Of the copious research on the labor market returns to college, very little has adequately modeled the pathways of non-completers or compared their outcomes with those of award holders. In this paper, we present a novel method for linking non-completers with completers according to their program of study. This method allows us to calculate the labor market returns to programs of study, accounting both for those who obtain an award and those who do not. We use a large dataset of community college transcripts matched with earnings data. We find that different classification systems--by algorithm, intent, or goal--yield very different enrollment patterns across programs. These classifications make a substantial difference to earnings patterns. Our results show that returns vary not only by program completion, but they also vary by program non-completion. Consequently, combining completers and non-completers yields a new pattern of returns. For some awards, this leads to wider earnings differentials. However, we find that the variance in returns by subject of study is reduced when we combine data on completers and non-completers. Finally, we find that progression in a program "per se" does not lead to higher earnings for students who do not complete (even as it demonstrably does for students who complete their program). If validated, these findings have significant implications for policies on program choice and on student retention policies. Appended is Figure A.1: Assignment of Students to Programs.