My base numbers are brand new sets of numbers based on points scored last year by players who are returning this year, last years defensive avg ppg, number of returning starters, and a number I plug in for special teams. These numbers cannot be back tested because they are a new base, arbitrary, and may not apply to all teams. However, they may be very relevant, they should be adjustable so as to bring them current by making weekly adjustments based on current statistics as compared to the base numbers (ly's performances). I am afraid that I may have problems with teams where the QB leaves but all the key receivers return and maybe vice versa. If a new QB takes off right away (look how Texas Tech reloads at QB) the numbers may flow right along. I am pretty sure that defensively I account for losing numbers of players but not necessarily the quality of the players although that should be reflected in the avg points allowed per game. After week one I will see what would have happened if I had accounted for All Americans who returned and/or did not return and take a look at the QB/receiver scenarios. In the case say of Memphis/Ole Miss I could see a 25 pt blowout because there are 18 factors involved. Also, the first couple of weeks there are many mismatches (strong vs. weak teams) and that data may only apply to those types of matchups. I am not rushing anything or forcing anything. I am just going to assess the numbers and see if they indicate any kind of predictable performance. This will not be a system. At best it will just make something pop out that might not otherwise have been detected by not comparing the base numbers. But to answer your question you would have to have old rosters, and go over all the statistics for every team for x number of years and account for players who returned and who did not for every one of those years. I do not think that much info is available so I will have to do the trial and error thing. I am not creating a formula I am merely observing numbers to see if they are accurate and can assist in predicting the outcome of certain matchups.