So how much time does Dabo Swinney get, anyway?

Outside of a few bloggers that have been against the Swinney hire from the start, there has been little reasoned talk of letting Swinney go at year's end. I think most people will agree that barring a complete implosion (a 3-9 finish would qualify as such in my book), Swinney isn't likely going anywhere between now and next season. But that does beg the question of how much time is reasonable to give him. I argued last week that it's always in the best long-term interest of a program to let a coach go as soon as its clear he's not going to consistently produce winning and hopefully championship-caliber teams. At the same time, you can't go around firing your coach every year. That kind of instability is murder on recruiting and depresses the fanbase, which drives down revenue for the athletics program. So at what point can you start to distinguish good coaches from bad ones?

Luckily, this can be tested pretty easily by comparing the year-by-year distributions of win percentages of head coaches who held long tenures at their schools and the distribution of win percentages for head coaches with short tenures at their schools. Of course, win percentage doesn't tell a coach's whole story, and length of tenure isn't a direct measurement of how "good" or "bad" a coach is/was. But I think it's fair to say that in general, wins ultimately determine tenure length and tenure length is going to correlate with perceived success of a head coach. So, with the help of Wikipedia and jhowell.net, I went back about thirty years for each school in the modern ACC and pulled out records for coaches with "long" tenures (I defined this as 7 or more years, ignoring interim seasons) and "short" tenures (defined as 4 or fewer years). I also added a couple of additional requirements/exceptions to the above: 1) If a coach moved on to the NFL or another school with a stronger football tradition, I included them in the "long" tenure under the assumption that these guys likely would have seen continued success if they had stayed on. This ends up being a pretty small group of coaches consisting of some Miami guys, Steve Spurrier, and Tom Coughlin. 2) I required that the "short" tenure coaches either resigned or were fired. This seems to strike more at the heart of what is trying to be measured, so guys that left because of retirement or health conditions, etc. weren't counted.

With that in mind, here are the lists I compiled. First, the long-tenure coaches:

And the short-tenure coaches:


WARNING, TECHNICAL GARBAGE (skip to the next paragraph if you don't care about this): Now its simply a matter of running a two-tailed Mann-Whitney U test to compare the win-distributions during each year. I went with a non-parametric test because I wasn't certain this would follow a normal distribution given the stratified nature of opponent strength for Division I teams. I went back later and checked and it turns out it pretty much does follow a normal distribution, so there you go.

So now we want to test whether the average record in the two categories of coaches are significantly different in each year. This would essentially tell us whether or not we can distinguish the performance of a short-tenure coach from a long-tenure coach in any particular year. When we apply the test mentioned above to year 1, the p-value, or the probability of observing the differences in the data across the two categories, is 0.18. A p-value below 0.05 is typically considered marginally significant. Apologies to the anti-Swinney crowd, but what this says is that no matter what record Clemson finishes with this year we have no way of determining whether Swinney likely to be short, long, or in-between tenure coach.

However, moving to year 2 the p-value falls to 0.024. I was actually pretty surprised by this, but it appears that the difference between the records of the coaches is marginally significant even after just two full seasons. In year 3, the value drops to 0.00024, a highly significant number. In year 4 the number is still significant but climbs back up to 0.044; I think this can largely be attributed to sample size issues, both within the short-tenure category and the difference in sizes between the two categories.

I think this can basically be interpreted as justification for letting Swinney go after two disappointing seasons. The two most likely scenarios at that point are either 1) he continues to struggle mightily and won't be around after two more years or 2) he manages to eke out 4-6 years of mediocre coaching, maybe managing a decent year or two in the process. As a passionate fanbase, neither of those scenarios are going to be deemed acceptable at Clemson; go ahead and cut the rope. On the other hand, if Swinney turns in a good record in year one or two, then the scenarios are 1) 4+ years of mediocre-to-decent coaching or 2) we're basking in the glory days of the Swinney years, and wondering what we ever so worried about back in year one. Now, scenario #1 is much more probable, but you're probably going to stick with Swinney for a few years in case #2 comes to pass.

The next obvious question concerns what winning percentage should we look for at the end of the year? I'm a little hesitant to point to an exact record, because while the means of the two distributions (.568 record for long-tenure and .370 for short-tenure) are significantly different, there is variance. Certainly, if he posts consecutive sub-.500 records, well, that pretty much cinches his doom. Even if he manages a 6-6 or 7-5 record on year, I think it makes sense to look at other factors like recruiting class strength, fanbase fervor, season ticket sales, etc. to make the final decision on Swinney's fate.

Tigermax: Edited (9:45am) to make the conclusions a bit clearer...