This seems like a pretty elaborate way of trying to understand whether a team wins when they are expected to. How is this an improvement from simply looking at a team's record vs. the spread? If they cover regularly it seems like they meet expectations and if they don't cover they are likely failing to meet expectations. I'd like to see your measure compared to data on spread covers. I bet they tell very similar stories.
I used the drive charts from SI.com and came up with the following. I only calculated points per possession (PPP) for UofM, OSU, and IU.
Number of possessions / PPP:
OSU - 57 / 3.46
UofM - 47 / 3.51
Indiana - 33 / 3.76
Because of limitation with the NCAA data you are using the number of possessions for each team is being underestimated. The problem however is that this is not constant across teams. While you underestimate the number of possessions for IU and UofM equally (-4), you underestimate the number of OSU possession by 6. What these results show is that this is affecting your calculations and the rankings of the teams. If we use the drive chart from SI.com UofM is more efficient offensively than OSU.
Granted this is a very small sample, but I think you'd find small, but potentially significant fluctuations in the rankings if you changed your calcuation methods.
I think that this measure is a wonderful idea and I apologize for being overtly critical. Just want to help make it as good as can be.
Here is an easy way to figure this all out (still somewhat labor intensive, but really not so bad). SI.com has drive charts for all games. You can simply count the number of "drives" for each team. This would obviously take a while given that we are 4 games into the season, but for each week after that I bet it woud only take 30 minutes to update each week. This way you wouldn't even need to account for turnovers, kickoffs, or punts. It would solve the end of half issue. It would take care of any other issue that would result from calculating it from end of drive outcomes.
I know this is more labor intensive (i.e., meaning you'd have to go back and look at the stats from each game), but I think it is the right way to calculate your measure.
I think you shouldn't focus on how a possion ends, but rather how it begins. I think the solution to the end of quarter dilemma and a more accurate measure of the number of possessions for a team is simply using the times the opponent kicks off, punts, or turns the ball over. That is if you want to know how many times TEAM X had the ball, you need to calculate how many times its opponent, TEAM Y, kicked off, turned the ball over, and punted.
Perhaps its not so much his inability to coach up his defense to stop the spread as it is his inability to recognize defensive talent/ identify potential player issues that has lead to the disparity between our offense and defense.
I think an additional point about defensive size, other than the fact that it has little explanatory power regarding defensive power is that just because two variables are correlated doesn't mean that one CAUSES the other. Rather, it simply means that they covary, but that covariance could be due to some unobserved 3rd variable. Statistically this kind of relationship is known as a "spurious relationship". For example, there is a strong correlation between ice cream sales and drowning, but that doesn't mean one causes the other. Indeed, both are related to a third variable...outdoor temperature. So even though it may look like two things are related they may not be.
It could also be that this is just a statistical anomaly. Its a 1 year study...this could just be a fluke.
My question is why would we expect this relationship? Why would having a heavier front 7 lead to having a better defense? The only argument I can think of is that the defense would be harder to push off the ball...but that applies primarly to the d-line and is really a testament to strength not necessarily girth.
Now, regarding the correlation between size of defense and defensive performance, the way to check against spuriousness is by running a multivariate regression with control variables. I think controlling for defensive scheme (3-4; 4-3; other.), offensive rank of opponents, team's offensive rank, weight of d-line only, opponent's % run/pass, and TO margin would be a good place to start.
I know such an analysis is probably not feasible, but I think that we shouldn't just assume causation here, however weak or strong the relationship
Recent Comments
This seems like a pretty elaborate way of trying to understand whether a team wins when they are expected to. How is this an improvement from simply looking at a team's record vs. the spread? If they cover regularly it seems like they meet expectations and if they don't cover they are likely failing to meet expectations. I'd like to see your measure compared to data on spread covers. I bet they tell very similar stories.
I used the drive charts from SI.com and came up with the following. I only calculated points per possession (PPP) for UofM, OSU, and IU.
Number of possessions / PPP:
OSU - 57 / 3.46
UofM - 47 / 3.51
Indiana - 33 / 3.76
Because of limitation with the NCAA data you are using the number of possessions for each team is being underestimated. The problem however is that this is not constant across teams. While you underestimate the number of possessions for IU and UofM equally (-4), you underestimate the number of OSU possession by 6. What these results show is that this is affecting your calculations and the rankings of the teams. If we use the drive chart from SI.com UofM is more efficient offensively than OSU.
Granted this is a very small sample, but I think you'd find small, but potentially significant fluctuations in the rankings if you changed your calcuation methods.
I think that this measure is a wonderful idea and I apologize for being overtly critical. Just want to help make it as good as can be.
Here is an easy way to figure this all out (still somewhat labor intensive, but really not so bad). SI.com has drive charts for all games. You can simply count the number of "drives" for each team. This would obviously take a while given that we are 4 games into the season, but for each week after that I bet it woud only take 30 minutes to update each week. This way you wouldn't even need to account for turnovers, kickoffs, or punts. It would solve the end of half issue. It would take care of any other issue that would result from calculating it from end of drive outcomes.
I know this is more labor intensive (i.e., meaning you'd have to go back and look at the stats from each game), but I think it is the right way to calculate your measure.
I think you shouldn't focus on how a possion ends, but rather how it begins. I think the solution to the end of quarter dilemma and a more accurate measure of the number of possessions for a team is simply using the times the opponent kicks off, punts, or turns the ball over. That is if you want to know how many times TEAM X had the ball, you need to calculate how many times its opponent, TEAM Y, kicked off, turned the ball over, and punted.
Perhaps its not so much his inability to coach up his defense to stop the spread as it is his inability to recognize defensive talent/ identify potential player issues that has lead to the disparity between our offense and defense.
I think an additional point about defensive size, other than the fact that it has little explanatory power regarding defensive power is that just because two variables are correlated doesn't mean that one CAUSES the other. Rather, it simply means that they covary, but that covariance could be due to some unobserved 3rd variable. Statistically this kind of relationship is known as a "spurious relationship". For example, there is a strong correlation between ice cream sales and drowning, but that doesn't mean one causes the other. Indeed, both are related to a third variable...outdoor temperature. So even though it may look like two things are related they may not be.
It could also be that this is just a statistical anomaly. Its a 1 year study...this could just be a fluke.
My question is why would we expect this relationship? Why would having a heavier front 7 lead to having a better defense? The only argument I can think of is that the defense would be harder to push off the ball...but that applies primarly to the d-line and is really a testament to strength not necessarily girth.
Now, regarding the correlation between size of defense and defensive performance, the way to check against spuriousness is by running a multivariate regression with control variables. I think controlling for defensive scheme (3-4; 4-3; other.), offensive rank of opponents, team's offensive rank, weight of d-line only, opponent's % run/pass, and TO margin would be a good place to start.
I know such an analysis is probably not feasible, but I think that we shouldn't just assume causation here, however weak or strong the relationship