Over the years, I've been surprised by data so often that I've almost stopped guessing what I'll find with any given analysis. I suppose that's why people analyze data, because conventional wisdom is so frequently wrong.
I'm also frequently surprised by the many questions that can be informed by data analysis.
I was reminded of both of these with an article in ESPN magazine. Several studies have looked at the characteristics of NFL coaches who are most successful, and they suggest the following criteria (quoted from the article):
1. They were between ages 41 and 49.
2. They had at least 11 years of NFL coaching experience.
3. They were assistants on teams that won at least 50 games over a five-year span.
4. They had only one previous NFL head-coaching gig.
Also, the best offense or defense coordinators don't necessarily make the best coaches.
Now, I would imagine that the study is bumping up against problems of statistical power, and there are certainly exceptions to these characteristics, but who knows, maybe the logic of Moneyball might be applied to coaches as well as players.
I'm also frequently surprised by the many questions that can be informed by data analysis.
I was reminded of both of these with an article in ESPN magazine. Several studies have looked at the characteristics of NFL coaches who are most successful, and they suggest the following criteria (quoted from the article):
1. They were between ages 41 and 49.
2. They had at least 11 years of NFL coaching experience.
3. They were assistants on teams that won at least 50 games over a five-year span.
4. They had only one previous NFL head-coaching gig.
Also, the best offense or defense coordinators don't necessarily make the best coaches.
Now, I would imagine that the study is bumping up against problems of statistical power, and there are certainly exceptions to these characteristics, but who knows, maybe the logic of Moneyball might be applied to coaches as well as players.
3 comments:
The article refers to "multiple studies," but here's my guess as to methodology: Someone threw a bunch of variables into a computer and came out with correlations or beta weights or some such. The dependent variable was probably W-L record and maybe post-season success. What they probably did not do was split the sample, derive the formula from one half and then test it on the other half. The formula may be very good for "predicting" successful coaches in the sample we have, but it's probably less powerful in predicting the performance of future coaches.
Two of the items are intuitive, but still it's nice to have statistical confirmation:
1. if someone was a good assistant coach, they'll probably be a good head coach
2. If someone was fired from more than one head coach position, he's probably not going to do much better with the next team.
Norv Turner has been one of the most successful coaches over the past three years with the San Diego Chargers. He was hired at 54 years old, had been the head coach of two previous NFL teams, and I doubt if when he was an assistant coach (for multiple teams) that he had five seasons where his teams won 50 games. He was considered an excellent offensive coordinator.
I doubt if he is the exception that proves the study, rather the study needs more work.
SFG
Yes, there are a lot of exceptions. I think that the studies were looking at association rather than determination.
Post a Comment