The 2003 Projections
by Tangotiger
Thanks to all the Primer readers for participating. Thanks also to Dan Werr and Chris Dial for their look at each player. A special thanks to Pete Palmer, Ron Shandler, Nate Silver, Dan Szymborski, and Ken Warren for providing me with their forecasts based on their engines. A Primer reader was kind enough to supply me with Diamond Mind's (DMB) projections. Another Forecaster, Danny Wind, supplied me with forecasts that were very different from the other 6 forecasters. While I will look at all forecasts at the end of the year, the 6 forecasters is the representative group that I will focus on in this article.
The following table shows each player's baseline forecast (3-yr simple average, adjusted for age), the average of the six systematic forecasters, and the average of the 165 Primer reader ballots.
Player Year of Birth 2000 2001 2002 Baseline Forecasters Primer Readers
Barry Bonds 1964 1.127 1.379 1.381 1.231 1.248 1.268
Jim Thome 1970 0.929 1.04 1.122 0.979 1.03 1.019
Gary Sheffield 1968 1.081 1 0.916 0.949 0.952 0.939
Troy Glaus 1976 1.008 0.898 0.805 0.949 0.883 0.887
Luis Gonzalez 1967 0.935 1.117 0.896 0.934 0.931 0.871
J.D. Drew 1975 0.88 1.027 0.778 0.895 0.865 0.873
Pat Burrell 1976 0.822 0.816 0.92 0.895 0.892 0.956
Moises Alou 1966 1.039 0.949 0.757 0.869 0.816 0.816
Richard Hidalgo 1975 1.028 0.811 0.734 0.858 0.821 0.798
Jeremy Giambi 1974 0.761 0.841 0.919 0.84 0.872 0.898
Sean Casey 1974 0.902 0.827 0.696 0.808 0.786 0.766
Roberto Alomar 1968 0.853 0.956 0.708 0.797 0.776 0.793
Jacque Jones 1975 0.781 0.751 0.852 0.795 0.822 0.814
Torii Hunter 1975 0.726 0.784 0.859 0.79 0.837 0.843
Rich Aurilia 1971 0.783 0.941 0.718 0.773 0.79 0.798
Rondell White 1972 0.867 0.9 0.666 0.77 0.726 0.776
Adam Kennedy 1976 0.703 0.69 0.795 0.766 0.74 0.75
Jeromy Burnitz 1969 0.811 0.851 0.677 0.741 0.737 0.724
Jeff Cirillo 1969 0.869 0.838 0.629 0.74 0.689 0.723
Jose Hernandez 1969 0.687 0.743 0.834 0.717 0.831 0.836
Marquis Grissom 1967 0.64 0.654 0.831 0.673 0.717 0.692
Freddy Garcia 1976 3.91 3.05 4.39 3.41 3.81 3.86
Javier Vazquez 1976 4.05 3.42 3.91 3.41 3.67 3.61
Ryan Dempster 1977 3.66 4.94 5.38 4.19 4.62 4.56
Kip Wells 1977 6.02 4.79 3.58 4.32 4.23 4.1
Chan Ho Park 1973 3.27 3.5 5.75 4.59 4.7 4.64
Matt Clement 1974 5.14 5.05 3.6 4.6 3.84 3.95
Jamey Wright 1974 4.1 4.9 5.29 4.76 4.89 5.1
Aaron Sele 1970 4.51 3.6 4.89 4.77 4.65 4.83
Shawn Estes 1973 4.26 4.02 5.1 4.91 4.34 4.79
Kenny Rogers 1964 4.55 6.19 3.84 5.35 4.41 4.52
Todd Ritchie 1971 4.81 4.47 6.06 5.62 4.77 5.16
League OPS 0.782 0.759 0.748 0.76 0.75 0.751
League ERA 4.77 4.42 4.28 4.49 4.312 4.31
Hitters
Remember that our group of players were selected based on the inconsistent results of their performance over the last 3 years. From that standpoint, we expect some disagreement of opinion, based on how you weight each year's performance level.
There wasn't much disagreement. Luis Gonzalez and Pat Burrell are the two main exceptions. In both cases, Primer Readers expect a continuation of the rise/fall of these players. The Forecasters on the other hand take it all in, and almost exactly match the Baseline Forecast.
Jeremy Giambi and Richard Hidalgo. While the Forecasters put more weight on their recent performancess, Primer Readers put much more weight in the 2002 performances.
Primer Readers and Forecasters were in agreement with Torii Hunter, Troy Glaus, Jim Thome, Moises Alou, Jose Hernandez: their recent performance is more indicative of their current talent level.
Running a regression analysis on the individual Forecasters, and averaging the results, the weightings used by the Forecasters was 60/21/18 (for years 2002/2001/2000). Taking the Primer Readers as a group, their weightings are 64/19/18. Essentially, not that much of a difference.
Forecasters v Forecasters
The largest disagreement among the Forecasters was with Jose Hernandez. Silver's PECOTA has him at .772 (essentially discounting his recent performance, and not considering Coors that much), while Dan's ZiPS took the opposite viewpoint. Marquis Grissom also confounded the group, with Palmer having him as an above average hitter (perhaps due to increased platooning), while PECOTA and ZiPS having him in the .670 range.
Barry Bonds polarized Palmer, Shandler, and Warren on one side (pessimistic, for Bonds anyway, at around 1.2 to 1.22), with PECOTA, ZiPS, and DMB on the other side (optimistic at around 1.27 to 1.31). Giambi had a similar effect, with Shandler, ZiPS, Warren, and DMB on the optimistic side. There are many many differences among the group, and when the season comes to a close, we'll see how they stacked up.
Pitchers
Interestingly, both Forecasters and Primer Readers project for the same league ERA of 4.31. And the agreement level is extremely high between individual pitchers. Garcia, Vazquez, Dempster, Chan Ho Park, Clement, and Rogers (6 of the 11) all were within 0.11 runs between these two groups. Wells, Wright, and Sele were within .21 runs different (one bad start). The two pitchers with the largest disagreement were Shawn Estes and Todd Ritchie. These two pitchers had a 2002 that was far worse than the prior two years, and the Primer Readers placed a great amount of weight on that.
Compared to the Baseline Forecast however, Primer Readers and the Forecasters do not agree. The Baseline sees Vazquez, Garcia, Dempster as much better than the Forecasters and Primer Readers see them. On the other hand, the Forecasters and Primer Readers see something extra in Matt Clement and Kenny Rogers.
Running a regression analysis, the Forecasters used a weighting of 62/26/12, while the Primer Readers used 60/21/19. While the Primer Readers used virtually the same process as with the hitters, the Forecasters discounted the year 2000 performance heavily. As well, while both groups regressed towards the mean for the hitters at about the same rate (about 15%, which confirms my findings in a previously published study), Primer Readers applied virtually no regression for pitchers, while the Forecasters regressed towards the mean at around 27%. (How many whiles was that??) It is interesting that with very different thought processes, both groups agreed so well on such a diverse group of pitchers.
Forecasters v Forecasters
The Forecasters, while as a group agree with the Primer Readers, could not agree among themselves on even 1 pitcher. The closest agreement came on Vazquez, with ZiPS checking him in at 3.47 on the one end, and Palmer checking him at 3.89 at the other end. On the agree to disagree scale was Kenny Rogers, with a full 1 run difference between the high and low.
Representative Forecaster
Which forecaster best represented the view of the Primer Readers? Ken Warren's hitter picks came the closest to Primer Readers, with 20 of the 21 hitters within .06 in OPS (Warren doesn't like Rondell). ZiPS came closest to Primer Reader for pitcher picks, with 9 of the 11 within .20 runs. Overall, ZiPS and Warren were the systems that agreed with the readers the closest. Each of the six Forecasters agreed more with the Primer Readers, than did the Baseline Forecast.
We'll check in at the end of the season to see how each group did, each Forecaster, as well as to look at all individual Primer Readers.
May 6, 2003 - Vinay Kumar
The part I find most interesting is Tangotiger looking at the Primer readers' forecasts, and reverse-engineering our yearly weightings and regression coefficients. I didn't think about that sort of thing when making my projections; they were more seat-of-the-pants than that, and I assume that they were for most people. It's interesting to see what weights and coefficients correspond to our seat-of-the-pants knowledge.
May 6, 2003 - tangotiger
(www)
(e-mail)
I didn't think about that sort of thing when making my projections; they were more seat-of-the-pants than that, and I assume that they were for most people.
I hope this is the case, as this is what I was hoping for. Can you get 100+ baseball fans to make seat-of-the-pants calls on extreme players, average them, and come up with something decent? We'll see in a few months...
May 6, 2003 - Tiberius
What a fascinating study. I am looking forward to the results. Maybe the Primer readers are more like Pecota; they look at similar players and then give a range of forecasts. We may do it by the seat of the pants rather than rigorous analysis, but my money is on the Primer voters.
August 22, 2003 - Scoriano
Player 2000 2001 2002 Baseline Forecasters Primer Readers Barry Bonds 1.127 1.379 1.381 1.231 1.248 1.268
Barry current actual 1.273 with a bullet.
December 16, 2003 - Scoriano
Bonds final is 1.278, beating the consensus-es.