Greg Mankiw writes the following:

If indeed a year of schooling guaranteed you precisely a 10 percent increase in earnings, then there is no way increasing education by a few years could move you from the middle class to the top 1 percent.
But it may be better to think of the return to education as stochastic. Education not only increases the average income a person will earn, but it also changes the entire distribution of possible life outcomes. It does not guarantee that a person will end up in the top 1 percent, but it increases the likelihood. I have not seen any data on this, but I am willing to bet that the top 1 percent are more educated than the average American; while their education did not ensure their economic success, it played a role.
Let me give you a couple examples. I am comfortably in the top 1 percent. I believe that Paul, with his Princeton professorship, regular Times column, speaking fees, and moderately successful textbook, is there as well. I suspect (although cannot prove) that if he and I had stopped our educations after finishing high school, we would not have been anywhere near where we are in the income distribution. If that is correct, might it be better to think of education as the key rather than focusing on the growing influence of oligarchs? [Emphasis added]

Thomas Lemieux’s work might be of interest, including his 2006 paper on “Post-secondary Education and Increasing Wage Inequality”:

I show that the changes in wage inequality and in the wage structure can indeed be described in a remarkably simple way. Like Mincer (1998) and Deschênes (2002), I find that the return to post-secondary education increased sharply while returns to lower levels of education remained relatively unchanged. Using quantile regressions, I show that the return to post-secondary education has increased even more in upper quantiles (like the 90th quantile). The return to post-secondary education has also increased in lower quantiles (like the 10th), but by a smaller amount than higher up in the distribution. But other than for this increase in the return to post-secondary education at various quantiles, the rest of the wage structure has remained remarkably stable over the last three decades. For instance, the experience-earnings profile, both at the mean and at various quantiles, has remained essentially unchanged since the early 1970s.
I then propose a simple explanation for these changes using a standard human capital model with heterogenous returns in both experience and education. As is well known in the returns to education literature (e.g. Gary S. Becker, 1967, David Card, 2001), there is not such a thing as a single “return to education”, but rather a distribution of returns across heterogenous individuals. This model implies that returns to education at all quantiles are linked, and that groups (like college post-graduates) experiencing relative increase in average wages should also experience increasing within-group wage dispersion. I estimate this model using a variance components approach and conclude that, holding the education and experience distribution for the workforce constant, increases in the return to post-secondary education accounts for most of the growth in wage inequality over the last three decades. As in Lemieux (2006a), most of the remaining change in wage inequality is due to composition effects. The increase in the return to other characteristics (experience, primary and secondary education, and unobserved characteristics) only accounts for about 10 percent of the growth in wage inequality. [Emphasis added]

Heterogeneous returns in experience and education is exactly what Mankiw has in mind.