Posted by Justin Kubatko on August 10, 2009
I have been working on a way to extend the Win Shares system to seasons prior to 1973-74 off and on (mostly off) for quite some time. Until last week, though, I had not been able to come up with a satisfactory solution. However, I had an idea last week that produced what I felt were satisfactory results, so today I would like to announce that the Win Shares system has been extended back to the 1951-52 season.
Full details are given in the article Calculating Win Shares, but I wanted to comment on a couple of things that people might view as flaws/bugs in the system.
First, if you take a look at the season leaders for Win Shares, you'll notice that the top ten are all from the 1960s and early 1970s. Does this mean the system is biased against modern players? No, it doesn't, because a key variable that needs to be considered is playing time. Here are the average minutes per game for the top five players in minutes per game over the years:
As you can see, the top players from the 1960s and early 1970s were averaging about 4 more minutes per game than today's stars. For that reason alone we should expect the top players from that era to have roughly 10% more Win Shares per season.
Second, NBA record keeping was notoriously spotty in the early years. For example, there are 62 cases from 1950-51 to 1961-62 in which we know a player played for multiple teams, but we don't know his splits for those teams. In other words, we know his statistics for the entire season, but we don't know his statistics while playing for Team A and Team B. Because of this, some players with unknown splits will be missing Win Shares data.