Customer Reviews

Win Shares

5 star:		(6)
4 star:		(6)
3 star:		(2)
2 star:		(0)
1 star:		(2)

 

 

Bill James presents his new player rating system in gory detail.
The Win Shares rating system was the basis for his player rankings in the New Historical Abstract.

James always seems a step ahead of everyone else when it comes to baseball analysis. The players' performance in a season is summarized by an integer number representing three times the number of wins a player contributed to his team. For instance Barry Bonds scored a 54 for 2001 -- tied for the third best season according to James.

Getting through the detailed explanation of the system (over one hundred pages) takes time. I wish it had been condensed some more. The fun begins when James uses his results to ask questions such as "Which players deserved postseason awards?"
or to compare players over different eras like Carl Yastrzemski and Chuck Klein.

Another emphasis in this book is on fielding stats. James spends a lot of time picking apart Total Baseball's fielding ratings and then attempting to prove his system is better.

There is also a 500-page reference section showing Win Share results for players on every major league team since 1876, career progressions for top players, career and single-season leaders, etc. Overall, a nice companion to the new Historical Abstract and Total Baseball.

Bill James latest book, Win Shares, is probably mostly of interest to really die-hard stathead baseball fans. Win Shares, as a tool for evaluating players, is of great interest to fans. It is a quantum leap forward in the analysis of ballplayers, allowing the evaluation of players based on a single number without the biases and problems built into linear analyis. And it is the first to correctly account for defense -- both in recognizing great defensive players and taking credit away from pitchers who thrived because of their defense.

This book, in particular, however, is one that all but the most diehard statheads will find a tad disappointing. The first 100 pages of the book is concerned with constructing the system and running it through three examples. I found this interesting because I was fascinated by the construction of this system. James does an excellent job of justifying his methodology and is sensitive to the fact that this is the first version of Win Shares -- it can stand improvement. But it might get tedious if numbers don't fascinate you.

The second section details why he made some of the decisions he made in constructing the system. The third section is what most people will find interesting -- a long section of Jamesian essays looking at ROY awards, MVP awards, comparing various players, finding good players on bad teams, extreme teams (all hitting or all defense), etc. Great stuff.

The fourth section is like Baseball Enyclopedia -- a listing of win shares for every team in baseball history, with detailed breakdowns for certain players. If you're fond of finding out who was the most valuable player on the 1982 Brewers or whether Don Sutton really WAS that good, you'll have fun. But if the baseball encyclopedia or baseballreference.com don't fascinate you, this won't either.

I feel I should defend this book from some criticism. Just because James' conclusions don't mesh with your prejudices doesn't make him wrong. James statement that baseball is 37% pitching will surprise some -- until you realize that pitching AND defense combined for 52%. Someone's got to catch the ball. The numbers don't lie on this one -- they are very clear and his win share system does predict the actual W-L record of teams very well. As for comparison of players . . . well, Jeff Conine WAS better than Troy Glaus -- in 2001. Conine played in a pitcher's park with a terrible offense. Glaus played in a hitter's park on a decent offense. But that's the beauty of win shares. If you look at their careers, Glaus is clearly the better player and has a MUCH brighter future, even if Conine was marginally better during his career year. The system is fair in evaluating players in different ballparks, different eras and on different teams.

For the die-hard stathead, this book is very good. For the more casual fan, I recommend the Historical Abstract. You'll find it an easier and more entertaining read.

:
People often look to new statistics to simply reaffirm what they already believe. Of course Babe Ruth's 1927 season was the greatest of all time. Of course, Bob Gibson's 1968 season was the pinnacle of pitching excellence in the 'live ball' era. Without a doubt, any number of Ozzie Smith's seasons would rank him in the very top of shortstops in terms of defense. And book after book comes out and tells us the same stuff we already have heard.

I, for one, have listened to enough people repeating these supposed "truisms" of baseball history to be exciting to another bandwagon statistician. The beauty of Bill James' Win Shares is that it presents a different way of looking at things, thus contributing in a new and exciting way to the discourse of baseball statistical exploration. And by the way, according to James's research, Babe Ruth's '27 season didn't crack James' Top 25. Two pitchers in the year 1972 alone rank better than Gibby's '68 campaign. And, no matter how hard you look, no Ozzie Smith season made the 10 ten for defense among shortstops.

James' system operates on the idea that individual statistics (such as batting average or home runs) should not be looked at in a vacuum. Obviously, a home run in Coors Field is less significant than a home run in Dodger Stadium. With more runs being scored, a home run in Colorado does less to contribute to a win than a home run in a traditional pitcher's park like Chavez Ravine. And batting .350 in 1932 when the National League average was over .300 means far less than it does in 2001 when the league average is in the .250 range. After fighting for years the problems that eras and parks present, James has extended the scope of statistical examination beyond solely considering production stats like slugging percentages and runs created, instead asking the simple and ultimate question: Who contributes more to producing a win? The idea of Wins created spans eras, positions, hitters parks, pitchers parks, good teams and bad teams. It includes pitching, hitting and defense and arrives at a single integer which can compare the seasons of any two players.

The idea of win shares presents an entire new methodology of statistical examination. The book, however, is not the final word on this new area of analysis. Bill James himself is careful to point out that there are some flaws, assumptions and imperfections in his work that will be smoothed out by he and others over the coming years. But it is exciting nonetheless to see the dawn of a new era in the field of baseball statistics.

Early criticisms of James' method have mainly been poorly reasoned, reflecting the some frustration at having subjective assumptions thrown out of whack. One reviewer on this very site has rejected out of hand James' entire work because he believes Troy Glaus is a better player than Jeff Conine and thus refuses to accept the fact that in 2001, Conine might have been more valuable. This is an easy trap to fall into. Win Shares makes no effort to proclaim who is an overall better player. Just because Jeff Conine out performed Troy Glaus in 2001 does not mean he will repeat that feat in 2002. And Conine's 2001 rating does not mean that he is somehow more valuable than Glaus in the long run. It simply says that he was more valuable over the course of a single season. And that's the fun part of the Win Share system. It demonstrates how, for example, Mo Vaughn won the MVP over Albert Belle in 1995, despite ranking 10th in the AL (and 2nd on his own team!) in Win Shares. It lets us compare Jimmy Fox to Barry Bonds.

But, as mentioned above, the system is not perfect. Certain results are so questionable that even at a quick glance, a reader can see a problem which needs to be explored further. In the ranking of the top 25 individual seasons in the post-1900 era, 16 come from the 25 year period of 1903-1928. Another 5 come from the 15 year span of 1942-1957. In other words, no seasons from 1929-1941 and only 4 from 1958-2001 cracked the top 25 (Bonds in 1993 and 2001, Mantle in '61 and Will Clark in 1989). While its possible that 84% of the top 25 seasons could have occurred in less than 40 of the 101 years of Major League Baseball, it seems unlikely that it's a simple coincidence. One is left to wonder how an era bias slipped into the equation.

Win Shares is original, exciting, complex, thought provoking, and fun. At this point, I would not use the data Bill James has collected by itself. But it certainly can immediately place itself alongside with OPS and Runs Created as a statistic that really tells us something of value. Win Shares is a challenging read. Over 120 pages of statistical method derivation and explanation is enough to intimidate even the most avid of amateur statisticians. But it's worth it, so don't be deterred. Once again, Bill James has pushed the envelope and encouraged us to examine old issues in new and fun ways.

The concept of win shares is wonderful and represents a fascinating step forward in sabermetric analysis. Any book that provides a single number to measure the value of each player for each season in baseball history is thought-provoking and worthwhile. Bill James's work on analyzing fielding statistics is incredibly useful and important.

I have a major criticism with the win shares system, however, and I hope I'll be able to express it adequately in writing. My problem is this. Win shares are based on the idea that a player contributes offensively to the runs that the team scores and defensively to limiting the runs the team allows. What James does, however, is start with the total number of wins for a team -- e.g., 97 for the 1977 Baltimore Orioles -- and then he takes as a fundamental premise that the Orioles as a team created 291 win shares. He then allocates those win shares to the offense, the defense, and the pitching, and then calculates each player's "share" of the team offense, the team defense, and the team pitching -- adding the total together for each player. For instance, Ken Singleton gets 36 win shares for the 1977 Orioles.

Here's the problem. Based on the total number of runs the Birds scored that year and the total runs they allowed, and using the "Pythagorean" formula popularized by James himself, Baltimore should have only gone 88-73 in 1977. So there should have been only 264 win shares to go around, not 291. That's a difference of 9%; accordingly, I would submit, Singleton should only get 33 win shares, not 36.

By contrast, take the 1972 Orioles, who actually went 80-74, but whose Pythagorean won-lost mark is 90-64. James's system allocates 240 total win shares to the team that year, instead of 270 -- a 12% percent difference. And therefore, Jim Palmer, who James gives 24 win shares for that year, should get 27. This makes a big difference; the James win share system gives Singleton's 1977 season a 12-win share advantage over Palmer's, but the Pythagorean system would peg the difference at only 6.

What's wrong, you ask, with assigning win shares based on actual wins instead of Pythagorean wins? Because in reality, the difference is simply a matter of luck. Over a five-year period, a team's actual and Pythagorean won-lost records will almost always be the same. The fluctuations come in one or two-year samples. Certain teams -- the 1977 Orioles, the 1975 Red Sox -- just get a lot of breaks. The ball bounces their way. And the win shares system should recognize that.

In fact, however, James not only doesn't recognize it, he compensates in the other direction. He gives Singleton (and all the 1977 Birds) extra credit for the fact that the team exceeded its Pythagorean projection. He briefly discusses this in the Snider/Mays essay in his book. I think this is a serious mistake.

The whole idea of tying the system to actual wins is a fiction, of course. Bill James doesn't know what Ken Singleton's statistics in 1977 were in the Orioles' 97 wins, as opposed to their 64 losses. It's theoretically possible that all of Singleton's production came in losses and was thus effectively useless to the team. We simply have to assume that every player's production is roughly proportionally distributed across the team's wins and losses; that assumption is built in to all analytical systems. That's why player evaluation is based on offensive runs created and defensive/pitching runs prevented. The Pythagorean system converts team runs scored and allowed into a won-lost record, and it is that record that should be the basis of the win shares system.

I have read a number of Bill James books, so before I started I knew what I would be getting into. Baseball as a sport is fairly unique in the ways it lends itself to statistical analysis, and people like Bill James are always trying to find better ways to use those statistics to measure the games of baseball. It takes Bill about 100 pages to fully define his system, which seems like overkill, but since he is setting up a system that covers baseball over more than 100 years, with all the differences in playing style and statistics that were kept there are a number of special rules. Bill explains them all in detail, but of course if you are uninterested in how catchers defensive ratings for players in the 1880's were calculated you can always skip that portion of the book.

The next section is devoted to defending his system, and various short pieces discussing different ways that his new statistical method can make things clearer.

The nice thing about this book is that people can actually demonstrate if the "win shares" concept is valid or not. There are a number of computer games that realistically can "replay" seasons. These simulations take the actual statistics, adjust for the ballpark effect and allow you to do a number of things. You can trade players, create your own pitching rotations etc. There are many people who are in leagues, in which you take over the job of the field manager/ general manager, make your own trades and try to win your own world series. After a few years, the players on the teams in the fictional leagues have no relation whatsoever to the players on the teams in the real major leagues. I am associated with such a league that has been in operation for over 20 years. It is as if you took the players in the real major leagues and drew them out of a hat to redistribute them. If the "win share" concept was not valid, "win shares" would not predict the number of wins for the teams in our league. However, "win shares" has a very high correlation in our league, the teams get very close to the wins that their "win shares" indicate they should win. It seems to me that this validates the "win share" concept.

This book will destroy some of the baseball myths that people believe. After reading this book, statements like "pitching is 75% of baseball" (which I am sure someone just made up without even looking at any evidence) can be totally disregarded. If you want a deeper understanding of baseball, you should read this book.

First, I am impressed by the quality and thoughtfulness of the reviews here.

Second, I agree with those who thought it was unnecessary for Bill to beat up on Pete Palmer's work. He did enough of that in the first Historical Abstract more than 15 years ago. I was embarrassed that he was still at it.

Third, I agree with James' assertion that in this work he has taken the analysis of fielding stats further than anyone has gone before. An absolutely brilliant and valuable contribution to the field.

Fourth, I think Win Shares has an enormous problem with interpreting and assigning relative value to pitching stats. I believe his requirement that Win Shares work for -- and be fair to -- players from any era undermines his method when it comes to pitchers. He acknowldeges the essence of the problem in explaining the adjustments necessary for pitchers from the mid 19th Century -- otherwise, all the best players in baseball history would have been pitchers from the 1800s. Isn't it simplistic to make only one such blanket adjustment for early pitchers and then expect the method to hold up across the years for everyone else, from Cy Young to Pedro Martinez? Bill has explained frequently that how pitchers are used is in constant flux. I think it might serve the accuracy of Win Shares if there were several such adjustments to support the evaluation of pitchers from different eras. Bill likes to point out how the logical flaws in other methodologies lead to laughably wrong-headed rankings. His career win shares list has Juan Marichal ranked below Tony Phillips. I loved Tony Phillips, but I cannot believe he contributed more toward the success of his teams than Marichal did.

Finally, Bill doesn't seem to enjoy the writing portion of his work as much as he used to, doesn't seem to let it rip with the abandon he did in the '80s. That's okay, that's probably inevitable. He is still a brilliant analyst and a fine writer.

I'm glad I bought the book.

A number of other reviewers have critisized Bill James' latest effort. While he does address the downsides to Pete Palmer's system, and then proclaim not to be "judging it," I don't feel that detracts too much from his work. On to some issues (good and bad) with the book.
Other reviewers have found issue with the fact that Win Shares are distributed without taking into account Predicted W-L records and the "luck" that determines the outcomes of some games. I disagree with that notion. James is not seeking to place a singular value on a player's season in a vacuum; rather, he determines a player's worth in the context of his team's performance. If you follow his method, he assigns "claim points" PRIOR to assigning Win Shares. These essentially determine the proportion of the team's success that is credited to a given player. If a team wins 90 games, regardless of their Pythagorean W-L record, their players did, in fact, win 90 games.
As a Twins fan, I will use 2001 as an example. Their record was 85-77, though the predicted record would be 82-80 based on runs scored vs. runs allowed. Corey Koskie is credited with 24 Win Shares in that season. Scaling down his Win Shares because the Twins finished higher than the Pythagoreon method makes little sense. First of all, who is to say the Pythagorean method tells us more than the actual record? A couple of blowout games can skew a teams predicted record, but losing by 5 runs is essentially the same as losing by 15. Second, keep in mind that James is not trying to say what SHOULD have happened; rather, he is analyzing what DID happen.
That said, Win Shares does fall short in some respects. It's most valuable when comparing the worth of players on one team; comparisons across different teams and seasons reduces its utility, in my opinion. James' decision to make each win worth 3 "Win Shares" is likewise questionable. While it is easy to convert those into a single-win system, the inverse would also be easy. James' system has the effect of emphasizing relatively small differences between players, which is just the kind of thing sabermetrics people are supposed to avoid.
Overall, though, there's a lot of good work here-especially on defensive statistical analysis. I suggest it to anyone who has either read James in the past or is interested in learning more about his chosen field.

This book is the sort of thing us stats geeks dream of. With Win Shares Bill James may have come closer than anyone else to developing a statistical system that objectively evaluates an individual player's performance. We already knew that Coors Field numbers don't mean as much relative to the rest of the league; but now we can figure out exactly what those inflated numbers mean. We can ask the question, "How good is Todd Helton?" and get a logical answer. Actually, he's pretty good, but he's no Lou Gehrig. More than that, we can ask "How good was Lou Gehrig compared with modern ball players?" Actually, he was pretty darned good, so good that James made him the top first baseman in his New Historical Baseball Abstract. This is not news, you may say. That's certainly right; but putting modern baseball players in a historical context, or vice versa, has been a deep well for fan argument for a hundred years or more. How do 1970's ball players stack up with 1930's ball players, or 1990's ball players? Win Shares gives an answer based on the logic that an individual's contributions to their team's win total is the most accurate way of measuring a player's performance. It's not a perfect system, which James takes great pains to illustrate; but he also illustrates at fascinating length how useful it is; as well as what it is (the complete statistical system is detailed in section II). If you are a baseball stats geek get this book.

Another great Bill James effort. His "New Historical Abstract" had introduced us to the "Win Shares" approach, and here, he fleshes it out. A real treat is that we even get ratings for middle-of-the-pack players; there aren't too many places where you can find meaningful ratings for guys like Omar Moreno, or Horace Clarke (who actually rates surprisingly well). Also the "letter grade" evaluations of players' defense are a fascinating treasure trove, and apparently the most correct defensive ratings yet, albeit with aberrations -- e.g., is Derek Jeter really a D+?

One negative: the author is harder on his colleagues than he needs to be. It's not that he isn't RIGHT about the flaws in their methods, but his forcefulness is off-putting, especially since (presumably) he's talking about his friends. (Hey, Bill, you don't want to turn into Howard Cosell!)

There are indeed flaws in the Win Shares method, but the author acknowledges them, and he emphasizes that his method, like all others so far, is a work in progress; in fact he seems unique among his colleagues in how prominently he acknowledges this. The Win Shares method seems a HUGE advance over everything else. Perhaps the largest and most important breakthrough is very simple, and I don't think the author highlights this enough: The method inherently places a premium on WINNING; if your team wins a lot of games, you tend to rate higher by this method than by others.

This book's ratings seem like the best of any so far anywhere, and the methodology is by far the most satisfying.

A reader said earlier, if someone other than James wrote this book, no one would care. This is probably true. However, that James had the determination to write it is in itself amazing. He is basically saying we need to take a step back from baseball statistics (and much of what he himself has worked on before) and get a better feel for what these statistics really mean. In this book James provides an outline of how statistics should properly be analyzed. Much of his analysis (especially fielding formulas) is subjective. He explains that many values are estimates. However, James always is sure to explain the logic for his analysis. Furthermore he states that he hopes and expects better estimates and methods of analysis to be found to evaluate players.

Win Shares is not the grand masterpiece many readers were hoping for. Instead it is a retraction of much of the previous statistical work of the last decades, to much of which James contributed, and an explanation of a better approach. This book shows James' determination to find truth within statistics, no matter what limitations there are on the system. Win Shares MUST use a lot of subjectivity and estimates to get at the truth of the statistics. It is sacrifice that is well worth the price.