Posted by Doug on August 18, 2008
If Ben Roethlisberger or Drew Brees or David Garrard or Eli Manning or Jake Delhomme were to go down with an injury and be lost for the season, how much would it cost his team in the W column by the end of the year?
Here's a quick study, made possible by the new QB start data.
The outline is this: take all teams who lost their starting QB for some games during the season, look at the team's points per game with the starting QB compared to its points per game with the backup QB, and then average the differences. This will give a rough estimate of how many marginal points per game a typical starting QB adds compared to a typical backup QB.
Seems pretty straightforward. But as you might imagine, there are a lot of details to consider.
Mainly, what is "a starting QB" anyway? For the 2007 Atlanta Falcons, which QB --- Harrington or Redman --- goes in the starter column, and which goes in the backup column? The answer is neither of the above. For the purposes of this study, I want to throw out teams like the 07 Falcons altogether, and consider only teams that had a clear starter who was performing reasonably well, and were forced to replace him with his backup because of injury.
Because I don't have historical injury info, I have to make some generalizations. So here are the teams I dug out of the database. I found all teams since 1990 satisfying the following:
1. the team's week one starter started at least 8 games and was above the league average in terms of adjusted net passing yards per attempt.
2. the team had other QBs who started at least three games during the season, not counting the potentially garbage-time-infested 16th game.
This query isn't perfect. I'm sure it will include a few teams I don't want to include and exclude a few teams I don't want to exclude, but I think it mostly hits the mark. In 2007, it gives us the Texans and Jaguars. Those are both teams we'd want to include. We also might have wanted to include the Rams, but the query doesn't pick it up because Marc Bulger's numbers were below average. If we relax the above average condition, we'll end up pulling in teams like the Raiders and Falcons, and I don't want to do that. If I'm asking how much a starting quarterback is worth, Josh McCown and Joey Harrington are not the starting quarterbacks I'm talking about.
In 2006, it gives us:
the Panthers, who were forced to replace Delhomme with Weinke for a few games.
the Redskins, who replaced Mark Brunell with Jason Campbell. It's questionable whether this should be included or not, as the Redskins were not winning when they made the switch, though Brunell's numbers were much more respectable than I remember them being.
the Eagles, who had a memorable several games with Jeff Garcia replacing Donovan McNabb. You might argue that Garcia is better than a typical backup, but that would be 100% revisionist. At the time, nobody thought the Eagles had a chance when McNabb went down. Nobody.
The point is, it's tough crafting a query that will pick up exactly the teams you want. The alternative is to go through the list by hand and decide which teams to include. This will severely limit the sample size (I don't remember exactly why Steve Pelleur replaced Steve Deberg in 1989) and will also introduce a lot of subjectivity into the process.
My solution is to use a rote query, but then try lots of different sets of parameters and see if they change the results. I tried using 110% of league average; I tried using 90% of league average; I tried requiring that the starter had a .500-or-better record in addition to above-average passing numbers; I tried requiring only seven games (instead of eight) for the starter and I tried requiring only 2+ games (instead of 3+) for the backup; I varied the minimum year. The results were surprisingly robust. The conclusions wouldn't change at all.
OK, take a guess. What's the number? How many points per game better is a team with its starting QB (as defined above) compared to its backup QB?
I'll post the answer tomorrow.