Jim Albright / the japanese insider
Description of the Rating Systems for Players
by Jim Albright
I'll start off with two tables which will be of greater general interest than the actual
description of the rating systems:
Now for the other table:
The Rating Systems for Players
When I did my first article for the Baseball Guru site, it was a ranking of the greatest players in Japanese history. I knew that because of many factors, the system used to arrive at those rankings was weak. A prime problem was the low values given to middle infielders and catchers. The article repeatedly reminded readers of that fact. When Bill James came out with his Win Shares book, I wondered if it might provide a solution. The first time I read it, it was clear I could not use the full system because I lacked a great deal of defensive data the system used. Further, using the full system meant rating all the players, and I have never had the time for such an ambitious project.
I put the thought aside for awhile, but in a rereading of the book, it dawned on me that his "short form" system could be adapted to give me a much improved system over the earlier "Greatness Points" system. I still had to jettison the idea of rating everybody to make the project manageable in size. I could normalize runs by comparing the runs scored and runs allowed by a team in a season to the historic NPB average of 8.81 runs per two teams in a game. This approach would enable me to use much more information in reaching the rankings, as the "Greatness Points" system used only league leading performances, awards, and career marks above certain cutoffs.
The short form system also gave me a framework to assign values per game played at various defensive positions. I realized I'd have to deal with situations where it was obvious a player had played multiple positions in a single game. When I looked further at James' book, I saw that he provided average defensive win shares for various ranges of defensive innings played. I realized I could improve my system by creating a curve to describe the data found by James, and then placing each player on the appropriate spot on the curve. This approach would still be a crude estimate of defensive value, but it would give credit to players who played more defensive innings at a position higher defensive evaluations than those who played less. I had to create a means to estimate the number of defensive innings, and did so.
At this point, while my emerging system was clearly inspired by James' short form system, it was just as clearly different and therefore needed to be called something else. I settled on Estimated Win Shares (EWS). I used a few tweaks from the complete win share system, such as counting one-half of unearned runs against a pitcher, on the theory the pitcher did something (such as let another man get on base or get a hit to plate the run) that contributed to the unearned run scoring. Since Japanese baseball has kept games finished in relief throughout its history, but only calculates saves since their adoption as an official statistic in 1974, I concocted a means to use the games finished data to more fairly rate star pitchers. In NPB, from its start in 1936 through much of the 1960's, used star pitchers to both start and, if they weren't starting that day, as relief aces.
Estimated Win Shares does a nice job of coming up with a career value, but merely adding up star players' EWS didn't separate great players from more ordinary ones as well as I wanted. After some experimentation, I settled on a framework inspired by what Bill James did in his latest Historical Abstract. I didn't use some of the fancy mathematical means he used (such as harmonic means) or some of the more subjective elements of James' system. I simply added to the career total of EWS the three best seasons in EWS from his career and the best consecutive five year total of EWS. Every season counts at least once, but a season which is among the three best can count three times (once for career, once as one of the three best, and once if it is a part of the best five year stretch).
That solved some difficulties, but left two other unresolved issues: what to do with major league service, and seasons lost due to WWII. The first was relatively easy to resolve. A player's MLB Win Shares as calculated by the full system would be used if the player had 50 or more EWS prior to playing AAA or MLB ball. If not, none of the MLB win shares would count. In order to emphasize NPB performance over MLB for the ratings, no MLB win shares would be used for the three best season or best five consecutive season bonuses. Further, win shares in MLB would count for the franchise/league the player left for MLB, unless (like Irabu), he returned to another team, in which case the two teams would split the MLB win shares. I am comfortable with this approach, especially when one considers that NPB has chosen to count hits and pitcher wins toward the Meikyukai, which is a group of the elite players from Japan, namely those with over 2000 career hits or 200 pitching wins.
The means of covering seasons lost to World War II is a little trickier. Any estimates derived by the means of covering lost seasons are ineligible to be used in the best 3 season and best five consecutive season bonuses, but those lost years are ignored completely for those bonuses to balance the scales for WW II era players. The rules for arriving at an estimate of EWS lost due to WWII are as follows: The gap much occur between 1938 and 1945. Further, there must be a season preceding the gap, and one after it, not later than 1946. I put this rule in because the issues of deciding when a player would have come up to NPB but for WWII or what he would have done in the remainder of his career had he not been killed or injured were far too speculative for me to deal with. I used 1946 as the end year because only a very few lost that season because of WWII, and researching out those cases was beyond my resources. The value given to each missing season is 1/3 the value of the season (the split 1937 and 1938 seasons were combined and viewed as one season for this rule) before the gap, 1/3 the season after the gap, and 1/3 the average value of a season in the player's career. This approach gets the player in the right time frame, and the combination of the three factors helps keep the adjustment from getting out of hand. I only used this adjustment for position players, because Craig Wright's research in his fine book The Diamond Appraised indicates pitchers who get time off due to wartime tend to benefit in career durability. Thus, I felt that pitchers already got their benefit from the wartime gaps, especially given the heavy pitching workloads given star Japanese pitchers of that time.
There were very large numbers of unearned runs scored in NPB before 1948, and I built in a factor which turned some of those unearned runs into earned runs for purposes of my calculations. Basically what I did was to bring those early seasons into line with historically normal ratios of unearned runs to earned runs.
I also expanded hitters' opportunities through 1946, due to the significantly shorter seasons up to that time. The 35 game 1944 season was doubled, and the other seasons (combining 1937 and 1938's halves into a whole season first) were brought up to 117 games or so. This number is still short of the historic norm of 130 games, but much closer to the mark. Again, I declined to make any such adjustment to pitchers because they were already pitching exceptionally heavy loads, and neither history nor common sense suggest they could have expanded those loads with a longer season without quickly experiencing adverse consequences. In fact, 400 IP seasons were very common then, in part due to the deadball style of play in Japan then)
I programmed all of my approaches into three basic spreadsheets, one for players who pitched and played some at one of the everyday positions. The others were one for pitchers, and one for everyday position players. The latter two were simplified versions of the first which were simply easier and quicker to fill out and use.
If anyone wants to go through the guts of those spreadsheets, I'll be happy to email those spreadsheets in Excel format to you and discuss the whys and wherefores in greater detail if you wish. Just email me here.
For the gaijin and team articles, I used only the rating system described above, which I call the rtg2 (for Rating #2) point system. I felt that with the lower level of accomplishment sometimes necessary to complete those teams that using arbitrary point bonuses for awards might create more distortions than they would resolve. At lower levels, players may fail to receive awards because better players get them, and so awards at lower levels are often as much a matter of chance as of skill. In turn, this means the effort to include such information is simply not worth the effort to do so. For the deade teams and the best overall player rankings, I included bonus points for awards actually received as follows: 3 for each MVP Award, 2 for each Best Nine Award or Sawamura (to a maximum of 2 points in a year), and 1 for each Gold Glove. When these bonuses were added to the rtg2 result, I called the figure Excellence Plus points. In the two cases where I included bonuses for awards, I felt the awards provided important information without seriously distorting the results at such high levels of achievement.
Finally, for the best overall, I also tried to fill in the blanks as much as possible for missing awards. The easiest rule to understand iss an MVP would get credit for a Best Nine Award as well if the MVP award came for a full season in which no Best Nine Awards were made. I felt that 3 points for a split season MVP was sufficient, however. Once those missing Best Nines were filled in, I allowed players who received multiple actual awards in a category (such as Best Nine or Gold Glove) to extrapolate the number of awards he would have received in the years in his career when the award in question was not made. He could only get credit for extrapolated awards for seasons in which he played in 85% of his team's games or pitched in over 30 games, and a further limitation was that he could receive credit for no more than 50% of the seasons meeting the other requirements. These rules prevented players from receiving credit for seasons they wouldn't have had a chance to win the award and/or to prevent extrapolating that a player would have won too many awards. Oh won his Nine Gold Gloves in the last nine years of his career, which were the first nine the award was made. If you don't put in the 50% limitation, he'd get credit for every season he would be seen as eligible, which is just too much to give him out of an extrapolation like this for my taste.
My sources for the ratings and the articles based upon those ratings are: