Training the Scientific Workforce: Does Funding Mechanism Matter?

Abstract

A National Institutes of Health (NIH) taskforce recently recommended decreasing the number of graduate students supported on research assistantships, and instead favoring traineeship and fellowship funding mechanisms. Using instrumental variables estimation with survey data collected from U.S. PhD-granting biomedical sciences departments and their newly-minted PhDs, we find that increases in these programs’ NIH-funded traineeships and fellowships do significantly increase programs’ total graduate enrollments, particularly of female students. However, PhDs who were funded primarily as research assistants are significantly more likely to take research-focused jobs in the U.S. scientific workforce after they graduate, as compared to PhDs who were primarily supported as trainees or fellows. The suggested policy changes thus may have unintended, negative consequences for scientific workforce participation.

1. Introduction

In FY2012, the U.S. National Institutes of Health (NIH) funded over $30 billion in health-related research, of which $17.3 billion—56 percent—went to support research at U.S. universities and colleges. The NIH Advisory Committee to the Director (ACD) recently tasked a working group to evaluate and make recommendations to improve the diversity and sustainability of the nation's biomedical research workforce. The ACD working group's final report, posted June 2012, recommends several policy changes, including some that would change how graduate students in biomedical sciences and related fields at U.S. universities are trained and funded. Pickett et al. (2015) reiterated one of these proposals among their eight consensus recommendations, stating: “Institutions and Federal agencies should shift support of trainees toward fellowships and training grants.” However, little evidence exists to help us understand how such changes might impact subsequent retention of completed PhDs in the U.S. scientific workforce.

In this paper, we combine survey data gathered from the universe of U.S. degree-granting institutions, from biomedical sciences departments and programs, and from individuals who earned PhDs in those programs, to explore how differences in students’ sources and mechanisms of financial support in graduate school may impact their early-career retention in the U.S. scientific workforce. Specifically, we assess whether U.S.-trained PhD students whose primary mechanism of financial support in graduate school was a research assistantship, teaching assistantship, personal or family funds, or some other form of support are more or less likely to transition after graduation into scientific research-focused employment, as compared to students graduating from those programs who were supported primarily as trainees or fellows.

Our paper builds on and extends prior studies in several ways. First, in contrast with prior studies that have examined overall stay rates for foreign students graduating from U.S. higher education institutions, in this article we consider more specifically new PhDs’ retention in the U.S. scientific workforce—that is, not only whether PhDs stay in the U.S., but also whether they choose jobs where their primary work activity is basic or applied research and/or development, after completion of their PhDs. We also expand the scope and population of interest for this question beyond foreign students on temporary resident visas, to consider and compare postdoctoral employment outcomes for U.S. citizens and permanent residents as a function of their graduate school funding mechanisms, as well.

Second, to better inform NIH policy with respect to the ACD recommendations, we focus on graduate training and workforce outcomes within biological and biomedical sciences, which have had relatively lower penetration by foreign PhD students as compared to many other S&E fields. We also explore possible differential effects of graduate student funding mechanisms for U.S. versus foreign students, with particular attention to the role research assistantships may have in encouraging or discouraging completed PhDs from joining the U.S. scientific workforce.

Finally, our empirical models account for possible bias that could arise due to unobserved university-, program-, or student-level characteristics. For example, if higher-ability students are more likely to receive fellowship funding, to desire research-focused jobs, and to obtain their preferred type of employment upon graduation, or alternatively if some institutions attract higher levels of R&D funding over time resulting in a greater share of students supported as RAs as well as better career placement assistance for their graduates, such correlations could cause us erroneously to conclude some mechanisms of support are more effective than others at promoting new PhDs’ transitions into the U.S. scientific workforce. We account for these possibilities first through inclusion of university and PhD major field fixed effects, then by using two-stage instrumental variables estimation.

2. Background

Doctoral students’ enrollment, retention, and timely completion of degrees have all previously been shown responsive to availability of financial support, but financial support for graduate students enrolled at U.S. universities can be—and often is—provided via multiple different mechanisms. Fellowships typically differ from other types of student assistantships in covering tuition and providing some stipend support, without expectation of services to be performed or subsequent repayment. In our data, among the 35% of U.S.-trained biomedical sciences PhDs who said their primary source of support was a fellowship or traineeship, 27% reported no other external source of funding, and 45% held neither a research assistantship nor a teaching assistantship.

By contrast, research assistantships are typically funded by faculty members’ externally-sponsored research project grants, with salary and other benefits (e.g., tuition waiver, health insurance, etc.) provided in return for work performed. Although over half (58%) of U.S.-trained biomedical sciences PhDs graduating between 2000 and 2010 report having held a research assistantship at some point in graduate school, as shown in Table 1 only 31% identified this mechanism as their primary source of support.

Table 1.

Descriptive Statistics for individuals completing PhDs in biomedical sciences or related fields at U.S. universities, 2001 – 2010

Variable	U.S. Citizens & Permanent Residents	Foreign Temporary Residents	Overall
Percentage of all biomedical sciences PhDs	73.0%	27.0%
Female	49.6%***	44.6%	48.2%
Average time to degree (years)	6.28***	5.70	6.12
Average age at graduation (years)	31.8	32.1***	31.8
Graduated from Carnegie RU/VH institution	84.3%***	81.4%	83.5%
Graduated from private institution	39.2%	39.0%	39.1%
Unemployed, still seeking work at time of survey	18.0%	20.0%***	18.5%
Primary funding source for graduate studies
Fellowship or traineeship	38.5%***	23.7%	34.6%
Research assistantship	23.8%	50.2%***	30.9%
Teaching assistantship	8.0%	9.6%***	8.5%
Other personal earnings, or family earnings, savings, or loans	7.4%***	1.9%	5.9%
Employer reimbursement	1.9%	--	1.4%
Foreign support	--	4.0%	1.1%
Reports definite plans to remain in the U.S., in a science R&D job, after graduation	51.1%	55.2%***	52.2%
... in a “postdoc” fellowship or research associateship position, conditional on having definite plans for U.S. R&D job	83.9%	87.0%***	84.8%

Descriptive statistics for the analytic dataset of 41,580 individuals completing PhDs in biomedical sciences and related fields across our panel of 121 PhD-granting U.S. universities. Data extracted from the NSF Doctorate Records File, under restricted use license.

While the tuition benefits and take-home salaries that research assistants (RAs) and fellows receive might ultimately provide graduate students with a similar level of financial subsidy, the incentives that each of these mechanisms creates for faculty interaction and the resulting qualitative experiences of students may strongly differ. For example, RAs typically gain exposure through their work to well-designed projects focused on significant research problems, and benefit from greater direct supervision and interaction with one or more senior researchers (Worthen and Gardner, 1988). Doctoral students funded as RAs are more likely to contribute to publishing research articles before graduation, as compared to students relying on other sources of funding (Buchmueller et al., 1999; Millett and Nettles, 2006). Research publication productivity among doctoral students has also long been promoted as an indicator of students’ professional development and socialization (Harnett and Willingham, 1979).

As Millett and Nettles (2006) discuss, RAs who work with faculty on externally sponsored research projects may attract greater hands-on involvement and training from faculty members, as the latter's professional success and subsequent funding streams will depend on their productive use of current financial resources. The faculty member thus has direct incentive to train and actively manage his or her RAs, and to have them participate in production of scientific publications. In addition to the structured development of knowledge and skills the RA's on-the-job training provides, the role-modeling provided by the faculty member over the course of the project may also enhance students’ progress towards self-efficacy (O'Meara et al., 2014). RAs may also benefit from greater professional socialization and relatedly achieve a greater sense of self-efficacy with regard to prospective scientific workforce employment. By contrast, the relatively greater independence a fellowship affords could leave a student more room to flounder.

The ACD report recommends that NIH shift its support for graduate student training to place greater emphasis on its existing traineeship and fellowship mechanisms, and reduce reliance (and total NIH expenditures) on graduate student RA positions (Tilghman et al., 2012; Pickett et al, 2015). This idea has been raised before: over a decade ago, the National Research Council (2000) made the same recommendation. For students, one presumed advantage of this shift is attenuation of the positive feedback loop between universities’ total research funding and graduate student enrollments (Blume-Kohout and Clack, 2013; Stephan, 2012). NIH-funded traineeships and fellowships also may allow greater agency oversight, for example due to the formal mentoring plans required for student trainees. Students with well-developed research agendas may especially benefit from the protected time these mechanisms provide to focus exclusively on their own dissertation research, potentially facilitating more timely degree completion.

Interviews with graduate student recipients of NIH-funded traineeships and fellowships show that being able to focus attention on their studies or dissertation research is the most widely valued aspect of these mechanisms of support (National Research Council, 2005). However, the same study also revealed that deficiencies in mentoring were second only to low stipend levels among students’ stated concerns, and the study further noted that traineeship and fellowship awards do not include financial compensation for faculty mentoring activities. Thus, while faculty PIs seem to have direct incentives to expend effort on training the graduate student RAs they employ into productive members of their research teams, there may be relatively little comparable extrinsic incentive for faculty members to invest their time in mentoring students who are supported on traineeships and fellowships.

Finally, NIH's traineeships and fellowships are currently limited to U.S. citizens and legal permanent residents—students who are more likely, overall, to remain in the U.S. after graduation than those on foreign temporary resident visas. From a policy perspective, putting greater emphasis on traineeship and fellowship mechanisms (along with increasing programs’ stipend levels) might encourage more U.S. students to pursue doctorates in biomedical sciences, improving long-run sustainability of the U.S. biomedical sciences research workforce (Freeman et al., 2009; Grogger and Hanson 2013; Grogger and Hanson 2015). Conversely, decreasing the availability of RA positions on faculty investigators’ research project grants might discourage U.S. departments from admitting foreign students, or discourage admitted foreign students from enrolling, due to foreign students’ having fewer alternatives for mentored research training and financial support. Supporting this notion, across our panel of 121 U.S. universities that grant PhDs in biomedical sciences and related fields, we find that an increase in a graduate program's share of students supported as RAs is significantly and positively correlated with higher proportional enrollments of foreign temporary residents. It is not clear, a priori, whether declining federal support would be offset by any increase in institutional funds for foreign students in these fields.¹

2.1 Funding Mechanisms and Postdoctoral Career Choices

Remarkably little evidence exists on factors affecting new PhDs’ choice to pursue research-oriented scientific careers. Sauermann and Roach (2012) report that more than 1 in 5 late-stage biology and life sciences PhD students at top-tier research universities viewed non-R&D focused jobs, for example teaching-focused faculty positions or other careers, as “extremely attractive” options, even though academic research jobs were most likely to be “strongly encouraged” by faculty mentors. However, 90% of PhD students in these fields still opined that conducting basic or applied research would interest them. Gibbs et al. (2014) likewise document declining interest in academic research careers among biomedical sciences PhDs between entry and graduation, with women and historically underrepresented minorities (URMs) also expressing lower interest in faculty positions at research institutions, overall, than non-Hispanic White and Asian American men. By contrast, students with higher research self-efficacy—that is, confidence in their abilities as independent researchers—more often preferred research careers, both within and outside academia.

One earlier study found that RA positions were more likely to encourage PhD students to shift their career goals towards research, than towards other goals (Worthen and Gardner, 1988). Relatedly, Buchmueller et al. (1999) found economics PhDs who worked as RAs—and particularly those who published as students—were significantly more likely to be employed in academic, research-oriented positions after graduation. Gibbs et al. (2014) found similar results for biomedical sciences PhDs, noting higher rates of first-author publication and greater advisor investment were positively associated with academic research careers. Taken together, it seems plausible that if RAs in biomedical sciences receive more focused attention and more opportunities to publish when working directly for faculty PIs on sponsored research projects, RA positions may position graduates better for research-focused postdoctoral employment.

2.2 Foreign Students’ Postdoctoral Stay Rates and R&D-Focused Employment

Among foreign students who earned PhDs in 2005 from U.S. universities’ life sciences programs, approximately three-quarters still were in the U.S. two years after graduation (Finn, 2010). Previous literature identifies two major factors that may contribute to foreign PhD students’ retention in the U.S. scientific workforce. First, much of the growth in foreign PhD student enrollment in recent years has been among students from low- to middle-income countries (Grogger and Hanson, 2015). Students from less-developed countries with low prospective earnings or few employment opportunities in science and engineering (S&E) fields may also have relatively stronger preference to remain in the U.S. after graduation (Bound, Turner, and Walsh, 2009; Grogger and Hanson, 2015).

Second, foreign students may have relatively greater retention in R&D-focused jobs after graduation than native U.S. citizens and permanent residents, due both to visa restrictions and to comparative advantage. Peri and Sparber (2009) observe that “highly-educated immigrants, relative to native-born workers, will have imperfect language skills, knowledge of local networks, and familiarity with social norms.” So, in addition to possibly enjoying a greater range of career opportunities if they remain in the U.S., foreign students with strong quantitative skills may choose to specialize in research-focused science occupations over teaching-focused, management, or other careers, to better leverage this comparative advantage (Hunt and Gauthier-Loiselle, 2010).

Finally, Grogger and Hanson (2015) also observe that foreign students who receive university support in the form of fellowships, scholarships, research assistantships, or teaching assistantships are more likely to remain in the U.S. after they graduate, and the authors suggest this correlation is due to unobserved ability. That is, if university-supported students have higher academic ability, and if foreign students with higher academic ability are also more likely to remain in the U.S. after graduation, then differences in unobserved ability may be the root cause of any observed difference in foreign students’ postdoctoral retention across funding mechanisms.

3. Data

Data for the empirical analyses that follow are drawn from three national surveys, including the National Science Foundation's (NSF) Survey of Earned Doctorates (SED), for which responses we used are contained in the restricted-use Doctorate Records File (DRF). We combined DRF student-level data on the respondents’ doctoral institution and graduation year with additional institution-level variables extracted from the NSF-NIH Survey of Graduate Students and Postdoctorates in Science and Engineering (GSS), and from the NSF Survey of R&D Expenditures at Universities and Colleges. These datasets are described in more detail, below.

3.1 NSF Survey of Earned Doctorates (SED) – Doctorate Records File (DRF)

The SED target population is the universe of individuals who earned research doctorates from any accredited U.S. institution, in a given year. Responses to the annual SED are added each year to the DRF data file. The DRF data used in this analysis were obtained from the NSF under a restricted-use license.

We began by identifying all students who graduated with research doctorates in biomedical sciences or closely-related fields, over the period 2001 through 2010. Key variables for our analysis include each student's primary mechanism of financial support while pursuing their doctoral degree, their citizenship status, and their postdoctoral employment plans. Postdoctoral employment plans of interest include respondents’ stated intent to remain in the U.S. after graduation, and—conditional on their having definite plans for employment after graduation—what their primary work activity will be (for example, basic research, applied research, development, design, teaching, human resources, etc.). Our analytic DRF dataset contains records for over 40 thousand individuals who earned PhDs in biomedical sciences and related fields, across 121 U.S. universities, between 2001 and 2010.

Like Finn (2010), we define graduating PhDs as having “definite plans” to stay in the U.S., and—in our case—as having “definite plans” to take R&D-focused jobs, based on respondents saying that both (a) they plan to work or study in the United States after graduation, and (b) they have already signed a contract, or otherwise have a definite, firm commitment for a specific position—including “postdoc” fellowships and research associateships—with a specific employer. As such, our outcome variable—having “definite plans” for an R&D-focused job in the U.S. scientific workforce—closely reflects actual post-graduation employment patterns, as opposed to just stated intentions or preferences. This assumption is further validated by Finn's (2010) finding using linked tax data, matched to SED respondents via their Social Security numbers, of very strong correspondence between foreign PhDs’ stating at graduation their definite plans to remain in the U.S., and those individuals’ actual subsequent U.S. employment. For this study, we coded all postdoctoral research associateships as R&D-focused employment, as well as all postdoctoral fellowships where no other primary work activity (e.g., teaching) was specified by the respondent. For individuals who indicated some other, non-“postdoc” U.S. private or public sector employment, we code as R&D-focused employment those who responded that basic research, applied research, or development would be their primary work activity. We exclude from the analyses that follow any respondents who indicated they were still unemployed as of the survey (“seeking position but have no specific prospects”), representing about 18% of new U.S. citizen and permanent resident PhDs, and 20% of new foreign temporary resident PhDs. We also exclude those who planned to enroll in another full-time degree program or other unspecified training or studies, as well as those who planned neither to work nor to study, for example due to family commitments.

Descriptive statistics for these data are presented in Table 1. About 1 in 4 students completing PhDs or equivalent research doctorates in U.S. biomedical sciences and related field programs from 2001 through 2010 were foreign temporary residents at time of graduation. Of those students, about 45 percent were female, whereas among U.S. citizens and permanent residents approximately half were female. Foreign temporary resident students were slightly less likely to earn their degree from a Carnegie RU/VH institution (81% versus 84%), and averaged four months older at graduation than their U.S. citizen and permanent resident classmates (32.1 years old, versus 31.8 years old), despite their relatively shorter average time to degree completion (5.7 years, versus 6.3 years).

Table 1 also presents for comparison the proportions of U.S. and foreign students, respectively, who received their primary financial support from each of several funding mechanisms, as well as the share responding that they intend to remain in the U.S. after graduation and have definite plans for post-graduate employment in a R&D-focused job. About 1 in 3 new PhDs in biomedical sciences and related fields (35%) received their primary financial support from fellowships and traineeships. However, while these were the most common mechanisms among U.S. citizens and permanent residents, among foreign temporary residents RA positions were most common, serving as the primary source of support for about half of foreign students. By contrast, less than one-quarter of U.S. students were supported as RAs, and a similar proportion of foreign students were supported by U.S.-based fellowships. Finally, just over half of all PhDs who completed their degrees during our study period reported having definite plans for postdoctoral employment in U.S.-based science R&D occupations after graduation, with foreign students more likely to report such employment (55% of foreign students, versus 51% of U.S. students).

3.2 NSF-NIH Survey of Graduate Students and Postdoctorates in Science & Engineering (GSS)

The GSS is an annual survey of departments and other S&E graduate-degree-granting programs at U.S. academic institutions. In contrast to the individual-level SED-DRF data described above, the unit of observation in the GSS is the degree-granting department or program. We use data from this survey first to describe in greater detail the sources of funding for each mechanism among biomedical sciences graduate students, including for example the share whose RA, traineeship, or fellowship funding came from the NIH versus from other federal and non-federal sources. The GSS data provide a clearer picture of departments’ relative reliance on different sources and mechanisms of support for graduate students, regardless of students’ degree completion outcomes. In particular, whereas individuals are only eligible to respond to the SED upon earning their PhDs, the GSS data include counts of all graduate students enrolled in the responding program. Furthermore, while the SED asks individuals to indicate their primary, secondary, and any other mechanisms of financial support used in pursuing their PhDs, in recent years the survey instrument has not collected the specific funding source for each mechanism. For example, it is not possible to distinguish between NIH fellowships and institutional fellowships in the SED data, but this distinction can be made in the GSS, hence our reason for including these department- and program-level data, and merging them with the SED data (as described below).

From these data, we extracted and summed counts of first-time, full-time graduate students enrolling in biomedical sciences and related fields (see Appendix for list of fields), as well as counts of full-time graduate students in those same programs whose primary financial support came from NIH-funded research assistantships, traineeships or fellowships, by university and year. We also constructed an additional covariate by summing first-time, full-time graduate student enrollments across all other STEM fields—that is, excluding listed biomedical and related fields—for each university and year. The “Other STEM Fields” covariate provides a time-varying control for unobserved university characteristics that may impact foreign students’ share of enrollment at a given university, as well as secular shifts in total foreign graduate student enrollments at U.S. higher education institutions over time.

The 2010 GSS data indicate that over 60% of graduate students supported on NIH fellowships were enrolled in PhD-granting biological sciences departments or programs, and another 24% were enrolled in medical, other life sciences, and psychology programs. About 10% were in chemistry, chemical engineering, and other/unspecified engineering fields (which notably includes biological and biomedical engineering). A similar pattern holds for NIH-funded trainees, with over 70% found in PhD-granting biological sciences departments or programs, and comparable but smaller percentages distributed in the other fields noted above.

NIH-funded research assistantships support more graduate students than fellowships and traineeships combined, with a somewhat broader array of disciplines represented among funded students and programs. Still, some 56% of NIH-funded research assistantships were housed in PhD-granting biological sciences departments, and overall 89% of NIH-funded research assistantships were in PhD-granting departments or programs in the same set of fields listed above. As such, although increases in a university's total NIH R&D funding may also result in support for some graduate students in “Other STEM Fields,” these students comprise only a small fraction of students funded by NIH, and furthermore, NIH supports only a very small percentage of the total graduate students in those other disciplines.

As shown in Table 2, for our analytic panel of 121 institutions granting PhDs in biomedical sciences or related fields, the average aggregate graduate student enrollment during the period 1998-2010 across those institutions was about 465 students, with over 90 students entering these programs for the first time each year. About 80% of students in these programs were enrolled full-time, and over three-quarters (77%) were U.S. citizens or permanent residents. Among full-time students in these programs, about 20% were supported by the NIH, over half of those on research assistantships. Only about one-third of NIH-funded students, representing less than 7% of total full-time biomedical sciences and related fields graduate enrollment, were supported by NIH traineeships or fellowships.

Table 2.

Descriptive Statistics for analytic panel of 121 U.S. universities with departments or programs granting PhDs in biomedical sciences or related fields, 1998 – 2010

Variable	Mean	St. Dev.
Total graduate enrollment, biomedical sciences fields	464.8	440.9
Full-time enrollment, biomedical sciences fields	379.4	364.9
First-time, full-time enrollment, biomedical sciences fields	91.4	98.7
U.S. citizens and permanent residents, percent of full-time enrolled	79.3%	11.8%
NIH-funded research assistants, percent of full-time enrolled	12.3%	11.1%
Graduate students funded as research assistants, all funding sources, percent of full-time enrolled	35.4%	19.1%
NIH-funded trainees and fellows, percent of full-time enrolled	6.84%	8.17%
Graduate students funded as trainees or fellows, all funding sources, percent of full-time enrolled	21.9%	20.6%
Percent of full-time students with any NIH support	20.3%	17.4%
Federal life sciences R&D funding, millions $2010	94.5	114.5
Private institutions	44 (36.4%)
Institution's Carnegie 2005 Classification: Doctorate-granting Research Universities – Very High	64 (53.7%)

Descriptive statistics for the analytic panel dataset of 121 PhD-granting U.S. universities, with data extracted from the NSF-NIH Survey of Graduate Students and Postdoctorates in Science and Engineering. Standard deviations are reported between panel institutions. Graduate student counts and percentages are for all PhD-granting programs in biomedical sciences fields at a given university, in a given year.

3.3 Analytic Dataset Construction

To construct our analytic datasets, we merged observations by university, field (biomedical sciences and related, or other STEM), and year, using institution FICE codes followed by hand-matching for the remainder. Our initial panel included 189 institutions that grant advanced degrees in biomedical sciences and related fields, but because our panel data empirical methods exploit variation within each university over time, we excluded universities that reported zero NIH research assistantships throughout our study period, 1998 through 2010, in the GSS data, as these provided no variation to exploit. Then, because the SED-DRF data only captures PhD completions (that is, we are unable to observe terminal Master's degrees), we excluded 33 institutions for which no PhDs were observed in biomedical sciences or related fields in the SED-DRF data during our study period. Together, these exclusions resulted in analytic datasets with, respectively, biomedical sciences PhDs and PhD-granting programs across 121 PhD-granting institutions.

4. Empirical Methods

Our empirical analysis proceeds as follows. First, we demonstrate empirically our intuitive notion that increases in total NIH R&D investment at a given university increase the share of students in biomedical sciences and related graduate programs funded as research assistants. Then, having established relevance of NIH R&D funding as a predictor of availability of NIH research assistantships as a mechanism for graduate student support, we evaluate how demographics of full-time graduate student enrollment change across universities’ biomedical sciences programs, given a unit change in the number of NIH-funded traineeships and fellowships, versus a unit change in the number of NIH-funded research assistantships. Finally, we combine institution-level changes in availability of NIH-funded research assistantships and R&D intensity (R&D expenditures per capita) as instruments to predict probability that a given newly-minted biomedical sciences PhD received primary financial support in graduate school from research assistantships (any funding source), versus from other mechanisms including traineeships or fellowships, personal or family funds, teaching assistantships, employer reimbursement, and so on. These instruments are used in each of the two-stage instrumental variables (IV) estimation strategies we use to predict probability of a U.S. R&D job: a standard two-stage IV linear probability model, simultaneous two-stage GMM IV system estimation, and Lewbel's (2000) special regressor method.

We begin with simple descriptive estimation of the change in full-time graduate student enrollment that occurs with a one-unit change in NIH-funded traineeships and fellowships, versus that for NIH-funded research assistantships. Due to strong evidence of first-order autocorrelation when estimating models with university fixed effects, we present results from the simple first-differenced equation:

$$F{T}_{u,t}-F{T}_{u,t-1}=\alpha +\rho (R{A}_{u,t}-R{A}_{u,t-1})+\phi (T{F}_{u,t}-T{F}_{u,t-1})$$ (1)

where FT_t is the number of full-time graduate students enrolling in year t, RA_t is the number of full-time graduate students supported on NIH-funded research assistantships in year t, and TF_t is the number of full-time graduate students supported on NIH-funded traineeships or fellowships in year t. In addition to estimating this equation for the full population of graduate students, we also estimate it separately for the subpopulations of foreign temporary residents, U.S. citizens and permanent residents, and U.S. women students. The purpose and advantage of equation [1] is that the measure of enrollment encompasses both first-time (new) enrollments, but also students retained in the program. If either ρ or φ is significantly less than one, this may indicate that NIH funding sources have crowded out other institutional sources of funding for graduate students.

Next, we evaluate how increasing the number of students supported by NIH on research assistantships affects the demographics of incoming graduate student cohorts. Because (as discussed above) demographic diversity and federal R&D funding may be endogenously determined, our identification and estimation strategies here provide two key improvements over those for equation (1). First, recognizing that the dramatic increase in NIH-funded research assistantships was—from the funding agency perspective—an unintended consequence of increasing NIH funding for R&D expenditures at U.S. universities during the “doubling” period, 1998 through 2003, our expectation and interest reside with universities whose funding and enrollment patterns were susceptible to changes in their R&D funding. In essence, Angrist and Krueger's (2001) Local Average Treatment Effect (LATE) describes precisely the effect of policy interest. We therefore employ instrumental variables (IV) estimation, employing two instrumental variables that were previously validated for use in instrumenting NIH R&D funding: universities’ predicted NIH funding, and their Congressional representation on appropriations subcommittee with responsibility for NIH Institute and Center budgets (Blume-Kohout et al, 2015). These two IVs are used to predict changes in the number of NIH-funded research assistants at the university, per equation (2) below.

$$R{A}_{u,t}=\beta NI{H}_{u,t-1}+\gamma C{R}_{u,t-1}+\delta O{F}_{g,u,t}+{\theta }_u$$ (2)

$$NewF{T}_{g,u,t}=\delta R{A}_{u,t}+\omega O{F}_{g,u,t}+{\eta }_u$$ (3)

Then, to control for unobserved characteristics that may influence universities’ relative attractiveness to foreign students and women, as well as overall university-year changes in total graduate enrollment, we add the contemporaneous “Other STEM Fields” covariate described above, OF_g,u,t, which is the number of entering full-time graduate students in other science & engineering fields for demographic group g, at university u, in year t.

4.1 Effects of Different Primary Funding Mechanisms on Early-Career Occupational Choice

Having demonstrated relevance of changes in R&D funding and availability of research assistantships for each university's share of graduate students primarily funded as RAs, we turn next to estimate separately for U.S. citizens and permanent residents and then for foreign temporary resident students the relative impact of RA funding versus other mechanisms of support on the student's probability of taking a R&D-focused job in the U.S., after completing the PhD.

We begin by estimating a linear probability model (LPM) with university and year fixed effects:

$$\mathrm{Pr}(USRnDJo{b}_i=1\mid {\mathrm{x}}_i,{\tau }_t,{\nu }_u,{\pi }_f)={\phi }_{1}R{A}_i+{\phi }_{2}T{A}_i+{\phi }_{3}E{R}_i+{\phi }_{4}S{F}_i+{\phi }_{5}O{T}_i+{\mathrm{x}}_i\mathrm{b}+{\tau }_t+{\nu }_u+{\pi }_f$$ (4)

This model predicts the probability that student i, who earned his or her research doctorate from university u in year t in biomedical sciences or a related field, reports definite plans to take a research-focused job in the United States after graduation, USRnDJob_i,u,t = 1, versus definite plans for some other employment in the U.S. or abroad, conditional on: the student's graduation year, t; the university awarding the student's degree, u; the student's PhD major field-of-degree, f; a vector of individual characteristics, x, such as student's gender and age; and the student's primary mechanism of financial support while pursuing the PhD (RA=research assistantship, TA = teaching assistantship, ER = employer reimbursement, SF = personal or family savings, earnings, or loans, OT = all other sources).

Inclusion of university, PhD major field of degree (biological or medical sciences, excluding animal and plant sciences; biochemistry or biophysics; animal or plant sciences; or biological or biomedical engineering), and year fixed effects control respectively for differences across universities in their placement rates for research-focused jobs, differences across subfields in propensity towards research-focused employment, and secular differences over time in the relative availability of research-focused jobs for new PhDs, including changes in visa availability for foreign temporary residents over time.

Finally, the empirical model for U.S. citizens and permanent residents also includes in x indicators for the student's race/ethnicity. Recent research demonstrates significant differences across racial and ethnic groups as well as by gender in career interests of biomedical sciences PhDs (Gibbs et al., 2014). The empirical model for foreign temporary resident students includes in x an indicator for having earned one's first bachelor's degree in the U.S., and a set of country of citizenship indicators to control for differences across countries in macroeconomic conditions, employment opportunities, and U.S. work visa availability. Along with the PhD field-of-degree indicators, inclusion of country of origin as an explanatory variable helps us to avoid potential confounding due to differences across fields in foreign students’ participation by country of origin, and correlated country-specific differences in stay rates.

The LPM in equation (4) is attractive due to its relatively straightforward interpretation, but it has well-known disadvantages, including heteroskedasticity, possible negative predicted probabilities, and inconsistent signs on the marginal effects (Greene, 2008; Lewbel et al., 2012). To accommodate heteroskedasticity, we estimate and report cluster-bootstrap standard errors robust to both arbitrary heteroskedasticity and two-way clustering on graduation year and detailed PhD field of degree. To assess consistency of the LPM estimates, we also estimate and present marginal effects from similarly-specified probit models, with dummy variables representing university fixed effects.²

We remained concerned that RA funding may be endogeneous, specifically that there might exist unobserved individual-level variation in student preferences for particular work activities that affect both the probability that their primary support comes from a RA position and the probability they take a research-focused job after graduation. Interestingly, when we employ two-stage IV estimation for the full model described above—including university fixed effects—we find no evidence of such endogeneity (p>.48). To the extent there is selection on preference for research activity, it seems to be revealed and fully accounted for through students’ choice of graduate program. We do, nonetheless, provide our results from two-stage GMM IV estimation in the Results section that follows.

For this approach, we needed instruments strongly correlated with probability of a student receiving primary support from a research assistantship, but that provided no additional explanatory power for students’ postdoctoral placements in R&D-focused jobs. Intuitively, if the university where a student earns his or her PhD experiences an increase in its share of graduate students funded by the NIH as research assistants during the period after he or she enrolls—which helps protect us against students’ selecting into programs on that basis—but before the student's year of graduation, then all else equal, we'd predict a higher likelihood that when he or she graduates, the student will report RA funding as his or her primary mechanism of financial support. We therefore propose as instruments the first through third lags, relative to student i's graduation year, of university u's percent of full-time biomedical sciences graduate students primarily funded on NIH research assistantships. In addition, because the number of RA positions overall (both NIH- and non-NIH-funded) is partly driven by changes in total life sciences research intensity, we also considered per capita life sciences R&D expenditures and per capita predicted NIH funding in each student's final year as candidate instruments. The relevance condition for these instruments is easily tested, and we do so, finding the first-stage F-statistics for the excluded instruments exceeds 16 for the full sample. To test the exogeneity condition, we take advantage of overidentification provided by the multiple instruments. Hansen's J-statistic indicates no reason for concern, as we fail to reject the null hypothesis of exogeneity with p>.38 in all cases.

Finally, because foreign students’ employment in the U.S. scientific workforce reflects both selection to stay in the U.S. after graduation and some preference for R&D-focused employment, either or both of which may be correlated with PhD students’ primary funding mechanisms, we conclude our empirical analysis with two-stage general method of moments (GMM) IV system estimation of two simultaneous equations, predicting respectively: (1) probability of remaining in the U.S. after graduating with a biomedical sciences PhD; and (2) probability of a U.S. R&D job, conditional on definite employment plans. These models include the same demographic variables, field of degree, and year fixed effects as described above. In addition, because we find PhDs who stay in the U.S. are more likely that those who leave to have definite employment plans for research-focused jobs, we include an additional indicator variable for taking any (U.S. or foreign) R&D job in the model predicting staying in the U.S., overall. We treat this variable as potentially endogenous. With this variable included, the set of university fixed effects becomes jointly insignificant (p=.32) for predicting foreign students’ probability of staying in the U.S. after graduation, so we exclude university fixed effects from that equation to simplify GMM estimation. However, we do retain university fixed effects in the second equation predicting probability of a U.S. R&D job. Excluded instruments for both equations include per capita predicted NIH R&D funding in the student's final year, and the first through third lags of the institution's percent of full-time graduate students supported as NIH-funded RAs.

When modeling binary outcomes in the presence of endogeneity, one common solution is to identify appropriate instrumental variables as we have done above, and then to use them to estimate IV probit models. However, this approach can be problematic if the endogenous explanatory variable is also binary—as in our case, with RA funding—as it can yield inconsistent parameter estimates (Lewbel et al., 2012). To address this problem, Lewbel (2000) presented an alternative estimation technique called the special regressor method, which was recently implemented for Stata (Baum, 2012), and is described in more detail in the Appendix. We validate our key results using this approach.

5. Results

We began by estimating effects of changes in NIH R&D funding on the number of full-time graduate students employed as NIH-funded research assistants the following year, using IGMM. Our estimated elasticity of R&D funding is positive and highly significant: 1.27, p=.001. T-test for unit elasticity fails to reject the null hypothesis: that is, our evidence suggests a 1 percent increase in NIH R&D funding in year t-1 yields a proportional, 1 percent increase in the number of NIH-funded research assistantships in year t.

Below, we describe our results from university- and student-level empirical analyses of the effects of research assistantships versus traineeships and fellowships on both the demographics of graduate students enrolled in PhD-granting biomedical sciences and related programs, and on probability of taking a R&D-focused job in the U.S. scientific workforce after earning a PhD from one of these programs.

5.1 Effects of Funding Mechanisms on Graduate Student Enrollment

Table 3 presents results from our investigation of possible crowding-out of alternative graduate student funding sources, when NIH support for graduate students increases. We find that NIH-funded traineeships and fellowships increase full-time graduate enrollment by essentially 1:1 (p<.05), with this effect concentrated (unsurprisingly) among U.S. citizens and permanent residents. Each additional NIH traineeship or fellowship is associated with a 0.885-student increase in full-time enrollment of U.S. citizens and permanent residents, of whom over half (estimated 54 percent, 0.476 divided by 0.885) are women. The small spillover effects we see for foreign students (0.143 foreign students added or retained for each NIH traineeship or fellowship, p<.10) may simply reflect departments’ increased capacity to move institutional, research assistantship, and other funds to foreign students when traineeships and fellowships are awarded to domestic students.

Table 3.

Changes in Total Full-Time Graduate Student Enrollment due to Changes in NIH-Funded Research Assistantships, Traineeships and Fellowships, Biomedical Sciences Graduate Programs, 1998-2010

	All Students	Foreign Temporary Residents	U.S. Citizens and Permanent Residents	U.S. Women
Graduate Students Supported by NIH as Research Assistants	0.516** 0.261	0.165*** 0.0382	0.284 0.227	0.183 0.179
Graduate Students Supported by NIH as Trainees or Fellows	1.082** 0.446	0.143* 0.0756	0.885** 0.380	0.476** 0.228
Observations	1452	1440	1212	1319

^***p<0.01

^**p<0.05

^*p<0.10

Results from first-differenced linear panel estimation. The dependent variable is the total number of full-time graduate students enrolled in PhD-granting biomedical sciences graduate programs at a given university, in a given year. Explanatory variables are the total number of students supported by NIH in those same departments and programs, in the same year, on research assistantships or traineeships/fellowships, respectively. Standard errors reported below each coefficient estimate are robust to both arbitrary heteroskedasticity and clustering at the university level.

Increases in the number of NIH-funded research assistantships have a relatively weaker effect on graduate students’ full-time enrollments. On average, universities must add two RA lines to increase total enrollment by one student (coeff. estimate 0.516, p<.05). Furthermore, we find no significant effect of RA positions overall for U.S. students’ enrollment, whereas the effect of an increase in NIH-funded research assistantships on foreign student enrollment is highly significant (coeff. estimate 0.165, p<.001). Interestingly, although foreign students represent less than a quarter of full-time enrollment in these programs, the effect of an increase in research assistantships for foreign students’ enrollment appears disproportionately high. Foreign students comprise about 32 percent of the full-time students added or retained, given a unit increase in the number of NIH-funded research assistantships (0.165 divided by 0.516 for total enrollments).

The correlational results in Table 3 do not control for unobserved heterogeneity across universities and programs in their attractiveness to particular student demographic groups, nor other trends over time in admissions of students by race, gender, and citizenship. To address these concerns, in Table 4 we present results from 2SLS IV estimation. As the first-stage F-statistics indicate, our two instrumental variables—predicted NIH funding and Congressional representation—are highly relevant predictors of changes in universities’ NIH research assistantships. In contrast, these instruments are relatively poor predictors of changes in NIH-funded traineeships and fellowships. We therefore limit our focus in this analysis only to NIH-funded research assistantships. In addition, to provide additional assurance that we are removing any secular demographic trends in enrollment that may be spuriously correlated with changes in universities’ enrollments of foreign students and women, we include as a time-varying covariate the number of first-time, full-time students from the same demographic group that enrolled in other STEM fields graduate programs at that university, in the same year.

Table 4.

Changes in First-Time, Full-Time Graduate Student Enrollment with Increase in NIH-Funded Research Assistantships, Biomedical Sciences Graduate Programs, 1998-2010

	All Students	Foreign Temporary Residents	U.S. Citizens and Permanent Residents	U.S. Women
Graduate Students Supported by NIH as Research Assistants	0.566** (0.254)	0.331*** (0.0689)	0.130 (0.219)	0.0799 (0.156)
Entering Graduate Students, Other Fields	0.158*** (0.0291)	0.0647*** (0.0122)	0.233*** (0.0378)	0.378*** (0.0465)
Observations	1,331	1,324	1,154	1241
First-Stage F-statistic	25.6	26.16	25.1	25.05
Hansen's J-statistic p-value	0.655	0.947	0.513	0.514

^***p<0.01

^**p<0.05

* p<0.10

Results from two-stage least squares instrumental variables (2SLS IV) estimation, with predicted NIH funding and Congressional representation as instruments for the total number of graduate students in the university's PhD-granting biomedical sciences and related fields graduate programs funded by NIH. The dependent variable is the number of first-time, full-time students entering those PhD-granting biomedical sciences graduate programs, by university and year. All variables are first-differenced. Standard errors reported below each coefficient estimate are robust to arbitrary heteroskedasticity and clustering at the university level.

We find that increases in programs’ NIH-funded research assistantships do increase the number of first-time, full-time graduate students enrolled in those programs. However, these effects are once again stronger for foreign student enrollment than for domestic students: for every three NIH-funded RA positions added, one additional foreign student is admitted and enrolled (point estimate 0.331, p<.01). On the other hand, we find no significant effect of an increase in RA positions on contemporaneous first-time enrollments of U.S. citizens and permanent residents, overall.

5.2 Impact of U.S. Citizen and Permanent Resident Students’ Primary Funding Mechanism on Early Career Occupation Choices

In Table 5, we present results from linear probability (LPM), probit, and two-stage GMM IV estimating the impact of U.S. citizen and permanent resident students’ primary funding mechanism on their probability of taking research-focused employment in the U.S. scientific workforce after graduating with their PhDs. Both the LPM and probit models include university fixed effects to control for university characteristics that may correlate with both the probability of RA funding and students’ subsequent employment. As described below, inclusion of university fixed effects had very little impact on our estimates of the effects of different funding mechanisms, though it did remove statistical evidence of endogeneity of RA funding.

Table 5.

Marginal Effects of Predictors of U.S. Citizen or Permanent Resident Students Taking a U.S. R&D Job, after earning a PhD in Biomedical Sciences or Related Fields, 2001 – 2010

	Linear Probability Model with University Fixed Effects	Probit Model with University Fixed Effects	Pooled Two-Stage GMM IV Estimation
Student's Primary Source of Financial Support (Reference Group: Traineeship or Fellowship)
Research Assistantship	0.0464*** 0.0074	0.0461*** 0.0070	0.112* 0.0673
Teaching Assistantship	−0.0810*** 0.0134	−0.0801*** 0.0139	−0.0621* 0.0351
Personal or Family Earnings, savings or Loans	−0.0765*** 0.0140	−0.0732*** 0.0131	−0.0489 0.0320
Employer Reimbursed	−0.0030 0.0230	−0.0054 0.0208	0.0388 0.0333
All other support	0.0039 0.0090	0.0030 0.0074	0.0242 0.0244
Female	−0.0118* 0.0062	−0.0117** 0.0060	−0.0108 0.0076
Age (Demeaned, Negative)	0.0064*** 0.0008	0.0058*** 0.0007	0.0066*** 0.0009
Observations	19,843	19,843	19,843

^***p<0.01

^**p<0.05

^*p<0.10

Binary dependent variable for definite postdoctoral employment in the United states with R&D primary work activity, versus other definite employment plans, among PhDs surveyed at time of graduation from U.S. doctoral programs in biomedical sciences and related fields. standard errors presented below each marginal effects estimate are robust to arbitrary heteroskedasticity and clustering on detailed PhD field of degree. For fixed effects models, standard errors are cluster-bootstrapped with 400 replications, with two-way clustering on detailed PhD field of degree and graduation year. Two-stage IV estimation uses five excluded instruments in the first stage: per capita predicted NiH funding and university-demeaned per capita total life sciences research expenditures in the student's final year, and the first through third lags of the institution's percent of students funded as RAs on NiH awards. All models also include indicators for race/ethnicity, PhD major field, and graduation year (not shown).

We observe that U.S. citizens and permanent residents supported as research assistants have significantly higher probability of taking U.S. jobs with their primary work activity R&D after graduation, as compared to trainees or fellows. With both LPM and probit models, we find the probability of taking a U.S. R&D job is about 4.6 percentage points higher among PhDs who were primarily supported as RAs, versus among those earning PhDs from the same program in the same year, but whose primary support came from traineeships or fellowships. Point estimates from LPM and probit models excluding university fixed effects (not shown) are strikingly similar: 4.7 and 4.8 percentage points higher probability of a U.S. R&D job for RAs, respectively. All reported estimates retained the same level of statistical significance in models with standard errors clustered instead on PhD-granting institution, instead of detailed PhD field and graduation year.

As noted above, we found no evidence of endogeneity of RA funding in the university fixed effects models. On the other hand, because prospective students desiring research-focused careers might also be attracted to programs featuring relatively greater or lesser reliance on RA support, relying on variation in outcomes within alumni of each institution to identify the impact of graduate student RA support may underestimate the total effect of these mechanisms for scientific workforce participation. We therefore also used two-stage IV estimation to estimate a pooled model, correcting for possible endogeneity of RA funding. Results from this pooled model are presented in the right-most column of Table 5, and can be considered as an upper bound for the total effect of RA versus traineeship or fellowship funding.

As shown in Table 5, among U.S. citizen and permanent resident PhDs, the estimated probability of taking a U.S. R&D job after graduation was 11 percentage points higher for RAs than for trainees or fellows graduating in the same field and year, conditional on the respondent having definite employment plans. Our alternative two-stage IV estimation using Lewbel's special regressor method on university-demeaned data yielded nearly identical results: the probability of a U.S. student with primary RA support taking a U.S. R&D job was 11 percentage points higher (p<.05) than for a U.S. student who was primarily supported on a traineeship or fellowship, and who graduated from the same program, in the same year.

5.3 Impact of Foreign Temporary Resident Students’ Primary Funding Mechanism on Early Career Occupation Choices

Table 6 presents similarly strong positive effects of RA positions for foreign students’ early career retention in the U.S scientific workforce. The LPM estimates 5.8 percentage points higher probability (p<.01) of a U.S. R&D job among foreign students primarily supported as RAs versus those supported primarily by U.S.-based fellowships.³ Once again, the similarly-specified probit model (not shown) yielded very similar results to the LPM. But in contrast with our earlier finding for U.S. students, here we observe no significant negative effect of teaching assistantships as a primary means of foreign student support, relative to fellowship funding. In addition, other graduate assistantships, dissertation grants, or internships appear to yield similarly higher probability of a U.S. R&D job among foreign students. Using Lewbel's two-stage IV special regressor method with university-demeaned data, the estimated effect was somewhat larger (albeit more similar to that found for U.S. citizens and permanent residents): 10 percentage points higher probability of a U.S. R&D job for foreign students supported primarily as RAs, versus those supported primarily on fellowships.

Table 6.

Marginal Effects of Predictors of Foreign Temporary Resident Students Taking a U.S. R&D Job, after earning a U.S. PhD in Biomedical Sciences or Related Fields, 2001 – 2010

	Linear Probability Model with University Fixed Effects	Two-Stage GMM IV Estimation
		Outcome: Stay in U.S.	Outcome: U.S. R&D Job
Student's Primary Source of Financial Support (Reference Group: Traineeship or Fellowship)
Research Assistantship	0.0584*** 0.0124	0.0226* 0.0123	0.381** 0.0978
Teaching Assistantship	0.0139 0.0194	0.0148 0.0117	0.238*** 0.0718
Personal or Family Earnings, Savings or Loans	0.0137 0.0403	0.0031 0.0282	0.238*** 0.0663
Foreign support	−0.233*** 0.0281	−0.256*** 0.0178	−0.0863 0.0559
All other support	0.0461***	0.0277**	0.232***
	0.0151	0.0123	0.0558
Female	−0.0176** 0.0073	0.0046 0.0049	−0.0105* 0.0057
Age (Demeaned, Negative)	0.0099*** 0.0015	0.0060*** 0.0009	0.0095*** 0.0010
Bachelor's Degree Earned in U.S.	0.0447** 0.0186	0.0748*** 0.0120	0.0456*** 0.0136
Observations	7,567	7,567	7,567

^***p<0.01

^**p<0.05

^*p<0.10

Binary dependent variable in columns (1) and (3) represents definite postdoctoral employment in the United States with R&D primary work activity, versus other definite employment plans. Columns (2) and (3) present results from two-stage GMM IV system estimation of simultaneous equations predicting binary dependent variable for definite plans to stay in the U.S. (column 2) and definite plans for a U.S. R&D job (column 3). Standard errors presented below each coefficient estimate are robust to arbitrary heteroskedasticity and two-way clustering on detailed PhD field of degree and graduation year. For the fixed effects model in column (1), these are cluster-bootstrapped standard errors based on 400 replications. Added instruments for the GMM IV system include per capita predicted NIH funding in the student's final year, and the first through third lags of the institution's percent of students funded as RAs on NIH awards. Models (1) and (3) both include university fixed effects. All models also include indicators for country of citizenship and year fixed effects, not shown.

The predicted probability of a foreign temporary resident student taking a U.S. R&D job after graduation can be described as the product of their probability of staying in the U.S. after graduation, and conditional on staying in the U.S., their probability of choosing an R&D-focused job. Two-stage GMM IV estimation of simultaneous equations predicting each of these outcomes separately finds that foreign students whose primary financial support came from RA positions, internships, or other assistantship funding are slightly more likely to remain in the U.S. after graduation, as compared to their foreign classmates graduating from the same program in the same year, but whose support came primarily from fellowships. We find no significant difference in stay rates for students with RA funding versus TA or other assistantship funding.

On the other hand, those whose primary support came from some foreign source (government, employer, etc.) were substantially less likely to stay in the U.S., perhaps due to conditions of those mechanisms of support. In addition, although foreign students with U.S.-earned bachelor's degrees and those younger than average at graduation were more likely to stay in the U.S., we observe no significant difference in stay rates for men versus women.

Taking these differences in stay rates across funding mechanisms into account via simultaneous equations GMM IV estimation, we find substantially higher propensity for U.S. R&D jobs among foreign students whose primary support came from a research assistantship versus any U.S.-based fellowship. Foreign students supported primarily as RAs had 15 percentage points higher probability of definite plans for a U.S. R&D job after graduation, as compared to foreign students graduating from the same program in the same year with any other mechanism of support (all differences statistically significant at p<.01). Compared to their foreign classmates primarily supported on fellowships, the probability of a U.S. R&D job was 38 percentage points higher. Like U.S. women, foreign women had slightly lower probability (1.2 percentage points, p<.05) of taking a U.S. R&D job after graduation.

The results summarized above are robust to several alternative formulations and estimation strategies, not shown. First, due to concerns about consistent IV estimation in the presence of a binary outcome variable and binary endogenous variables, as noted above we re-estimated our models using Lewbel's special regressor method, described in the Appendix. For U.S. citizen and permanent resident newly-minted PhDs, we estimated the conditional probability of a U.S. R&D job after graduation is 11 percentage points higher (p<.05) for those primarily supported as RAs, versus on either traineeships or fellowships. For foreign students, we similarly estimate the probability of a U.S. R&D job is 10 percentage points higher (p<.05) for those primarily supported as RAs versus on U.S.-based fellowships.

Second, because temporary “postdoc” positions comprise 84% of new biomedical sciences PhDs’ first postdoctoral employment in US R&D jobs, and because for visa reasons these temporary positions may be easier for foreign workers to obtain, we re-estimated our models to predict regular employment in a non-“postdoc” R&D-focused position, versus all other definite employment. Descriptively, we do find that foreign temporary residents are significantly more likely to take research postdoc positions than otherwise-similar U.S. citizens and permanent residents, resulting in foreign students’ 1.6 percentage points lower probability of “regular” (non-postdoc) research-focused employment after graduation. However, consistent with our results above, we find that both U.S. and foreign PhDs are more likely to obtain regular R&D-focused employment in the U.S. scientific workforce if they were primarily funded as RAs, versus as trainees or fellows. For U.S. citizens and permanent residents, this appears to reflect an overall greater tendency towards research-focused positions among those supported as RAs, as similar positive impact is found on probability of a research-focused postdoc versus other employment. Among foreign students, however, the benefit of RA support over traineeship or fellowship funding appears to reside specifically in their higher probability of “regular” research-focused employment after graduation, as there is no significant difference in probability of taking a research-focused postdoc position versus other employment among foreign students supported as trainees, fellows, or RAs.

Finally, although NIH-funded fellowships and traineeships do not allow departments to require additional work (e.g., as teaching assistants or research assistants) from awardees, many students do rely on a mix of funding mechanisms for their financial support in graduate school, as we discussed in section 2. To address possible contamination of our RA “treatment” group and trainee/fellow “control” groups, we re-estimated the models limiting the base group to individuals with traineeships and fellowships who reported no TA or RA funding support, and separated out students who reported serving only as RAs (with no TA, trainee or fellowship support) from those who reported RA positions in conjunction with any of these other types of support. As one would expect given the results reported above, the estimated effect of RA funding becomes even more stark. For U.S. citizens and permanent residents, the LPM with university fixed effects estimates 9.3 percentage points higher probability of a U.S. R&D job for students purely funded as RAs, versus those purely funded as trainees or fellows. Moreover, those who received support from RA positions as well as from traineeships, fellowships, or other assistantships had 2.8 percentage points (p<.001) higher probability of a U.S. R&D job, compared to those funded only as trainees or fellows. Among foreign temporary resident students, the LPM estimated effect likewise increases to 6.9 percentage points (p<.001) for those solely supported as RAs versus those supported solely on fellowships.

6. Conclusions

In this paper, we demonstrate that graduate students’ mechanisms of financial support matter not only for the number and demographic mix of students enrolling in U.S. doctorate-granting programs, but also with respect to individual PhDs’ early career plans for employment in the U.S. scientific workforce. First, we observed that the number of NIH-funded research assistantships at PhD-granting institutions increases proportionally with increases in NIH R&D funding, such that a 1% increase in R&D funding yields, on average, about a 1% increase in graduate student RA positions. These changes in institutions’ NIH-funded research assistantships, in turn, yield disproportionate contemporaneous effects on enrollment of foreign graduate students, with 1 additional foreign student enrolled for every 3 RA positions added. By contrast, we observe no significant effect of institution-level changes in RA positions (as driven by changes in R&D funding levels) on enrollment of U.S. citizen and permanent resident students. However, both U.S. and foreign students have significantly higher probability of taking a research-focused job in the U.S. scientific workforce after graduation if their primary funding mechanism in graduate school was a RA position, as opposed to traineeship or fellowship funding.

Foreign temporary residents are significantly more likely than U.S. citizens and permanent residents to take “postdoc” positions after their PhDs, comprising almost 30% of postdocs versus 22% of new PhDs in other employment. Because visas for such temporary positions are relatively easier to obtain—especially for those in academia, medical schools, or university-affiliated research institutes—this outcome should not necessarily be interpreted as evidence of foreign students having greater long-run attachment to scientific R&D jobs. However, whereas RA positions appear to increase probability of both regular (non-postdoc) research-focused employment and research-focused postdocs among new U.S. citizens and permanent resident PhDs, for foreign students we find the RA difference is driven by their higher probability of regular, non-postdoc employment in the U.S. scientific workforce. Further examination reveals many of these individuals are hired by industry rather than academic employers, suggesting RA positions may play a particular role in foreign students’ professional socialization as employees in a U.S.-based scientific research enterprise.

From a policy perspective, NIH-funded traineeships and fellowships clearly do provide a more direct means for controlling total graduate enrollment levels, with each additional traineeship or fellowship at a given institution associated with one additional graduate student enrolled. That is, we find no evidence that traineeship or fellowship funding crowds out other institutional funding sources for graduate students. These NIH traineeship and fellowship enrollment effects unsurprisingly are concentrated among U.S. citizens and permanent residents, to whom these mechanisms of support are restricted, though the small positive spillover effect we observe on enrollment of foreign students suggests some shifting or substitution of funding sources may occur across students within each university. By contrast, about half of NIH-funded RA positions seem simply to crowd out graduate student support from other funding sources.

U.S.-trained PhD students in biomedical sciences and related fields who were primarily supported as RAs in graduate school have between 4.6 and 11 percentage points higher probability of taking a U.S. R&D job after graduation, as compared with those supported on traineeships or fellowships funded by NIH or other U.S. (non-foreign) sources. One possible explanation for this early-career outcome gap is that productive research assistantships may provide a sort of commitment device for graduate students, increasing their involvement and professional socialization. Early work by Worthen and co-authors (1976; 1988) emphasizes the importance of genuine apprenticeship experiences to future scholar-scientists, and finds research assistantships do facilitate and encourage PhDs towards research careers. However, RAs who do not participate in writing research proposals or articles, who do not present their work at conferences, or who do not ultimately progress towards complex analysis or design of experiments may not receive these supposed benefits.

One limitation of this analysis is the unexplored possibility that disparities may exist within graduate programs in unobserved quality of PhDs’ faculty advisors. If higher-quality advisors are also more likely to get research grants, and if faculty with more research grant funding find their time and labs saturated with graduate research assistants leaving them no time to mentor other doctoral fellows, it is conceivable that doctoral fellows may be left to seek supervision from less research-active or less experienced faculty mentors. Further research is needed to disentangle the relative contributions of faculty supervisor characteristics such as research and mentoring experience, and students’ perceptions of embeddedness versus isolation.

Our findings present clear directions for future research, focused towards better qualitative understanding of how these mechanisms of graduate student support differ in terms of students’ mentored research experiences, their engagement with peers and more senior researchers on project teams, and their familiarity with different research environments and possible career paths. Even if traineeships and fellowships provide a very effective means of controlling total enrollment in biomedical sciences graduate programs, if these mechanisms are less efficient in producing scientific researchers, then we need to understand to what alternative occupations and careers those trainees and fellows are attracted, and why. This is not to say these alternative careers are inferior or undesirable; however, to the extent that federal agencies intend to use these mechanisms to stabilize and broaden diverse participation in scientific research jobs, it is important to understand the outcomes that result. With traineeship and fellowship mechanisms more often targeted towards women and underrepresented racial/ethnic minorities, our finding that these mechanisms are less likely to result in postdoctoral employment in the scientific workforce is particularly troubling. Future research will examine longer-term career outcomes, occupational activities, employer sectors, and job satisfaction associated with these alternative career paths, and their implications for broadening participation in the U.S. scientific workforce.

Highlights.

• NIH-funded research assistantships disproportionately support foreign students

• Graduate students’ funding mechanisms affect their early-career job choices

• Research assistants are more likely to take U.S. R&D jobs after graduation

Appendix

In the Special Regressor Method approach, the special regressor, V, is an exogenous regressor, independent of and additive to the error term. As shown in Lewbel et al. (2012), although the SRM estimation equation is similar to equations found in more common maximum likelihood and control function models, here V is separated from the other exogenous regressors:

$$D=I(X^{\prime }\beta +V+\epsilon \ge 0)$$ (5)

To improve estimation and interpretation of results, we further require that V be distributed with zero mean, that it have wider variance and distributional support than our outcome variable, and finally that it have significant positive correlation with the outcome variable. But, unlike an instrumental variable, which must pass an exclusion restriction—IVs must not have any explanatory or predictive power for the second stage regression—the special regressor V is appropriately and necessarily included in the second-stage model, as a predictor of our outcome of interest.

The SRM approach then incorporates both first-stage estimation with the excluded IVs as in usual 2SLS IV models, plus the special regressor added in the second stage, to improve consistency of the estimation results.

Following the requirements above, to implement the SRM approach, we transform the age variable, subtracting each student's age from the mean age for completing PhDs who share the student's citizenship status, to generate our candidate for V, demeaned negative age.

2SLS IV linear estimation demonstrates a statistically significant positive correlation between V and taking an R&D-focused job after graduation, and we find no statistical evidence suggesting age is endogenous. This proposed V also has much wider variance and distributional support than the R&D job outcome variable. As such, demeaned negative age appears to satisfy the requirements for use as a special regressor.