Before I get to that, I want to give a little background. I'm of the ever shrinking school of thought that a quarterback should not be judged by his win-loss record. Working backwards, I decided that quarterbacks have much more control over their team's offensive output than they do over the actual game results. That's why my first article/analysis/"journey into loser-dome" is trying to find the stats that can predict offensive output.
I started off by looking at a couple of basic stats:
- Yards per attempt
- Interception %
- Average points given up by opponent (with the game being analyzed backed out)
The first thing I did was look at the correlation of each of these individual statistics with the team's scoring output. The numbers used were the averages for the 2008 season.
The 3 graphs show the correlation between each of the 3 variables above and team scoring. If there is a correlation, we'll see a linear pattern, with the data points arranging themselves diagonally from bottom left to top right (or, if it is a negative correlation, from top left to bottom right).
The results were a little surprising to me. Basically, the correlation of team score and the quarterback's yards per attempt is significantly stronger than the other two. Interception % ahad almost no predictive value and average defense only offered a little more.
____________________________________________________________________
Applying "Model" to Game Level Yards Per Attempt
Let's take this a step further and apply the formula from the correlation of average yards per attempt and team score (Table 1-a) and apply it to individual games in 2008. The formula was y = 3.9572x - 5.7059 (x is yards per attempt, y is team score)
The resulting graph has an R squared of 0.2746. This means that 27.46% of the variance can be accounted for by this one variable. While that's not great, when you consider all of the factors that go into a 60 minute football game, being able to explain one-quarter of the variance team output with one metric is pretty surprising.___________________________________________________________________
Do We Even Have to Look At The Score?
We know that there is some predictive value to comparing a quarterback's yards per attempt to the team's scoring. So, what if we look at how frequently the quarterback with the higher yards per attempt wins individual games.
Teams with a favorable yards per attempt won at a 71% clip (150-61). Only the Seahawks (2-6), Rams (0-3), Chiefs (1-3), Raiders (2-4), Lions (0-1), Bengals (1-2) and Packers (5-6) had losing records in games where they had better yards per attempt. Only the Titans (5-2) had a winning record when they had an inferior yards per attempt.
All of this says to me that a player who has a better yards per attempt statistic has a better chance of leading his team to victory.
What does this all mean? Not sure.
____________________________________________________________________
One Last Thing
Below is a listing of quarterbacks and their yards per attempt.
Hi Luis,
ReplyDeleteI saw your comment, and I'll definitely link to you the next time I get around to updating my links.
Interesting to note that there's a (very, very slight) positive correlation between Int% and points scored. Traditionally, one would expect there to be a notable negative correlation.
Thanks for posting. The positive correlation was odd. I'm actually enjoying this too much I think.
ReplyDeleteSo I got to thinking...maybe interceptions have such little correlation to scoring because they're really very rare?
ReplyDeleteTake blocked field goals, for instance. They probably have no appreciable correlation to scoring because they only happen so infrequently. Same with safeties or punt returns for TDs.
There were 465 interceptions last year, out of 31,681 offensive snaps. That's an interception on every 1.47% snaps. Might be that's too infrequent to register a major influence, and it could also be that "gunslinger"-types who tend to throw ints. are also good at leading their teams to touchdowns, while "game managers" are the opposite.
Good point. Maybe I could just check the correlation - or add to the yards per attempt model - a 0 for 0 or 1 ints and a 1 for 2+.
ReplyDelete