Abstract:
|
To examine the potential for enhancing the value of existing information on the process and outcomes of doctoral training in the United States, we explored linking the Survey of Earned Doctorates (SED) survey data to a variety of administrative records: university employee data and Proquest dissertation data. Three pairwise linkages were performed to link the three data sources using probabilistic record linkage algorithms on individual names. With the levels of overlap between the source data unknown, we explored different strategies of limiting the universe of possible matches using the university occupational classification, source of financial support reported in SED and field of degree. The linkage efficiency and linkage sensitivity and specificity were studied under different settings. To address the most apparent source of linkage errors from individuals with common names, we applied name frequency as a parameter for record linkage model probabilities. We then assessed the linkage quality and generalizability by inspecting the linked records, analyzing differences between the linked and unlinked records, and utilizing data borrowed from other pairwise linkage in evaluation
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.