I was saying that it would be interesting to create a Win Shares stat for football. That way it someone was responsible for 26% of the teams wins and the team won 11 games he would get more credit than someone with the exact same production (26% of the teams wins) on a 6 win team. The biggest issue I can see is that the small sample size for games played creates a high variation in win-loss records. In baseball it isn’t possible for a team to win 0% or 100% of their regular season games, but obviously in football it has happened twice each so it’s definitely possible.

One of the things that is important in using statistics is determining if there is a selection bias. The list for "Here are QB records against teams with winning records (MIN 50 starts):" has a selection bias in that QB who start 50 games are going to be some of the best QB’s in the game.

The second bias you are using comes your statement “The vast majority of those QBs … are within 5 spots of where they could be listed all time”. QBs who are on teams that win lots of games are also likely to win against winning teams. Since your main evaluation of QBs is based on who wins games, then of course a list like this will look to you like the QBs are list in approximately the right order.

I also agree with you on the Ben Roethlisberge thing. QB rating is a flawed stat. But it is not because it doesn’t take wins into account. It is because it doesn’t take the components of winning into account. Here’s a situation for you:

Steelers are backed up near the goal line in a 3rd and 15 situation and Ben throws a 14 yard completion. No first down, time to punt.

QB rating takes that 14 yard pass and says “That increases his completion percentage and increases his yards per completion, give the man some extra points”. A good system would say, “14 yards completion on 3rd and 15 gives you some credit for doing an above average play, but you do not get the extra credit for a successful play” and if the completion had been 1 yard longer instead of adding a tiny bit (like QB rating) it would add a sizable chunk for completing an unlikely play for a key first down (since you are no longer backed up).

Of course Ben did not do a lot of these things to help his team win, so he’s a pretty bad example. There were about 24 other QBs in the league that did more things to help their teams win (in the regular season) than Big Ben. So in this case the QB rating happens to have ranked him correctly. If you want to know who on the Steelers offense helped them the most to their wins, that was Mewelde Moore (13th most productive RB) and Hines Ward (7th most productive WR).

Obviously Griese wasn’t as good as Elway so the QB position was definitely downgraded. I never said that wouldn’t affect the team at all. In 1999 there were definitely a couple of games that because of the importance of the QB position some key mistakes lost them some games. There were also games (like the opener against Miami) where the defense lost the game. There are a number of games where the running game was real bad while Griese seem to be doing okay. There was even one game where Griese did well, but Brister laid an egg. It sure seems to be like there is a lot of complex circumstances that scream to me that Griese replacing Elway is not the reason why they lost 8 more games in 1999. It was part of the reason, because Griese was not as good as Elway (I’m sure the stats back this up), but not the entire reason.