The science, technology, engineering, and mathematics (STEM) workforce plays a crucial role in generating economic growth. Understanding the dynamics that shape this workforce and its effects on science, innovation, and economic productivity is essential.

Intellectual Merit: We construct a new data platform that enables tracking the careers and measuring the labor market outcomes of STEM PhD graduates and postdocs. The project uses detailed administrative data from a set of large U.S. research universities to develop and validate machine learning algorithms to identify STEM postdocs and postdoc training spells in the U.S. Census's Longitudinal Employer-Household Dynamics (LEHD) database. The LEHD is a longitudinal matched employee-employer database containing most U.S. workers and their employment spells.

We compare labor market outcomes and career paths of postdoc-trained PhDs and their non-postdoc-trained counterparts. As postdoctoral training is training in basic research skills, we investigate the association of postdoc training and university employment after training. For PhDs employed in industry (about half of postdocs take non-academic jobs) we evaluate the association between postdoctoral training and employment in firms that are engaged in research and development and patenting, in startups, which are often at the forefront of technological and scientific innovation, and in high productivity firms.

We estimate postdoc- and non-postdoc-trained earnings differentials eight years post-PhD by employment sector (university, industry, government, and nonprofit). Despite evidence of greater ability, postdoc-trained biomedical doctorates working in nonacademic jobs tend to earn less than their non-postdoc-trained counterparts. We show that postdoc earnings penalties can be explained nearly entirely by task mismatch, that is, the extent to which the tasks required in their employment differ from the skills learned within the postdoc or prior work.

A separate analysis focuses on the careers of academic PhDs in STEM. In collaboration with another research team, we develop novel measures of research productivity that are based on researcher publications (as opposed to earnings) but go beyond publication and citation counts. These metrics capture the breadth of the researcher's scientific impact based on the range of fields that cite the researcher's work, whether their work employs the best and latest ideas, cites the best and latest research, and whether it draws from multiple disciplines. We use these metrics to show that, among biomedical scientists after controlling for selective attrition, research productivity qualitatively declines over the career. This complements the large literature on scientific careers that focus on publication quantities over the career. We also compare careers by gender, race, and ethnicity. Using markers of career progression in publication records, we show that biomedical scientists who are female or are from underrepresented groups are less likely to show career independence relative to non-Hispanic White men.

We have conducted an analysis of the determinants of STEM labor demand in industry, focusing on the complementarity between firm R&D and STEM employment. We estimate models of STEM labor demand based on state-of-the-art econometric methods that are only possible with our new longitudinal data. We also examine the disemployment effects of COVID on STEM workers. STEM jobs showed smaller employment declines than non-STEM during the Great Recession and COVID-19. We assess how demographics, education, job tasks, remote work capability, industry, and the importance of STEM knowledge to the job contributed to this resilience. Notably, the job relevance of STEM knowledge was a key factor, while remote work feasibility and industry, which have been shown to be important in other studies of the pandemic and employment, played smaller roles.

Broader Impacts: The project creates a new comprehensive panel data set of PhDs and postdocs containing demographic, employment, and employer information. Important advantages over extant data in answering critical STEM PhD workforce questions are that it is comprehensive, including all doctoral graduates of US universities, and is linkable to other important state and federal databases, most notably detailed Census data on the employers. Some of the analyses that are possible with this information are displayed by our work. We hope its success will motivate federal and other agencies to pursue additional data linkage efforts, such as linking science funding agency administrative data to Census and tax databases. Such efforts would aid in formulating and evaluating STEM education and workforce policies (e.g., the many programs designed to increase representation in basic science).

Our documentation of earnings differentials and career paths for persons with doctoral and postdoctoral STEM training and our analysis of the determinants of STEM worker demand are pertinent to education professionals and young persons who are considering their educational and career choices. Our work on the STEM workforce and the pandemic shows that STEM training, even among the non-college educated, may protect those who have it from job loss during recessions. Our work showing the decline in scientific creativity as scientists age has implications for funding decisions at the NSF, NIH, and other institutions that support science.

Last Modified: 03/01/2024
Modified by: Gerald R Marschke