Yeah, this is a bit like comparing 538/Pollster/RCP to Gallup/PPP/Marist. The former aren't possible without the latter. And I cringed a bit where he said "My personal projections (Larson) took 2nd" given that his "personal projections" are just an averaging of other projections.

Dan, how do you use Pitch F/X in your projections? IIRC, Steamer's use of velocity was credited for its strong pitcher projections the past couple years. Do you use more than velocity?

I assume the target audience for this is traditional 5x5 fantasy players since the projection systems are evaluated using those categories. A title that makes that clear would have been a good idea. If this was meant as a serious evaluation of how well these projection systems predict performance, it's not well done because it's focusing on categories that don't matter much.

I look at two main bases of comparison: the first is the Root Mean Squared Error both with and without bias. Bias is important to consider because it is easily removed from a forecast and it can mask an otherwise good forecasting approach. For example, Marcel projections show very little bias, giving them a low RMSE, but are very poor at predicting variation among players, meaning that it’s not a terribly good forecast if you’re trying to rank expectations of future performance.

Eek! The only "bias" which should be removed is if the projection systems are projecting to different offensive environments (which they often are ... would be good if all the systems just adopted a standard, e.g. the previous year's averages). But if the bias is from other source then it's crucial to retain it in any evaluation. And a low bias method is only problematic for projection if it's lower bias was just random -- if Marcel is consistently lower bias, that's a huge advantage. (I suspect it's not and it is about the different contexts being projected to. Also the author seems to be suggesting that the variance in Marcel is so large that it outweighs the bias advantage -- i.e. it has a large RMSE anyway -- which would be a legit reason it would perform poorly.)

Also please stop averaging ranks as being a useful way to rate things. It assumes that each of the categories is equally important (which in 5x5 fantasy I suppose they are but not in any context that really matters) and that the difference between a rank of 4 vs. 5 (where the actual difference could be tiny) is the same as the difference between 1 and 2 (where the difference might be substantial). Add up the projections to get projected Rbat, Rbase, etc. and go from there.

Danny, I made that same point in my comment--my forecasts are probably say 85% intelligent (rather than rote equal-weighted) averaging of other systems, and 15% extra goodies. I do use more than just velocity, yes. Shoot me an email if you're curious about the details.

Is this contest judging on a mix of rate and counting stats?

Yeah, it's 4x4 fantasy evaluations. ZiPS always gets reamed in the ones with counting stats, as I make no attempt to tailor the playing time projections to my prediction of playing time.

ZiPS always gets reamed in the ones with counting stats, as I make no attempt to tailor the playing time projections to my prediction of playing time.

Mine always did too, until I mixed my rate stats with Tango's wisdom of the crowds playing time estimates. I try to answer the question "Who should play" while not caring so much about "how much will he play".

That's worked out OK for me.