I don’t think with your process.

The only thing it prevents you from doing is saying that a GB is .045 runs or whatever.  It means nothing without a context.  If the context is that an average PA is worth .12 runs or something, then fine. 

So, we just have to be careful when you take the .045 out of its specific context of what you are trying to do, and putting it into some other discussion.

It’s alot easier to keep things with a baseline where the average PA = 0 runs.