THE STUDENT-LEVEL DATA provided by the 13 universities consist of two parts: data collected upon students' entry into graduate school, and yearly reports of their progress. Aware at the outset that the quality of the data would depend on having clear rules for data collection and reporting, the Mellon Foundation established standardized formats for both. Each university was tasked with reporting raw data on every student registered in the relevant departments.1 The Foundation took responsibility for making all statistical calculations to ensure that they were uniform and that reliable comparisons could be made. Each year, the Foundation transformed the data it received and provided summary tables to the universities and departments for their own use.
To preserve confidentiality, the records were provided anonymously to the Foundation. The universities assigned identification numbers to the students for the purpose of connecting their records year by year. The Foundation agreed that any publication of the data or analyses would be presented in such a way that neither individual students nor individual departments could be identified.2
1 Additional student data were added from the records kept by the Woodrow Wilson Na-
tional Fellowship Foundation on all students to whom portable Mellon Fellowships in the
Humanities had been awarded. Since many of these students studied in treatment and com-
parison departments in the GEI and had been deemed especially promising, having such
information would permit controls to be introduced for assessed student quality. However,
this particular analysis has not been undertaken to date.
2 The Foundation's requests for data were reviewed by one university's institutional re-
view board, and the other universities subsequently accepted this agreement as binding. A
separate file—which, by agreement, was accessible only to one of the authors and her as-
sistant, as well as several staff members at Mathematica Policy Research, all of whom had
completed human subjects protocol training—was later constructed that contained the
names of the students and their identification numbers. This file was used to locate ad-
dresses for individuals for whom the institutions were unable to provide current addresses,
so that these individuals could be included in the Graduate Education Survey (described
later in this appendix).