accelerated somewhat in the early 1990s at the same time that the number of new graduates (many of them foreign students) has increased rapidly. For a variety of reasons, the number of academic positions and traditional industrial research positions is steady or shrinking, in accord with anecdotal reports that an unusually high number of new PhDs had to change career plans on graduating or after several years in postdoctoral positions.
To some extent, the science and engineering employment situation is cyclical, and it might already be adjusting. The recent recession ended slowly, but economic growth has resumed, and the demand for skilled people is increasing even in some industries that have undergone substantial reduction and restructuring. In addition, the high rate of increase in the number of PhDs awarded to foreign citizens in the United States, which averaged more than 12% per year in the late 1980s, began to fall after 1990 and was 0.3% in 1993 (calculated from Table 3 in NSF, 1994f). The number of doctorates awarded to foreign citizens with temporary visas fell slightly in physics/astronomy, chemistry, environmental sciences, and computer sciences from 1992 to 1993 (NRC, 1995: Appendix Table A-2).
However, we have already cited some indications of basic structural changes that lower demand, including cuts in defense spending, industrial restructuring, and reductions in growth of federal R&D spending. There is no evidence that these trends of the last several years will end soon. Thus, even if PhD production does fall in the near term, science and engineering graduate students might do well to prepare themselves for an increasingly diverse set of career paths.
Supply-demand models are not now adequate for predicting whether there will be an undersupply or oversupply of trained scientists and engineers (Fechter, 1990; Leslie and Oaxaca, 1990; NSB, 1993; Vetter, 1993). That conclusion was also expressed by the panel on estimation procedures of the Committee on National Needs for Biomedical and Behavioral Research Personnel, which found that previous supply-demand models for basic biomedical, behavioral, and clinical research scientists had not proved accurate (NRC, 1994b).
At least two types of limitations of such models severely reduce their reliability, especially over the 5- to 10-year periods needed to carry out graduate-education plans. Internally, they are not based on an adequate understanding of the behavior of the students, faculty, and other people whose collective decisions affect the supply of new scientists and engineers; externally, they cannot always predict the impact of major changes in key variables outside the graduate system itself that affect demand for scientists and engineers.
For example, predictions of a huge oversupply of scientists and engineers in the early 1970s did not come true, because as a result of the predictions the students changed plans, administrators reduced programs, and graduates found new ways to use their trainingall behavioral changes that were not included in the models. More recent studies have forecast