asked a series of demographic and employment questions. The response rate for the survey in 1995 was about 85% after second-wave mailings and telephone interviews; this was about a 30% increase in the response rate over 1989. Although the reduction of the sample reduced the overall number of responses from 1989 to 1995, it is believed that the increased response rate improves the quality of the data. However, the change in the survey produced a potential disjunction between data collected before 1991 and those collected since.
The sample is stratified across three variables: field of degree, sex, and a combination variable that includes degree field, sex, handicap status, ethnic group, and nationality of birth. The results of the survey are statistically analyzed to translate the data into weighted numbers for the entire population. From the weighted results, the doctorate workforce in S&E can be analyzed across different dimensions by looking at different demographic and employment characteristics and by taking different cohorts. That provides for both longitudinal and time-series analyses. However, in the analysis, one must take into consideration the change in sampling frame, the increased response rate in 1991, and the fact that some cells in an analysis could contain very few actual responses, in that the sample is only about 8% of the S&E workforce.
Data available from the SDR up to 1991 are field of doctorate and employment, sector of employment, geographic location, primary work activity, federal support, tenure status, salary data, and ethnic data. However, the 1991 SDR was administered in the fall, not the spring; some data points are not directly comparable with those from other survey years. The 1993 questionnaire incorporated substantial changes from earlier ones. In particular, the questionnaire before 1993 asked for data only as of a specific time, but the 1993 questionnaire asked for some retrospective employment information. There was also a change in the field employment questions, with much broader definitions of job categories, such as "biological scientist", as opposed to, for example, "ecologist" in the earlier surveys. As a result, the number of people in postdoctoral positions might have been slightly overestimated. In 1995, additional questions concerning detailed retrospective descriptions of the time spent in postdoctoral training were added.
The SDR is a sample survey of about 8% of PhD awards, and the number of responses might be low in some cases. A weighting formula is used to adjust the sample to the complete population. For example, a weighted response of 39 unemployed life scientists from the 26 high-quality institutions in 1995 corresponds to five responses; the 20 people working outside S&E in the same population is based on three responses. In the experience of the National Research Council's Office of Scientific and Engineering Personnel who have worked with these data for many years, a response of 10 or more provides a good estimate for a category. Although the sample is small and the analyses must be used with care, the sampling and weighting methods have been carefully developed to provide the most statistically valid results possible.
The National Science Foundation (NSF) conducts various surveys and data-collecting procedures as part of its responsibility in monitoring the state of science and engineering development in the United States. The survey that pertains most closely to graduate and postdoctoral training is the annual Survey of Graduate Students and Postdoctorates in Science and Engineering.