There are two ways to build an atheoretical statistical profit-making machine. In one, you look at a bunch of inputs--past returns, earnings misses, financial statement ratios--and then group into bins, and look at future returns. In another, you look at the best and worst performing stocks, and then try to identify their characteristics. I find the latter approach rather poor, because let's say you find that stocks with really high returns tend to have ratios of Accounts Payable/Sales that are above average. It may be that this ratio is associated with greater volatility, so that it generates both massive success and failure. From a portfolio, average, or expected basis, it is unclear whether this is good or bad. That's all an empirical issue, so you might as well be doing analysis from explanatory data to outcome from the beginning.
So I heard about this new book, Who Are You and What do You Want? and the authors noted they talked to all these really successful people, and then looked at their characteristics, as if this was a really great way to discover great habits. I'm not convinced. Say, you find that being really individualistic makes Steve Jobs a great CEO. What about all the other big dreamers? Perhaps, statistically, this is a bad strategy because for every Steve Jobs who happened to have an Apple, there are millions of cubicle drones who are despised and unrewarded for their seeming impracticality.
The problem is, 'success' is difficult to measure, so we have a tendency to choose people who are super successful, so that there is no debate over whether they are successful. Unfortunately, then the sample size is so small and skewed it's not valid for identifying a pattern.
But then again, the authors run a huge business consulting company that involves flattering CEOs so they send their subordinates to their little retreat. I suspect that's one of the main reasons it is a common way to structure such a book. The idea that interviewing CEOs is a way to find wisdom is very convenient.