SUPPORTING “THE BEST AND BRIGHTEST” IN SCIENCE
AND ENGINEERING: NSF GRADUATE RESEARCH
FELLOWSHIPS*
Richard Freeman, Tanwin Chang, Hanley Chiang
NBER and Harvard
Presented at the SEWP Research Conference
Oct 19-20, 2005
Cambridge, Massachusetts
* Supported by a grant from the NSF in collaboration with the Council of Graduate Schools (CGS). While serving as an
AAAS/NSF Science and Technology Fellow, Dan Stanzione was instrumental in facilitating this study from the NSF side.
Jason Abaluck of Harvard assisted in preparation of the data. We thank Myles Boylan of the NSF for leading a thoughtful
review of a short version of part of this work given at the AEA meetings.
Stipends: Funding Modes for Graduate Education
•
•
•
•
•
Regular payments (distributed to approx. 2/3 of S&E grad students)
Three main types of stipends:
– Fellowships: competitive awards (13% of FT grad students)
– Teaching Assistantships: teaching required
– Research Assistantships: research required
Multiple Funders:
– Government: NSF, NIH, etc.
– Universities, Foundations, Individuals
Postdocs may also receive stipends, often via fellowships
Other forms of funding:
– Self-financed: savings/family
– Loans
– Traineeships
Stipends: Potential Supply-Side Policy Tool
• Time in graduate school/postdoc represents up to ¼ of a
scientist’s career. (6 years grad school + 4 years postdoc = 10 years out
of 40 year career)
– Salary over first ¼ will have a non-trivial effect on lifetime earnings –
larger than later $ because of discounting
– Signalling effect to student can also be important
• Government fellowships may be restricted to
citizens/permanent residents, primarily affecting US
individuals (Gov’t awards ~1/3 of S&E Fellowships)
• Quantity and price dimensions – policy can alter number of
awards and value of awards
NSF Graduate Research Fellowship Program (1952 - 2004)
30
25
2500
Annual Stipend Amount
20
2000
15
1500
10
Awardees
5
0
1950
1000
500
1960
1970
1980
1990
2000
The number and value of GRF awards vary over time.
0
# of People
Thousands of Constant Dollars (2003)
3000
Primary Data Source:
NSF’s Cumulative Index (CI), including Follow-Up Data
Summary for Key Data in Total Time
Period
the GRF Datafiles
Panel Ratings
Reference Score
GRE Quant
GRE Verbal
GPA
Ethnicity
Gender
Data
Available
1976 - 1993
1994 - 1998
1999
2000 - 2004
1954 - 1993
1994 - 1998
1952 - 2004
1952 - 2004
1955 - 1998
1976 - 2004
1981 - 1987
1988 - 2000
2000 - 2004
1952 - 2000
2001 - 2004
Records
Overall Award
Rate
1952 - 2004
271,391
13.50%
Mean
STD
Full Range
328
315
3.11
2.90
24.2
1.91
691
606
3.49
115
100 to 600
118
100 to 600
1.02
1.00 to 6.00
0.85
1.00 to 4.99
8.7
10 to 70
0.65
1 to 7
100
200 to 800
104
200 to 800
0.40
1 to 4
(1, 2, 3, 4, 5)
(1, 2, 3, 4, 5, 6, 7, 8, 9)
(A, B, C, D, E, F, H, I, J)
(t, f) or (y,n,b)
(m, f)
(m, f, b)
GRF Awardees as Proportion of S&E Bachelors (1952 - 2004)
8
Thousands of S&E Bachelors
Bachelors
7
400
6
5
300
4
200
Awardees
3
2
100
1
0
1950
1960
1970
1980
1990
2000
GRF Awardees per Thousand Bachelors
500
0
The number of awards per S&E baccalaureate has shifted downwards
Huge Changes in Distribution of GFRP Awardees by Field:
Decline of Physical Sciences
100%
Mathematics
Computer Science
80%
Physics
Geoscience
60%
Chemistry
Engineering
40%
Life Science
20%
Social Science
Psychology
19
52
19
55
19
58
19
61
19
64
19
67
19
70
19
73
19
76
19
79
19
82
19
85
19
88
19
91
19
94
19
97
20
00
20
03
0%
Because of Changes in Applicants by Field
100%
Mathematics
Computer Science
80%
Physics
Geoscience
60%
Chemistry
Engineering
40%
Life Science
20%
Social Science
Psychology
19
52
19
55
19
58
19
61
19
64
19
67
19
70
19
73
19
76
19
79
19
82
19
85
19
88
19
91
19
94
19
97
20
00
20
03
0%
What if NSF Did Not Regulate the Percentage of
Awardees in Each Discipline?
Our research implies that a shift to pure “measured scholastic”
achievement would not necessarily benefit the hard sciences at the
expense of the more qualitative sciences.
Using linear probability model, found only modest shifts among
most disciplines. The biggest increases in awardees were for
Psych and Social Science at the expense of Engineering! Why?
Mean Applicant Quality
Engineering
Psychology
Soc Science
Quant
728
653
667
Verbal
573
615
642
GPA
3.67
3.57
3.60
Refs
0.05
0.27
0.19
Analyzing the effect of changes to stipends on S&E
workforce: Quantity (N) and Quality (Q)
When the number or value of awards changes
– What happens to the number and quality of applicants,
awardees, and graduates?
• N = SN(#awards, $awards, X)
• Q = SQ(#awards, $awards, X)
10000
9000
8000
7000
6000
5000
4000
3000
2000
1000
0
1968
1.2
1
0.8
0.6
0.4
0.2
Stipend/Alternative
Appplicants
GRFP Applications and Relative Value of GRFP
Stipend, 1968-2004
0
1973
1978
1983
Applicants
1988
1993
1998
2003
Stipend/Alternative Wage
Source: NSF, Division of Graduate Education, Cumulative Index of the GRF Program and related datasets.
Bachelor’s degree data tabulated by National Science Foundation/Division of Science Resources Studies; data
from Department of Education.
10000
9000
8000
7000
6000
5000
4000
3000
2000
1000
0
1979
4
3.5
3
2.5
2
1.5
1
0.5
1984
1989
Applicants
1994
1999
Unemployment Rate (%)
Appplicants
GRFP Applications and College Graduate
Unemployment Rate, 1979-2004
0
2004
Unemp Rate
Source: NSF DGE, Cumulative Index of the GRF Program and related datasets. Bachelor’s degree data tabulated by
National Science Foundation/Division of Science Resources Studies; data from Department of Education.
Unemployment data from estimated from the annual Current Population Survey (CPS) Outgoing Rotation Group,
Bureau of Labor Statistics.
800
1.2
780
1
0.8
760
0.6
740
0.4
720
700
1965
0.2
1975
GRE Quan
1985
1995
Stipend/Outside Salary
GRE Score
GRE Quantitative Scores of Awardees and
Relative Stipend Value, 1968-2004
0
2005
Stipend/Outside Salary
Source: NSF DGE, Cumulative Index of the GRF Program and related datasets. Salary data estimated from the
Integrated Public Use Microdata Series (IPUMS) of the March Current Population Survey.
720
710
700
690
680
670
660
650
640
630
620
1965
GRE Verbal
1.2
1
0.8
0.6
0.4
0.2
1975
1985
Detrended Verbal
1995
Stipend/Outside Salary
GRE Scores
GRE Verbal Scores of Awardees and
Relative Stipend Value, 1968-2004
0
2005
Stipend/Outside Salary
Source: NSF DGE, Cumulative Index of the GRF Program and related datasets. Salary data estimated from the
Integrated Public Use Microdata Series (IPUMS) of the March Current Population Survey.
Linear Probability Model for the Impact of Scholastic
and Demographic Variables on GRF Awards.
1955-98
Offered Award
1976-98
0.082
(0.001)
0.034
(0.001)
0.060
(0.001)
0.089
(0.002)
0.074
(0.001)
0.035
(0.001)
0.066
(0.001)
0.095
(0.003)
0.033
(0.002)
0.087
(0.004)
1976-98
0.170
(0.001)
0.022
(0.001)
-0.010
(0.001)
0.019
(0.001)
-0.010
(0.003)
0.037
(0.002)
0.077
(0.003)
Yes
Yes
207498
0.198
Yes
Yes
107658
0.1793
Yes
Yes
107597
0.2731
Panel Rating
Reference Score
GRE Quant/100
GRE Verbal/100
GPA
Female
Minority
Field Effects
Year Effects
Observations
R-squared
Source: NSF, Division of Graduate Education, Cumulative Index of the GRF Program and related datasets.
Determinants of the Number of GRFP Applicants, 1969-2004
Dep Var: ln(applicants in academic field in
current year)
(1)
(2)
(3)
(4)
Log(Bachelor's Degs) by Field in Current Yr
Log(Stipend/Outside Salary) in Previous Yr
0.195
(0.057)
0.996
(0.084)
Unemp Rate for College Grads Age 21-25
0.304
(0.063)
0.916
(0.060)
0.049
(0.013)
Unemp Rate for All College Grads
0.298
(0.062)
0.852
(0.059)
0.516
(0.066)
0.772
(0.056)
0.104
(0.024)
0.094
(0.022)
0.349
(0.054)
Yes
Yes
234
0.9634
Log(Awards/Bachelor's Degs) by Field in Previous Yr
Field Effects
Field x Time Trend
Observations
R-squared
Yes
Yes
324
0.8931
Yes
Yes
234
0.955
Yes
Yes
234
0.9561
Source: NSF, Division of Graduate Education, Cumulative Index of the GRF Program and related
datasets, as described in text. Outside salary are earnings of college graduates aged 21-25. Outside
salaries and unemployment rates from Current Population Survey
Determinants of Awardee Achievement, 1969 - 2004
GRE Quant GRE Verbal
Log(Number of Awards) by
Field in Current Yr
Log(Stipend/Outside Salary) in
Previous Yr
Field Effects
Field x Time Trend
Observations
R-squared
GPA
-10.6
(1.80)
-21.8
(3.30)
-0.087
(0.01)
29
3.3
35.4
6.1
-0.007
0.017
Yes
Yes
324
0.8943
Yes
Yes
324
0.684
Yes
Yes
270
0.7354
Source: Tabulated from NSF, Division of Graduate Education, Cumulative Index of the GRF Program and related
datasets, as described in text. Outside salary are earnings of college graduates aged 21-25, tabulated from Current
Population Survey.
Quality of GRFP Applicants on the Margin of
Getting and Award, 2004
GRE Quant
GRE Verbal
800
750
700
650
600
550
1
2
3
4
5
6
7
8
9
10
Source and Notes: All persons to
the left of the line were given
awards. All persons to the right
of the line did not get awards.
The numbers relate to groups of
50 persons, ordered by the
estimated probability they would
win an NSF award. The 5th group
consists of 50 awardees with the
lowest probability of getting an
award, the 4th group consists of 50
awardees with the next lowest
probablity, and so on. The 6th
group consists of the 50 nonawardees with the highest
probablity of getting an award.
The probabilities are predicted
values from an OLS regression of
an award receipt dummy variable
on panel rating, female dummy,
underrepresented minority
dummy, and eight field dummies.
Units for Stipend Budget: Dollars per $10 Million GDP Dollars
Fellowships Fraction of Bachelors choosing to apply
to GRF vs. total GRF stipend budget/GDP.
50
45
40
Bottom Line
Message:
35
30
Students respond
to spending on
Government
Fellowships
25
20
15
10
5
0
1950
1960
1970
1980
1990
2000
Applicants per 1000 S&E Bachelors
Stipend Budget (Normalized by GDP)
Source: NSF DGE, Cumulative Index of the GRF Program and related datasets. Data on the Gross Domestic
Product (GDP) from the Bureau of Economic Analysis, an agency of the U.S. Department of Commerce.
Conclusions
• Verified that primary determinant of awardees are
academic achievement.
• Verified that women and minorities have higher chance of
winning fellowship, compared to white men of similar
attributes.
• for every 10% increase in the stipend value, the number of
applicants goes up by 8-10%.
• the supply of applicants contains enough qualified
candidates to allow for a sizable increase in the number of
awards without greatly reducing measured skills.
Further work to see if results of this study generalize to the
population of stipend grantees requires detailed consideration of the
entire market for all stipends … NSF are important but there are
many other awarding agencies and groups.
Fixed Rule Policy Scenario:
# of awards 0.41% of S&E Bachelors Degrees
$ of awards 115.8% of alternative wage
ACTUAL PROGRAM CHARACTERISTICS
Period
1968-1971
1972-1975
1976-1979
1980-1983
1984-1987
1988-1991
1992-1995
1996-1999
2000-2004
(Annual Averages)
Awards As
Stipend as Average Quality of
% of S&E
Awards
% of Alt.
Awardees
Bach
Wage
Degrees
Quant
Verbal
921
0.35%
55.70%
745.2
692.8
511
0.16%
66.00%
751
693.8
504
0.17%
55.00%
753.1
703.6
460
0.15%
56.30%
746.9
693.1
520
0.16%
82.00%
759.6
696.5
807
0.25%
92.20%
766.3
692.9
794
0.22%
84.50%
762
681.2
820
0.21%
72.10%
751.1
660.3
915
0.23%
93.40%
753.2
633.9
HYPOTHETICAL PROGRAM
CHARACTERISTICS
Awards
1082
1285
1237
1261
1342
1330
1493
1580
1640
Predicted Quality of
Awardees
Quant
764.7
751
765.2
757.1
759.6
767.7
764.4
757.9
753.3
Verbal
715.2
693.8
710.4
696.6
688
690.1
678.6
662.8
628.8
Notes: “Hypothetical” characteristics denote those predicted to prevail if awards as % of S&E bachelor’s degrees had been maintained
at 0.41% (its 1968 value) and stipend as % of alternative wage had been maintained at 115.8% (its 2004 value). Hypothetical changes
in GRE Quant and GRE Verbal are based on coefficients in columns 1 and 3 of Table 7. In the calculations, actual numbers of
bachelor’s degrees in 2002-2004 were imputed from 2000-2001 field-specific growth rates in earned bachelor’s degrees, and actual
alternative wage in 2004 was imputed from 1983-2003 growth rate in salary of 21-25 yr olds with bachelor’s degrees.