Go straight to the Projection Update!
Go to the previous projection results update
It's October 8th and the season is over. It's time to look at how the various projections have done.
As you may or may not remember, at the beginning of June, I thought it would be kind of fun to see how my projections had done so far, compared to some of the more high profile names out there. I cautioned that it was just June, and lots could happen to the stats of various players.
I also decided to post some more projections of players who have, during the season, managed to obtain a position where they are currently getting some playing time. This happens usually when someone gets injured, but sometimes they just win the job. On the same page, I've also finally added the projections for the hitters lucky enough to play their home games in Coors. Yes, I now have the projections up for all of the Colorado regulars. These projections do not use any of the current season performance (minors or majors) as a basis. They appear the same as if they were posted before the season started. They can all be found here.
I'll also add that I won't use any of these projections when evaluating how well the projections have done. Obviously if they're not up by the time the season is underway, it's hard to fairly include them in the analysis.
I'm going to do a quick comparison of mine, with those of Stats, Inc. (as published in their yearly handbook), CBS Sportsline, Diamond Mind Baseball (A computer simulation game), Baseball Prospectus, John Mosey's Site and Mitchel Lichtman (MGL). I am simply going to compare OPS at this point (in the coming off season months I'll dissect my projections bit by bit to try and find ways to improve them).
What I did was keep track of the projected OPS for each player from each source, and use the least squares method to obtain a correlation coefficient between the projected OPS for each source and the actual current OPS. (For all of you who don't speak math: I looked at what was projected and used a valid statistical technique to compare it with what has happened). I'm only doing the comparison for the top 170 players in At Bats plus walks that I projected, as has been the policy all year. This will insure that we have a representative sample. The results:
|Source||Corr. Coeff.||Average Error|
|Diamond Mind Baseball||0.664||0.069|
The first thing I'd like to state is that I think all of these are pretty good, and better than I'd thought the systems would do. Now, I'll state very clearly here that between all of the various systems, those aren't very big differences (they shouldn't be as the projections were all reasonably close to each other). I'll also state that CBS Sportsline's projections pretty clearly look to be just simple estimations based on fairly recent performance without taking into account minor league numbers. That fact makes it pretty difficult for their's to compete (and frankly I don't think they much care about their projections anyway).
That said, I'm very pleased. I know both Stats, Inc., Baseball Prospectus, DMB, John and Mitchel all put a whole bunch of effort into their projections, and I was pretty sure that for this first year they'd spank mine pretty hard. Although I finished with the highest correlation, you shouldn't look at the results of mine, STATS or Baseball Prospectus' projections as anything but a tie here. Honestly, I think correlation coefficients above .720 here are excellent and I'm terribly surprised I did that well.
This at the same time proves nothing and proves everything. What it doesn't prove is that my projections are better than everybody else's or that someone else can make that claim. From what I've seen they can't. What it does prove is that past performance is a fair predictor of future performance.
It's worth noting that the average projected OPS of the group for each source was .814 (me), .806 (Stats), .818 (CBS), .817 (DMB), .818 (BP) .843 (Mosey) and .827 (MGL). Whereas the actual average OPS of the group was .839. Credit goes to John for pegging the level of offense better than the rest. I'm not sure whether this is due to increased offense this year, or shaping, or whatever. If I had to guess, I'd say it's partially increased offense, and partially that there's a minor bias due to the fact that hitting well early on probably increases playing time.
As the offseason continues, you can expect to see a more in depth evaluation of my 2000 projections, some possible improvements for 2001, a look at what goes into projecting hitters, comments and feedback from readers and of course The 2001 Player Projections which I hope to get out in early February of 2001.
Thanks for reading along and feel free to write me with comments.
|Home Page||Hitting||Pitching||Macro Baseball||FAQ||Links||Contact|
Proud member of