This is not a scientific setting where variables can be controlled. Here the use of stats works well over time proving the hypotheses. Passing more effectively than the other team is what wins 85+% of the time. You must have an effective running game in order to pass well, speaking in general and looking at not multiple games but seasons work of data. It's important to understand we are looking at data spread across multiple teams and seasons. With a large enough sample you can draw some conclusions.
You make observations then test them with the data. When they have a high correlation then you have something to hang your hat on.
i agree that an effective running game is needed to pass well, and an effective passing game is needed to run well.
also agree an effective passing game is essentially offensive efficiency.
these are common sense and to argue these points is plain silly, with possible exceptions being an amazing OL or someone like brady for a qb.
however, every time this type of topic is argued on this board, it becomes a statistical argument over common sense.
i started out this thread talking about # of possessions in a game, and it degenerated into an argument into running the ball, again.
if you try hard enough, you can manipulate statistical tests to prove most opinions.
for my final project in an econometrics class, i had to create a test/argument that world war 2 did not affect durable consumption.
is that silly enough for you?
i actually got an A for the project.
only took a few days on either TSP or SPSS.
one problem with using stats with the available data set is that the cowboys team is unusually constructed.
the other is that the data set is pretty small while there are many many variables.
after all, what we care about is the cowboys, and all other teams are not really important.
with the investment the team has made in the OL and offense in general, we are an outlier statistically.
to use a limited data set in an outlier situation is not robust.
you can draw some conclusions with some statistical tests.
i suspect they work pretty well for predicting the future success of college players.
but for what we are talking about here, it is a far more complicated situation.
you would have to take into account the particulars of the team e.g. romo's 'health', the existing investments already made.
i would be surprised if these are actually accounted for in existing models.
your data set is not big enough - this is not baseball.
if you include too many years, your data set is flawed because of the changes in the game including the rules of the game.
if you include all the teams, it washes out the individual differences in the teams - particularly the makeup of the cowboys with the investments in OL etc.
you can try to incorporate the strengths and weakness of the team in your model as additional models.
i suspect you would not get much more predictive benefits other than saying 'draft well and pick FAs wisely'.