THE BOOK--Playing The Percentages In Baseball

If the goal of a park factor is simply to put into context a player’s total contribution in scoring runs or preventing them as a proportion of the runs it takes to win, then a simple adjustment applicable to all players is appropriate.  A player who creates on average half a run a game is less valuable if he is playing in a park where it takes 5 runs a game on average to win than if he is playing in a park where it only takes 4.5 runs a game to win.  That adjustment is relatively easy to make and can apply to all players pretty much uniformly.  The kind of close analysis of different sorts of park effects you describe can be interesting and fun, and can help explain the details of individual player performance in a granular way (if, say, you are curious about why player x hits more triples than player y).  But for comparison of player values, in terms of win contribution, it seems that a basic adjustment for the overall run-scoring environment in each park is quite sufficient, and its wide use for this purpose in sabermetrics is quite appropriate.

That’s the “backward-looking” goal.  If the goal is “forward-looking” (say trading players), then you need to do it at the granular level.

In short: it all depends.

Park effects are not even required for the goal birtelcom mentions. Adjustment for the player’s overall run environment, not broken out by home and away, would be sufficient.

But Joe, don’t you need a park effects number to calculate a player’s “overall run environment”? A large portion of what constitutes one player’s run environment as opposed to another’s is in the park effect.  You are right that once you have used the park effect number to calculate that overall win environment you don’t need (for “backward-looking” purposes, as tango well puts its) to break numbers down thereafter into home and away, but the park effect still needs to be in there.

It would be sufficient to compare player X’s 100 runs created to player Y’s 80 runs created if you know that player X created his runs in a season where his team plus opponents scored 1600 runs, but player Y’s environment was 1400 runs. You can compare their contributions to winning in those environments without worrying about how the runs were distributed between their home and away games.

I agree, Joe.  I would only point out that within the same league during the same year the primary variable between the 1600 run environment and the 1400 run environment is almost certainly going to be what we think of as the basic park effect, as for the most part the rest of the run environment will be largely constant (absent huge differences from unbalanced schedules—which are not likely given the probability that park effects of the visiting parks a team plays in will mostly even out for all practical purposes over a season even in an unbalanced schedule).  Thus if you have the kind of table of basic park effects with which we are familiar (more runs scored in Coors, fewer in Petco, etc.) and a table of overall league scoring in each season, you can pretty easily do a substantially reliable run context adjustment for any player without figuring a “run environment” for every player in every season.

Bill James talked about this in the original Historical Abstract.  That is, if you are playing in a park where there are 9 runs scored in the game, does it matter if that park is Coors or the Astrodome?

The answer is yes!  To a point.  Why?  Because you need to breakdown the split between offense and defense.

An Astros team that participates in Astrodome games for 9 runs is likely to play in 12 run games on the road.  A Rox team in a 9 RPG Coors game is going to play in 7-run games on the road.

If you look at the park as its own universe, you have to then presume that offense and defense will each get 50% of the credit.

This is not necessarily right or wrong.  But, this is a huge caveat.  Going forward with the “backward-looking park-is-a-universe” scenario, everyone needs to be well-aware of that.

Actually Bill James’ original Historical Baseball Abstract [1984] was the direct inspiration for my remarks:

Inventing a typical Jim Rice season as an example, James wrote:
“What I have done this time is just ignore the league and place the player in a “team” context. In the past, for example, I would have reasoned this way with regard to the Rice example:
a. The average American league team scores 4.5 runs per game; the initial context, then, is a 4.5-run-per-game context.
b. The Red Sox play half their games in Fenway Park, which inflates offense (and thus reduces the value of each run) by about 14 percent.
c. The value of Rice’s runs must be reduced by 7 percent to compare him to the rest of the league.
What I have done this time is simply ignore the league and start with the team. The Red Sox played 162 games and scored 830 runs; their opponents scored 790. That is 1620 runs in 162 games, or 5.00 runs per game; that is the context in which Rice contributed his 7.00 runs per game.
I started this, initially, because it seemed preferable to facing the incomprehensible task of figuring accurate “park effects” for every team (...)
But after I had been doing this for a few hundred players, I realized that it was not only an acceptable substitute, but actually a preferable alternative.” [p.298]

James continues with examples showing how park and league adjustments just end up equivalent to working with the player directly in the team context.

If I follow what you are saying above, tango, you are explaining why park effects tables are calculated the way they are: by comparing run scoring in Park X with runs scored in road games played by the Park X home team.  This way, a Rockies team that has excellent pitching and poor hitting will still produce a strong run-enhancing park effect for Coors.  That’s because you are comparing this team’s combined offensive and defensive performance in Coors, where run-scoring will be relatively higher, with its combined offensive and defensive performance on the road, where run-scoring will be relatively lower.  If you merely look at “run-scoring environment” without this direct comparison of team home games and team road games, a great-pitching, poor-hitting Rockies team will produce the illusion that its hitters are performing in a league average run scoring environment instead of the reality that they are perfoming (in half their games) in a run-enhancing environment.

birtel: you got it.

Joe: that’s the exact passage I was remembering.  But, for reason that birtel is mentioning, and that I showed wrt to Coors/Astrodome, you have to be verrrrrry careful.  There’s a caveat that comes with treating the park as its universe.

I for one like it very much, since it works wonderfully at a career-level, since it’ll be rather difficult for Larry Walker, Jim Rice, or Jose Cruz to continually play for a team that is pitching or hitting heavy.  At a career level, it’s probably rather fair to consider that the off/def split is even.  And, for that reason, it makes the “park considerations” far simpler if we proceed using the “Rice” model that James is discussing.