program are set by each institution and, in some cases, by each separate graduate program.
Third, even if data could be collected or compiled using common definitions, the diverse structure of graduate programs would make comparison difficult. Rates of attrition would be meaningful only in the context of a particular program's admissions policies (open or selective) and its willingness to permit students to reenter who have dropped out. For example, less selective programs might expect higher rates of attrition, while programs with lenient reentry rules might have lower rates of attrition but longer time-to-degree.
A case in point is the Association of American Universities (AAU)/Association of Graduate Schools (AGS) Project for Research on Doctoral Education. With longitudinal data on individual students from 1989 to the present, this project has the potential to yield systematic, ongoing information on graduate student attrition, at least at major research universities. It is constrained, however, by many of the difficulties of comparability and definition discussed here (see Text Box 8).
In the face of such variability across programs, the panel concludes that it is not feasible or cost-effective to attempt to coordinate current efforts to collect information about graduate attrition and degree attainment in order to generate national estimates.
Having said that, we encourage individual institutions and researchers to continue studies of graduate attrition. Such studies can provide valuable insight to institutions so that students who leave do so by design and not through inattention.
TEXT BOX 8: THE AAU/AGS PROJECT FOR RESEARCH ON DOCTORAL EDUCATION
Hosted by the Educational Testing Service, the AAU/AGS Project for Research on Doctoral Education collects and maintains longitudinal data on individual students in Ph.D. programs in ten selected fields at approximately forty research universities (see Appendix A for details). The project has many of the features that would be required of a national, systematic effort to study graduate student attrition: it is national in scope; it includes a range of science and humanities fields; it collects from institutions data on individual graduate students; and it maintains a longitudinal data base of changes in students' degree status over time. Data are available from the fall of 1989.
One of the most difficult challenges in data analysis, as reported by project staff, is the variability in definitions of graduate student enrollment and degree status. The institutions participating in the project, and even departments within the same institution, define student status differently. They also may impose different requirements for the time limit allowed for degree completion, the need for students to be registered continuously, and the procedures for reentering after having "stopped out."
In this data base, students in a degree program who have not graduated and not officially left the program are divided into "active" and "inactive" students. Inactive students are defined as those who are not enrolled but are eligible to enroll, as far as is known by the institution. Inactive students may or may not be making progress toward their degrees; they may be studying for examinations or writing their dissertation; they may have left the program without informing anyone; or they may have taken a temporary break. The ambiguity of the status of ''inactive" students makes it difficult to determine an exact number of students to include in calculating attrition.
Over time, of course, most students in a given cohort will have completed their degree requirements or left the program. In the meantime, though, researchers wanting to use these data to study attrition will have to establish their own method to categorize the various types of inactive students.