Does Coach Age Predict Performance? (Yes, but be careful.)

By David Roher

I’m introducing the Speed Limit Coaching Corollary. If the coach of your favorite team is older than 55, or if your team is about to hire someone who’s older than 55, there’s a good chance you should start preparing for a frustrating stretch of football. — Bill Simmons

When I was developing the overweight coaches study, I figured it’d be a good idea to record each coach’s age. Given the tendency for weight problems to occur in our more advanced years, age could have been a confounding variable, one that affected both RAP and whether or not the coach was overweight.

It wasn’t. But I realized that age would be just as interesting to look at alone. After all, Bobby Bowden is stepping down as Florida State’s head coach after the Gator Bowl, and there’s a lot of thought that his age has prevented him from leading FSU effectively. And unlike the weight analysis, where I had to act like I was the only the judge on a version of “The Biggest Loser” without any scales, the number in question is freely available for every coach. So I decided to look at age using Simmons’ above column as the framework for the 2008 and 2009 seasons. He was focusing on the NFL, but D-I football is a good substitute.

The result? The Sports Guy is definitely on to something: among older coaches, there is a negative correlation between RAP and age. But correlation does not equal causation, and there are plenty of exceptions. Much older coaches like Bowden don’t have much to do with it and thus still shouldn’t be discriminated against.

The first thing I did was examine whether there was any correlation between RAP and age, without doing any additional categorizing. Nothing doing:

A scatterplot of coach RAP against their age. Nothing significant here. Note Bowden and Joe Paterno’s 2 seasons on the far right.

What if we do the same thing, except this time only looking at coaches over 55? Do we find anything? Yes. When I removed all coaches 55 and under, as well as the outlier ages of Joe Paterno and Bowden, there is a strong negative correlation between RAP and age. The R2 is .22, meaning that age explains 22% of variance in RAP. For the statistically inclined, the linear regression produced a P-value less than .0001, indicating a highly significant relationship.

The same plot as above, with all coaches not between the ages of 56 and 75 removed. The red is a best fit line, and the grey area around it is the confidence interval.

I’ll be honest: I wasn’t expecting anything near significance, let alone a relationship as clear as the one above. At the very least, I thought that selection/survivor bias in the hiring/firing process would take away anything. The hubbub over Bowden’s retirement and FSU’s ineffectiveness this year reminded me too much of the Mark Mangino situation, where the media connected two things that didn’t necessarily have anything to do with each other. But the above chart shows that something is up.

What I still don’t want to see, though, is people using this finding to condemn older coaches. First of all, as I’ve said before, RAP is flawed, and there might be skills that older coaches perform that aren’t reflected well here, or confounding variables that RAP doesn’t filter out. Second, and more importantly, Bobby Bowden and Joe Paterno, the oldest coaches in the Football Bowl Subdivision, produced significantly above-average RAPs in each of the last two seasons.

What that tells us is that the effect of age on coaches is exactly like the effect of age upon athletes. There is undoubtedly a general negative trend between age and production on the field after a certain age. However, age affects different people in different ways, and to differing degrees. Most athletes couldn’t be effective at 40, just like most coaches couldn’t be effective at 80. But we have guys like Randy Johnson, who struck out 173 batters at age 45, and (sigh) Brett Favre, who is one of the best QBs in the NFL at 40.

Advanced projection systems, like Nate Silver’s PECOTA, that use comparable players to produce their analysis have nightmares (yes, computers have nightmares) about these types of guys. They have no comparables. Neither do Paterno and Bowden. The best fit line like the one above is useless when coaches excel past age 65 (despite recent relative FSU struggles, Bowden won a national championship at age 70). They’re on their own curve.

What this study does tell us is that we’d be better served by changing our focus from Bowden and Paterno to coaches in their late 50s and early 60s. Some might begin to really drop off during this time (but some will stay the same or get better). We should still be very careful: after all, 78% of variance in RAP is explained by factors other than age. But the Sports Guy has the right idea with his Speed Limit Corollary — just use it sparingly and intelligently.

Within a couple weeks, I’ll post some matched pairs analysis, where I look at how coaches improved or declined over a small period of time.

This entry was posted in NCAA Football, NFL Football and tagged , , , , , , , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s