Okay guys, bear with me here, but I wanted to crunch some numbers on our last 10 years.
First I wanted to look at our adjusted offense and defense rankings (based on who you play and how you play) against our rank and win percentage (all pulled from KenPom). Generally, these give you some idea of how we've fared over the last decade.
I decided to plug these numbers into a basic OLS regression to see what sort of results we got.
For win percentage (because rank is a bit tautological), we get the following models:
Defense only:
Win% = .6868 - (.0007*defensive rank)
This doesn't show as statistically significant (which could be due to the small number of observations at only 10, and because of the lack of variation in our rank not giving us enough information), but it does suggest that higher defensive rank (which is a worse defense in this case) correlates to a lower win percentage. By way of example, if we're ranked 100th in defense that formula would give us an expected win percentage of .6168, while a defensive rank of 300 would give us an expected win percentage of .4768.
Offense Only:
Win% = .6990 - (.002*offensive rank)
This
is statistically significant at the .01 level. This means that if we're ranked 100th in offense, we would have an expected win percentage of .499, while an offensive rank of 50th would give us an expected win percentage of .599.
These are simple models, so I added a little extra (still super simple) to them. Somewhat intuitively it could be that we need to account for both of these items at the same time. Consider that if we look at what happens when we control for the other we get the following model:
Win% = .7966 - (.002*offensive rank) - (.0004*defensive rank)
This has an adjusted r-squared of .9296 (which basically just says offensive rank and defensive rank somewhat obviously explain a lot of what is going on with the win %). Again, offense is statistically significant while defense is not, ceteris paribus.
So an offensive rank of 159 and a defensive rank of 280 like we have this year would yield an expected win percentage of .367. Our actual win percentage this year is .389. That's not a bad prediction for a simplistic model.
Further, I thought it might be interesting, rather than holding changes in offensive and defensive rank at zero while examining the other, to look at how the combined total of offensive rank and defensive rank helped explain win % (so a rank of 100 in offense and a rank of 200 in defense, is a total rank of 300).
Running that regression, we get the following model:
Win% = 1.0062 - (.0014*total rank)
This variable is statistically significant, as well. So plugging in this year's numbers we would have an expected win percentage this year of .3916 (against an actual of .389).
The ultimate story here, depending on which model might hold up more strongly over the long run (if either), would be that the last model suggests if we could get our total ranks below about 350, then we should expect a winning record.
Just food for thought because I had an hour on my hands here at lunch.
Edit: fix image links