I was asked a number of times last week about Arizona’s drop from a solidly above-average team to a subpar one. The answer to that, and many more of life’s mysteries, lies in how we treat quarterbacks. Not only do we rate each team’s offense and defense, we (attempt) to separate QB performance from offensive performance using some fancy-dancy methodology called hierarchical linear modeling. Without going into to much detail, we’re able to do this because teams don’t use the same QB every game. Injuries and ineffectiveness lead to QB changes, and those changes allow us–with the help of lots of historical data–to see how much QB play actually matters.
When we grade out an offense’s games during the current season, we look at performance relative to expectation. That expectation is based on our QB rating, offense rating, and the opposing defense’s rating. We have no effective, practical way of separating QB performance from general offensive performance during the season (relative to expectation), so we don’t try to. Our QB rating is only used for the prior. Since the weight of our prior (relative to in-season surplus performance) decreases as the season wears on, and we don’t distinguish QB from general team performance during the season, our ratings are much more responsive to QB changes early in the season. (Our QB prior does change slightly throughout the season, but that’s based solely on factors such as number of starts.)
Turning back to the Arizona Cardinals, Drew Stanton comes in, Carson Palmer departs. We have a good idea of Carson Palmer’s value, since he’s played quite a lot of football (and played for a few different teams). We have a pretty good idea of Stanton’s value as well. There are a few factors that inform our prior, not all related to actual on-field performance. We know that he is not the starter, and is only starting out of necessity; we know how well he played in the past (poorly) and how much he’s played in the past (4 starts prior to this season, all 2010 and before); we know where he was drafted (2nd round) and how many years he’s been in the league (since 2007). Based on all this, our prior for him is pretty low. And while one could critique our methodology and say that our rating on Stanton needs to be updated (a quick google of “Drew Stanton” turned up an NFL.com story headlined “Drew Stanton must remain Cardinals’ QB over Carson Palmer”), Stanton’s performance this year would likely hurt his rating. Despite two wins in two starts, Stanton’s home-adjusted play success rate of 38.5% is third-worst in the NFL over the last two games, and his passing efficiency is significantly below average. The Cardinals have won in spite of Stanton, not because of him. Winning the turnover battle 5-0 and returning a punt for a TD are an excellent way to win football games, but they are simply not predictive. If the Cardinals want to sustain their success, their best bet is having a healthy Carson Palmer under center (well, there or in the shotgun).
For this week, Arizona’s rating assumes a 50% chance that Carson Palmer starts. If Palmer starts, Arizona’s rating jumps to +2.4; if Stanton is pressed into duty yet again, that rating drops to -1.0. Other QB changes/notes: JAC (Bortles for Henne), MIN (Bridgewater for Cassel), TB (Glennon for McCown), and STL (Davis remains starter).
A few other random observations from this week’s ratings:
- Style points do count in our ratings. A few teams lost and gained ground (STL, OAK, WAS) while others won and lost ground (DAL, NE, PHI).
- Ugly games (both teams’ ratings dropped): DET/GB,
- Pretty games (both teams’ ratings improved): CLE/BAL, ARI/SF
- Biggest gainer: ATL +4.13
- Biggest loser: TB -3.97