NYRoyal 1-0 ACLU $10 !!!: Place your Wagers HERE !!!

At least some of them. ZiPs uses batted ball data to a certain extent, for sure. I suspect PECOTA does, as well. Not sure about CHONE — I think Rally is pretty suspicious of PrOPS and stuff like that as a projection tool (and there are good reasons for that), and in any case, over a number of seasons, luck tends to even out. Regression to the mean and to groups of players based on sophisticated similarity scores, weight, height, etc., takes care of part of that.

Interesting that if you look at Teahen’s PrOPS for the last three years, weight the averages, then adjust for age, he ends up pretty much where’s he’s projected to be - a .750.775 OPS player.

So, look at his LD% and so forth to try to figure out if one or both may be outliers, if he looked unlucky one year and lucky the other, it is most likely inbetween, and so forth.

The projection systems (in various ways, to varying degrees do in fact look at that kind of luck and other kinds as well, like HR/FB luck. As Dan Szymborski, the creator of ZiPS said on this very website, “It starts getting very dangerous when you start throwing out outliers. Now, you can look at the data and see certain things within the data that may be outliers, but you don’t want to begin chucking out entire seasons of actual play that happened. In this specific case, ZiPS does regress Teahen’s HRs from 2006 more than usual because of an extremely high HR/FB ratio (17). It’s also the case the opposite way in 2007 when Teahen had an extremely low HR/FB ratio (6). Teahen finally had a normal season in 2008 (11%).”

The point is not that data should not be counted, its that the three year projection system wieghts get thrown way off by outlier data points because the systems because there are only three data points.

I don’t even think that all or perhaps even most systems limit themselves to just three years anymore. I believe for players of Teahen’s age, he looks at four years, which would include every data point in his major league career.

A projection system that takes into account likely outlier data and assigns it a lower wieght should be a better projection system. Is it really beyond projection systems to take into account normal major leauge player baselines and then have a computer analysis of the data find probable outliers?

Yeah, they do that. I really think you should read up more on what actually goes into the advanced projection systems like PECOTA, ZiPS, and CHONE. In varying degrees, they look at batted ball data and the luck that can be found therein, and PECOTA looks at comps as a baseline that they regress player projection towards.

Essentially, it would just be using advanced metrics to smooth out some of the jagged raw data that a projection system uses. We have this data already, why not use it in the projection systems?

Yeah, all of the above mentioned projection systems do this in one way or another. It appears that you assume all projection systems are just a simple regression model of the player’s last three years of stats. That describes the Marcel projection system, which everyone recognizes (including the creator) is significantly limited. The other, better projection systems use some type of multi-year data regression as the starting point and then do considerably more.