Self-Identified College Peer Groups and Derived Rankings

The Integrated Postsecondary Data System contains a rich set of information about U.S. colleges and universities. As part of this information, institutions can identify a ``custom comparison group'' of peer institutions. 80% of the nearly 1500 institutions we consider in this work provided such peer data. We explore the use of these data to discover self-identified peer groups and derive a ranking system. The work focuses on the use of mutual peers, where a pair of institutions explicitly identify each other as a peers. In the dataset we found 23% of identified peers are reciprocated as mutual peers.We first identify completely-connected graphs where each member of a group identifies every other member of the group as a peer. The largest completely-connected peer group we identify is for 10 public institutions. The largest completely-connected private institution peer group consists of seven institutions. Overall, we identify 29 completely-connected peer groups with at least four members, which include a total of 132 institutions.We next develop a clustering algorithm that uses the completely-connected peer groups as seeds to generate strongly-connected clusters, which include a total of 311 institutions. Some of these clustered institutions do not themselves identify peers. The three largest clusters include one with 39 national public research institutions, one with 44 national private liberal arts institutions and one with 27 national private research institutions. The influence of the Association of American Universities (AAU) is particularly prominent in the clusters with almost all AAU members in either the national public or private research cluster.We conclude our work by exploring the use of peer data to rank institutions based on identified peers that are or are not reciprocated as a mutual peer. This approach is motivated by previous work that found an institution's identified peers tend to be a combination of who they perceive they are and who they aspire to be. We investigate the creation of a ranking system based on self-identified peers not because another set of rankings is necessarily needed or will result in a ``better'' ranking system. Rather we use it as an opportunity to understand how results from such a ranking system compare with existing rankings.We applied our algorithm to three groups of institutions: public PhD/Master's, private PhD/Master's and private Bachelor's with the University of North Carolina, Stanford University and Carleton College as the top-ranked institution in each of these respective groups.While the rankings within each group are interesting, our focus is not on the specific ranking of a particular institution, but rather on how the rank order of institutions in a group compares to other rankings. Using the well-known U.S. News & World Report (USNWR) rankings, we compare top-50 rankings in our three institution groups with similar USNWR ranking groups. We observed that just over a third of peer-derived and USNWR rankings are comparable for the public PhD/Master's institutions. However more that 75% of rankings are comparable for the private PhD/Master's institutions and two-thirds of the top-50 private Bachelor's institutions have comparable rankings.The results of the work do show that the peer data can be used to discover self-identified peer groups---both those that are completely-connected as well as grown into strongly connected clusters. The latter approach can be used to include institutions that they themselves do not explicitly identify peers. The ability to construct a peer-derived ranking system that yields a significant percentage of comparable rankings to the well-known USNWR rankings demonstrates the implicit importance of peers in such rankings and the likely importance of rankings in identification of an institution's peers