It is a well-known fact that the college share in advanced countries has been increasing in the past several decades. This is generally attributed to the cohort succession model, whereby successive cohorts obtain more education. Despite this consensus, using a shift-share decomposition we show that a large and increasing fraction of the aggregate increase in the college share is driven by the within-cohort component in the United States. This is in contrast to the general view in the literature that once individuals quit school, they never return to acquire more schooling. We use panel data from the NLSY79 and NLSY97 to analyze the causes of delayed education and its consequences. Preliminary results suggest that late graduates lacked the liquidity to go to college earlier. In terms of returns to late education, we show that late graduates experience a significant increase in earnings after attending college, although the returns are lower than for early college graduates, especially for men. These results suggests that an analysis of cross-sectional data may lead to bias in the estimated returns to education and experience.