THE BOOK--Playing The Percentages In Baseball

One-line summary?

Use the run environment of the PARK vs. other parks in the league to calculate the park effect, ignoring that the PARK run environment could be heavily influenced by the home team’s offense/defense.

http://www.fangraphs.com/blogs/index.php/win-probability-update/

Read the comments.

I really don’t disagree with anything there.

The BlueJays scored 359 runs at home and 355 on the road.  They allowed 289 runs at home and 321 runs on the road.

In all:
home games: 648 runs
away games: 676 runs

I can’t tell how many innings they played in all.  Let’s say it was 1450 innings at home and 1400 on the road.  So, per 9 innings, it’s 4.0 at home and 4.35 on the road.

So, what happens is that all games played at the Skydome is treated with a run environment of 4.0.

Where’s the problem?  Well, the long-term park factors implies that the Skydome is really run-neutral.  And since there are, I dunno, 4.6 runs scored per game this year in the league, the Skydome should be evaluated as a 4.6 runs per game park.

However, by only looking at how many runs are scored in the park (4.0), we are treating that as the universe.  This means that it looks like it’s a pitcher’s park.  And Halladay will get nicked here a good deal (0.6 runs per 9 IP, or .06 wins per 9 IP.  If he has 120 IP at home, that’s a total of .8 wins that he’s being undervalued.) On the other hand, all the Jays hitters will go up in value, since they all look like they are playing in a pitcher’s park.  (And for argument’s sake, let’s say they are truly playing in a run-neutral park.)

As I noted in my original example, this is exactly what was expected.  Treating the park as its own universe means that all the players are expecting to see 4 runs per game for each team.  And, we are FORCING the offense and defense to be equally responsible for that.

It is this last assumption that may be dangerous.  We know it’s not true in the sense we are most used to.  But it may be true in the sense of treating the park as its own universe, where everything adds up, and the offense and defense is equal.

So few parks deviate enough from the mean to make a serious dent in a player’s line that it’s hardly even worth making park adjustments for maybe two thirds of the teams. It seems to me that it would be easy enough to take each league’s R/G and apply a simple park factor.

The one concern is always Coors, and since Fangraphs goes back to 1972, and eventually back to 1954, there will be issues with the Astrodome among many other parks.

There’s a “best” way to do it, and a “quick” way to do it, and the bridge between the two is simply programming time.  So, it’s a matter of where along the quick-to-best scale David wants to go, and the closer he gets to “best”, the more time it takes.

Oh yeah, I forgot about the other 34 years. :-\

34? 36. I combined the two years to get 1974.

Here’s an interesting scenario that you may find fascinating.

Say you have two teams that are heavily overbalanced toward the pitching side.  They have a +200 run differential, but, they both play in neutral parks, both have league average offense, but both have standout pitching.  So, in the MLB universe of 30 teams, it’s the pitchers that absorb the lion’s share of the value.

For illustrative purposes, the league scores 4.5 RPG, these two teams score 4.5 RPG, and these two teams allow 3.25 RPG.

These two teams then enter the playoffs, and play each other.  They end up scoring 3.25 RPG each.  And naturally, allowing 3.25 RPG.

In this universe of the two-team playoff, we have no idea at all the balance of power.  Are their parks both pitcher-friendly or hitter-friendly?  Do we split the credit toward the defense or offense?

Indeed, it is only because we know how these two teams operate in the 30-team universe that we wish to give extra credit to the defense in allowing 3.25 RPG when they face each other.

But suppose they play each other 50 times, or 100 times.  Why do we necessarily care how they do in some other universe?

So, in one respect, the universe is only that in which the game is played.  And to that end, we should always give the credit 50/50.

Clearly, once we take them out of their own universe, and put them in a 30-team universe, the balance won’t remain 50/50 in terms of talent.  But, for the game in question, when all we have is two teams, does it really matter what happens to the other 28 teams 3 months earlier or in 2 weeks?

This was the Bill James argument from 25 years ago.  I like the argument, which is why I like the idea of the quick park factors.

I agree that it is an interesting argument.  As Bill said, he started using it because he didn’t want to mess with park factors for all history, but then it grew on him.  Obviously he doesn’t use the approach anymore, but I think he did use it in the contemporary Abstracts for at least a year.

I think of it a lot in regards to Win Shares, when Win Shares will zero out an entire offense if it is below 50% of the league average.  From the perspective of what the metric is trying to do, this is fine, as it’s a necessary constraint to enable the pitcher’s values to remain equal regardless of how bad their offense is, and is no different conceptually than giving out negative absolute runs to awful hitters.

But from another perspective, it is bothersome, which is that it assumes that the pitcher’s value should be constant.  From a certain perspective, though, it doesn’t really matter how your offense and defense rate relative to the league.  The crude way I’ve always thought about Tango’s “two-team universe” scenario is: If you score one run and allow two, you’re going to lose, even though your defense has been good and your offense has been bad.  One can say, “Well, the offense was -4 runs relative to the league and the defense was +3”, but one can just as easily imagine taking that run allowed off the board and changing the whole game.

If a team takes a 1-0 lead into the bottom of the ninth and Joe Borowski comes into the game and blows it, he’s the goat, even though the offense has been worse relative to the league than the pitching has (of course, as an individual, Borowski’s performance is the most harmful, whether you measure that context-neutrally or certainly it will be by WPA).  But the pitching unit can still be said to have outperformed the batting unit, yet the first reaction will be to lament the two runs allowed, not the four additional runs that would have been scored with average output.