from Amazon
Wednesday, September 29, 2010
PECOTA and Marcel on Ichiro
Colin shows us how the new PECOTA sees Ichiro compared to the old PECOTA. Let me show you his chart (2003-2010), and I’ll add the Marcel line:
AVG OBP SLG Results
0.330 0.374 0.428 Actual
0.319 0.368 0.425 Marcel
0.320 0.360 0.420 New PECOTA
0.308 0.354 0.404 Old PECOTA
As you can see, Marcel did great.(*) Now, is this a bad sign for New PECOTA? Not really. PECOTA lives and breathes on comparables, and if you have someone who is unique, like Ichiro, well, that’s where it’s going to lose. It’s going to win elsewhere, and it’s going to lose in places. Ichiro is one where it loses, only now, with the new PECOTA, it doesn’t lose so much.
What do I think Colin did? Well, for someone who is 100% unique, I think Colin would weight it as 100% Marcel (or Marcel-like) and 0% old PECOTA. For someone who is 0% unique, he flips it around. Maybe Ichiro is like 90% unique, and so, he gets most of the Marcel forecast and part of the old PECOTA. Something like that.
Anyway, what I would look forward to seeing is from Colin, and Brian, and Rally, and MGL, is historically, which kinds of players are they nailing better than Marcel. All Marcel data is here.
(*) Also note what Marcel does. It takes 5 parts year T-1, 4 parts T-2, and 3 parts T-3, and 240 PA of league average performance. In effect, Marcel has no choice but to pretty much always come up with the career line with too much regression. (The regression amount is too much at a career level because we are always adding 240 PA each year, when we should only add 240 PA ONCE.) Basically, Marcel lags a bit on the career average. So, the better test would be to compare year-to-year what each system does.
So I did. Here is the RMSE of new PECOTA and Marcel, 2002-2010 Ichiro:
BA OBP SLG
0.027 0.025 0.029 Marcel
0.027 0.030 0.027 new PECOTA
And the average of the absolute differences, year by year:
BA OBP SLG
0.021 0.021 0.022 Marcel
0.020 0.022 0.021 new PECOTA
Really, just nothing there in terms of difference.
Anyway, kudos to Colin for publishing the data to make the comparison easy for me.
Recent comments
Older comments
Page 1 of 344 pages 1 2 3 > Last »Complete Archive – By Category
Complete Archive – By Date