OK - a lot of questions and feedback here, so I'll try to answer as much as I can.First - Wood. We pulled projections from the site, and it's possible that we just missed yours. We pulled projections in mid August, so if they were up then then they should have been included. We can rate them now if you want, if you have a version that was in place on August 17th or so (although because we can't verify the last date they were updated we would just post the score here rather than on the site).Second - Why did we rate a bunch from Football Guys, and only one from most sites? Simply because your sites allow you to get a draft package that gives you a choice of projections. Most draft packages include only one projection set - so that's the one we rated.Third - Scoring. Because we scored projection sets by summing the squares of the differences, we ended up with total raw scores for the projections that varied between 461,000 and 764,000. We scaled the scores so that the set scoring 461,000 scored 100, the 764,000 set scored 0, and every other set scored proportionally between them. The formula for converting a set's raw score to its final score was something like [1 - (rawscore - 461,000)/(764,000 - 461,000)]. A score of zero didn't mean the projection set had nothing right - just that it was the worst in the group.Fourth - Why release now rather than at the close of the season? Simply because this project was a lot of work, and we didn't get around to evaluating the sets until now. We were hopeful that this would still be in time for the upcoming season - next year we hope to be more on top of it.Fifth - Why sum of the squares? Simply because in the plethora of stats classes I took, variance was usually dealt with as the square. Of course, squares are also used in regression analysis, but that fact doesn't invalidate their use in variance calculations. In principal it makes everything positive numbers, but I never really understood why using absolute value wasn't just as good. I suppose that tradition is the answer for why we did it this way, but it's probable that their use has some better reason for that in traditional variance calculations.Sixth - Injuries. This was a major issue, and we were very tempted to deal with the issue in the following way (in fact, parts of the guide still refer to this method): For injured players, points per game is not a true reflection of value, nor is total points. The true reflection of value is the points that player accumulated during the games he played, plus replacement value on the games not played. We felt that true replacement value was best measured not as "waiver-wire level", but at "backup level". To clarify what that means, if you are in a 10-team league playing two RBs, and you figure that players will keep three running backs on their bench in general, the waiver wire RB is the 51st RB or so, and the backup level RB is 21st or so. So the method here, for a player that played ten games, would be to take the points they collected in those ten games and add those points to the points that a backup RB would have collected (the ppg for a backup RB times six).That calculation is a little complicated because it is reflexive (just who the 21st RB is requires some ranking, and a ranking requires assessment of waiver wire value), but it can be done reasonably well. There is a complication in that some injuries can't be predicted - the player is played by the fantasy manager, and then turns out not to play - so replacement value has to be downgraded a little, but that's not terribly difficult to estimate either. We didn't do it this way because it complicates the analysis significantly, and doesn't usually change the results a flat ton. The fact is that it's unusual that a player turns out to have a lot of value whose injury significantly affects the number of games he plays in. I guess we can rerun the stuff using this method, but the reason we abandoned it was because it was a ton more work and didn't really change much at all.OK - let me know if you have any other questions - I'll check in later today.
I'll bet you could send in a request to have your projections evaluated after the fact. Maybe they will honor your request?
