BP Comment Quick Links
October 24, 2001 Aim For The HeadThey CringedA month ago, my column "Making Statheads Cringe" did exactly that, generating a ton of responses. Unlike most other columns, reaction to the column was sharply split, as evidenced by the following two reactions:N.J. wrote:
Thanks. Outstanding article. By the way, you and BP do a terrific job. D.S. wrote:
Very surprising to see such crap from such a respected source. GIGO.
This week, I'll review and respond to some of the other comments:
T.D. wrote:
It was charitable of you to answer N.J., but you've done more harm than good by indulging him. You've created the Game-Winning RBI on andro and attached your name (and BP's) to it! How is this new freak stat any better for evaluating hitters than "Wins" and "Saves" are for pitchers? Yee-uck!
I agree that Contribution is not much better than Wins or Saves as a
player-evaluation metric. However--and I think this gets to the crux of the
negative reaction to the column--Contribution is not, and should not be used
as, a player-evaluation metric. If you recognize that the question being
asked is "How should credit for team wins be divided?" rather than
"How good was this player's performance?," then the resulting
answer that Contribution gives us, while still imperfect, may not bother you
as much.
The Contribution exercise is based on a team-outcome valuation system, which
runs contrary to the context-independent player-outcome systems
predominantly used in sabermetrics. However, I believe it's worthwhile to
occasionally break out of the box and examine how the baseball world looks
through other lenses.
B.S. wrote:
The problem with context-based evaluation, of course, is that the context is variable within the game itself. A hit by player A that contributes to a lead in the fifth inning is suddenly rendered "valueless" in such a system when the starter fades in the sixth inning and the team falls behind. And of course, when some other player on the team hits a two-run homer in the eighth, player A's hit is suddenly a positive contribution again (as long as the closer doesn't blow the save). B.S. has correctly identified one of the main issues with using the system I presented as a player-evaluation metric, although calling contribution "completely independent" is not accurate. Contribution is a post-hoc division of credit based on team results. Until the team has created something of value--a win--there's no way to assess who was most responsible for creating it. B.W. wrote:
How could you publish your most recent piece on "win contributions?!!" Given that there is no statistical evidence for an ability (year-to-year correlation) to hit above one's average in situations that "contribute" more to a win, your metric measures luck (both the luck of being on a good team and the luck of getting one's hits in close games).
I don't believe that something shouldn't be written because someone might
misunderstand or misuse what it says. Contribution is absolutely a measure
of luck, in combination with player skill, teammate skill, in-game
interactions, strength of schedule, park factors, and a slew of other
factors. It's a descriptive metric--a record of what happened (weighting
certain events deemed critical more heavily than others), rather than an
predictive one to determine the likelihood of a repeat performance, or an
analytical attempt to measure the relative isolated value of an individual
performance. And as such, the lack of evidence for consistent year-to-year
ability isn't a factor because we do not require such a characteristic.
James Kushner wrote:
Your article "Making Statheads Cringe" is a fun bit of research [.] Perhaps we should extend the logic involved...
James has taken the Contribution idea to its logical (some would say
logically absurd) conclusion, and we could in fact apply the margin of
victory for the divisional races or wild card as well as the in-game score
differential. The level of granularity is different, but the concept is the
same. Runs are tactical goals towards the strategy of winning the game. Wins
are the tactical goals towards the strategy of winning the division.
In a perverse way, this approach would defuse the often contradictory MVP
arguments about whether it's "more valuable" to win a race by a
large margin. For example, we often hear the argument that Mo Vaughn
was the MVP in 1995 over Albert Belle because the Red Sox wouldn't
have made the playoffs without him, while the Indians won their division
handily. Conversely, you also hear that Ichiro is the AL MVP this year
because he was the spark plug for the record-setting Mariners, which ignores
that Seattle won by such a large margin that removing any one player would
not have cost them anything in the final standings.
A.P. wrote:
The idea that the MVP award should go to the "most valuable" and not necessarily the best player seems reasonable to me. I was wondering if we might extend the state beyond just wins. The purpose of winning games in the regular season is to make the playoffs, so would it be possible to adjust the games won by how close the race turns out to be? That is, can we measure the player who was most important to his team in the regular season? Off the cuff, it seems like this adjustment would put Juan Gonzalez in a commanding lead over Bret Boone. A couple of people wrote in with the same idea, and the following table is thanks to A.A., who did the leg work for me when he wrote in:
Here are your results, but sorted by Contribution/Win which, I think, is a better indication of value. Granted you lose some of the importance that the player may play in winning more games for his team, I think this is outweighed by the exclusion of his team's pitchers, i.e., they play a role in the win total and so they help someone like Boone, while hurting someone like Brian Giles or Todd Helton.
American League Shawn Weaver wrote:
Nice article on importance and Contribution. However, I wonder why we attach importance to runs only in wins? Isn't it important for a player to contribute to effort, even if his teammates fall short? I think an examination of situations such as "late and close" might add some insight. Of course, "late" to me is not as important as "close," because if you score more runs than the other team by the third inning, you still win.
The basic assumption behind Contribution is that we're trying to allocate
credit for a successful team outcome-- the win. One could compute the
fraction of run scoring a player should be credited with in a loss, but
scaling that fraction by the number of wins the team had in that loss yields
zero in all cases. This is a top-down approach towards awarding credit
(starting with the team's value created and figuring out who is
proportionally responsible), versus the bottom-up approach of a context-free
model, in which each outcome a player creates has some fractional win value
which can be aggregated across a season.
Roy White wrote:
I don't see any reason to cringe at the idea of evaluating a player's performance according to its significance within a game; we don't get into the miry slough of clutch hitting unless we carelessly ascribe performance in a given context to clutch ability.
Roy has made the same point I tried to make above (and probably said it
better). Contribution is not a measure of clutch ability, it is a measure of
significance to the ultimate game outcome (with the caveat that we're
looking at offense only, not pitching, of course).
M.S. wrote:
Although I have been firmly in the sabermetric camp, I think you have stumbled onto something. You have captured what goes on in the right brain when thinking about the MVP: runs that win games. The method even devalues runs scored by teams with good pitching. Jason D. Scott wrote:
Nice job on the "Aim for the Head" this week. I understand the frustration that a lot of sabermetricians have with this type of analysis, but for years (although I love the work that BP does) there have simply been times when hardcore stat analysis just doesn't add up to the lived experience of watching a player contribute to a team's wins.
Both of these previous comments touch on the difference between potential
results and actual outcomes. At the risk of annoying the physicists among
our readers (or losing the interest of the non-physicists), I'm reminded a
bit of the principle of superposition--each player in the game produces a
contribution that has an effect on the probability of winning, somewhat
analogous to a wave function. Add up these "wave functions" for
each team, and you get a result that expresses how likely the team is to win
with these particular sets of contributions, yet at this point it's still
unknown whether the team actually wins (much like the fate of Schrodinger's
cat inside the box). However, the wave function only collapses to the actual
result when the game is played (or the box containing the cat is opened).
While this has been a fun diversion down a path not often
travelled by most baseball analysts, I don't expect to revisit
Contribution on a regular basis. It is not useful for the primary goal of most
sabermetric inquiry -- the assessment of player ability to help teams win
games. It's inferior to various measures that to attempt to describe
such abilities in several ways (including those mentioned by the readers
quoted in today's column), and thus there's no reason to prefer it as an
analytical tool (despite some correlation with context-independent value
measures). It's in the same category as an "analysis" of who had the most
Hall of Famers as teammates -- a way of looking backward at what has
happened, and categorizing events in some interesting, but not necessarily
predictive, way.
Keith Woolner is an author of Baseball Prospectus. You can contact him by
clicking here.
Keith Woolner is an author of Baseball Prospectus. 0 comments have been left for this article.
|