Covering the Front and Back Pages of the Newspaper
March 8, 2005
BASEBALL: EWSL Explained
A reference post explaining the status of Established Win Shares Levels as of March 2005.
EWSL is short for "Established Win Shares Levels." EWSL combines two statistical measurements originated by Bill James: Win Shares (WS) and Established Performance Levels (EPLs).
Win Shares seek to measure a player's total contribution to a team's bottom line win/loss record, in the case of non-pitchers through combining batting and fielding contributions. The system makes the assumptions that a team's total wins can be rationally connected to its runs scored and allowed. Thus, each player is assigned a share of the team's total wins based on his contribution to scoring and preventing runs. As a result, a team's total "Win Shares" will always be equal to three times its number of wins (1 share per win would be too small to quantify the differences between players).
I can't explain Win Shares here at any length, but the relatively uncontroversial batting components of the formula are based on James' well-known "Runs Created" formula. The parts of the formula that are more debatable are the was in which the system divides defensive credit between pitchers and fielders and among fielders, including the fact that it gives extra credit to pitchers who have a lot of saves. Anyway, WS is widely accepted as a good, if imperfect, shorthand for a player's value.
EPLs take a weighted measurement of a player's accomplishments in any given category over the prior three years, giving the most weight to the most recent. Here's how EPLs work: entering 2001, Manny Ramirez had smacked 45 homers in 1998, 44 in 1999, and 38 in 2000, so his EPL was ((38 x 3) + (44 x 2) + (45))/6 = 41 (rounded off). In other words, Manny entered the 2001 season as having established himself as a 41-homer guy. Fortuitously enough, that's exactly how many homers he hit in 2001. It doesn't always work that way, but EPL is a pretty good shorthand for what a guy has proven he can do at the major league level.
In simplest (raw) form, EWSL simply takes a player's WS totals over the past three seasons and runs them through the EPL formula to get an EWSL.
The First Level Adjustments
Raw EWSL is not a projection of the future but a record of past performance, but obviously a major reason to look at EWSL, and a major reason why I've added a variety of adjustments to EWSL when compiling them into team totals, is to provide a look at a team's available established talent for purposes of estimating where they stand entering a season. To translate raw EWSL into something useable on a team level, a number of adjustments must be made.
Not all players have a three-year major league track record to work with, and that's where the first level adjustments come in. It's a judgment call, but if a player has only two seasons with any sort of significant playing time (more than, say, 100 plate appearances, although that's not an ironclad rule), I'll just use the prior two seasons rather than all three, and divide by 5 rather than 6. (Note that this is just for guys with no previous major league experience, and not for guys who were hurt; if a player appeared more than three years ago or if he missed most of a season with injuries, that season goes into the calculations. What I'm looking to exclude is seasons when the player was playing but not in the majors.) Those players are marked with a #. Players similarly rated on just the prior season - not divided by anything - are marked with a *.
For true rookies with no track record, I've used arbitrary adjustments based on projected playing time, rather than work in my own assessments of the player's quality. Those players are marked with a +. For the reasons explained in this post, I'm now using 12 WS as the write-in figure for rookies who are projected to play every day, 6 for rookie bench players under age 30, 3 for rookie bench players age 30 and up, and 4 for rookie pitchers, both those projected for a rotation slot and those in the bullpen.
The Second Level Adjustments
As explained in this post, players between the ages of 21 and 28 systematically outperformed their EWSL in 2004, while players between the ages of 29 and 39 systematically underperformed their EWSL. If you think about it, this is entirely unsurprising: young players tend to improve, older players tend to decline, and EWSL amplifies that process because a 28-year-old player's EWSL carries his stats from ages 25 to 27, while a 35-year-old player's EWSL carries his stats from ages 32 to 34. But in baseball, it can be a long way from 25 to 28, or 32 to 35.
Based on my study of my results from 2004, explained at the link above, I'm applying a second-level age adjustment this season, adjusting EWSL by the following factors by age:
Those are the actual results from 2004, with four exceptions. First, I'm not using these adjustments for players who are already getting the rookie adjustment; that would be double-counting. Second, I'm using the age 21 adjustment as well for players under 21 who aren't rookies, since there's insufficient data on players that young. Third, the actual factor for age 35, 0.52, struck me as too dramatic to be a reliable predictor; I've added an arbitrary "fudge factor" to bring that up closer to the other age factors, while keeping it the lowest. Fourth, it can't be realistic to expect players 40 and over not to decline, despite last year's stellar performances (in a small sample size) by the very aged; I dropped that from 1.00 to 0.85 to get a more realistic assessment. Even that may be too generous, but then the over-40 population is still heavily weighted towards the best pitchers in baseball.
One thing I don't have a systematic way of dealing with is previously healthy players who are already injured entering the season. Those are dealt with on a case-by-case basis, and where possible noted in the team comments.
Putting It All Together: The Team EWSL Figures
I'll continue last season's method - partly intended to save labor and avoid embarrassing mistakes in estimating who's going north with which team - of rating 23 players per roster, 13 non-pitchers and 10 pitchers. This gives us a number somewhere in the vicinity of an actual win total for a team, and avoids over-crediting teams whose benches are deep in discarded ex-regulars. In theory, I'd have a playing-time adjustment to factor in the natural limits on how many at bats a team can have, but to my mind it's a fair tradeoff that teams with exceptional depth get a bit of an advantage, since that depth often comes in handy during the regular season.
To make the method more transparent, I intend, in this year's ratings, to list EWSL in three columns: raw, adjusted, and then age-adjusted. I'll be eliminating the "weighted age" figures, since the age adjustments take care of that and those were a beast to figure out once I worked in the adjustments, although when I have more time later in the season I may go back and compute team weighted ages.
If you want a look back, here's last year's EWSL reports:
Anyway, EWSL is the most ambitious project I do around here, and I hope to make it a regular practice to get all six divisions done by the start of the season. Hope you find EWSL useful and interesting.
UPDATES: If you're wondering why the team totals sometimes don't add up, remember that the reported EWSL are rounded numbers, but team-level compilations are from the unrounded figures, and the individual adjustments are also made from unrounded totals.