Planning early for careers in science, RH Tai, CQ Liu, AV Maltese, X Fan

Tags: baccalaureate degree, nonscience, baccalaureate degrees, science career, mathematics and science, AAAS, science degree, achievement scores, baccalaureates, Dependent variables, students, students' demographics, students', Regression analysis, cluster sampling, academic backgrounds, science achievement, Eighth-grade math, National Research Council, probability, National Science Board, science-related, career expectations, mathematics courses, National Center for Educational Statistics, National Education Longitudinal Study, light line, life science, physical science and engineering degrees, National Science, dark line
PHOTO CREDIT: PHOTOS.COM Life science Physical science/engineering Downloaded from on July 10, 2007
career choice Planning Early for Careers in Science
Robert H. Tai,* Christine Qi Liu, Adam V. Maltese, Xitao Fan
Young adolescents who expected to have a career in science were more likely to graduate from college with a science degree, emphasizing the importance of early encouragement.
Concern about U.S. leadership in science has captured the national spotlight once again (1). The physical sciences and
engineering are at particular risk, with declines
in the number of earned doctorates in these
fields among U.S. citizens and permanent resi-
dents in the past decade (2) (figs. S1 to S3).
Recommendations for
Enhanced online at content/full/312/5777/1143
improvement focus on education, particularly in improving the
number of teachers and
the quality of teacher
training for primary and secondary schools (1).
This is an attractive but expensive approach.
How important is it to encourage interest in
science early in children's lives? How early in
their lives do students decide to pursue a science-
related career? We used nationally representative
longitudinal data to investigate whether science-
related career expectations of early adolescent
students predicted the concentrations of their
baccalaureate degrees earned years later.
Specifically, we asked whether eighth-grade stu-
dents (approximately age 13) who reported that
they expected to enter a science-related career by
age 30 obtained baccalaureate degrees in sci-
ence-related fields at higher rates than students
who did not have this expectation. We analyzed
students in the United States for years 1988
through 2000 and controlled for differences in
academic achievement, academic characteris-
tics, and students' and parents' demographics.
Survey and Analysis We used the National Education Longitudinal Study of 1988 (NELS:88) for this study. Designed and conducted by the National Center for Educational Statistics (NCES), NELS:88 began in 1988 with a survey of 24,599 eighth graders. Researchers conducted additional surveys in 1990, 1992, 1994, and 2000. The overall sample size after five surveys was 12,144 participants. Our analysis focused on those students who responded to the question about their age 30 career expectation as eighth graders in 1988 and who
The authors are at the Curry School of Education, University of Virginia, 405 Emmet Street South, Charlottesville, VA 22904­4273, USA. *To whom correspondence should be addressed. E-mail: [email protected]
MULTINOMIAL LOGISTIC regression analysis
Independent variable
Coefficients of nested models Baseline 2 3 4 Final
took into account students' backgrounds and natural propensities. For example, students with stronger performance in science and mathematics may be more likely to major in the sciences. We
0.6 0.7 0.7 0.6 0.7
Life sci. (0.2) (0.2) (0.2) (0.2) (0.2)
expectation Phy. sci. 1.7 1.4 1.2 1.2 1.2 /engr. (0.2) (0.2) (0.2) (0.2) (0.2)
therefore included four covariate groups to account for (i) academic backgrounds (science and mathematics achievement scores); (ii) students' demographics (gender and ethnicity); (iii) students' academic characteristics (enrollment in
Covariate groups
advanced versus regular mathematics and science
Student demographics
+ +++
classes, attendance in these classes, and studentreported attitudes toward mathematics and
Achievement scores
+ + + science); and (iv) parents' background (highest
Academic characteristics Parent background
++ +
educational level and professional versus nonprofessional employment) (6). Our analysis focuses on the independent vari-
Regression analysis results. P < 0.001 for all data shown; + indicates inclusion of covariates in the model; standard errors are shown in parentheses; n = 3359. Dependent variables: nonscience = 0, life
able derived from the NELS:88 survey question: "What kind of work do you expect to be doing when you are 30 years old?" Students were then given a list of employment options and required
science = 1, and physical science/engineering = 2. to select only one. We categorized the responses
See supporting online material for more details. into two groups: science-related and nonscience
career expectations, creating the Career
also obtained baccalaureate degrees from 4-year Expectation independent variable (4).
colleges or universities by 2000. This reduced the
We applied multinomial logistic regression,
sample to 3,743 participants. The sample was fur- which handles categorical dependent variables
ther reduced to a final size of 3,359 participants, with more than two outcomes. Our analysis
because 384 participants were missing data in one included two outcome comparisons in earned
or more of the variables used in the analysis.
baccalaureate degrees: (i) earning degrees in life
These variables included scores from mathe- sciences versus nonscience areas and (ii) earning
matics and science achievement tests (designed degrees in physical sciences/engineering versus
by the Educational Testing Service) that were nonscience areas. We assessed the degree to
administered in the first three surveys of data col- which the independent variables could predict
lection, when students were mostly enrolled in these two comparisons. In the NELS:88 sam-
the 8th, 10th, and 12th grades (3, 4).
pling design, two analytical issues require
The baccalaureate degree concentrations-- special attention: (i) the effect of purposeful
which were coded into
three broad categories
of physical science/en-
gineering, life science,
and nonscience--result-
edinacategorical depend-
ent variable (tables S1
and S2 and supporting
online material text) (5).
The independent variables used in this analysis
0 15 25 35 45 55 65
0 15 25 35 45 55 65
came from data col-
Mathematics achievement score
Mathematics achievement score
lected when participants Estimated probability comparisons. Probability that students who, in eighth
were enrolled in the grade, expected (dark line) or did not expect (light line) a science career would
eighth grade.
achieve a life science degree (left) or a physical science/engineering degree
In our analysis, we (right). Blue arrow designates the average mathematics achievement score. SCIENCE VOL 312 26 MAY 2006 Published by AAAS
Earned baccalaureates (%) Downloaded from on July 10, 2007
oversampling of some ethnic and
minority groups and (ii) the effect
of multistage cluster sampling
on standard error estimation. We
followed the NCES guidelines
by using sampling weights for
statistical analyses (3). We ac-
counted for the complex sampling
design by using the STATA 9.0
statistical software package (3, 7).
Results and Discussion
Our analysis began with a base-
line model that included only
the Career Expectation inde-
pendent variable and continued
with successively more complex models systematically
<_30 35 40 45 50 55 60 65 <_30 35 40 45 50 55 60 65
accounting for each of the four
Math achievement score
Math achievement score
covariate groups (see table on
page 1143 and table S7).
Proportion of earned baccalaureates. Degrees in life science (light
As more independent vari- green), physical science/engineering (dark green), and nonscience
ables were included in the nested fields (gray). Students who in eighth grade expected a science degree
models, the coefficient remained are shown on the left (n = 337); those who did not expect a science
unchanged for the life science degree are shown on the right (n = 3022).
outcome. For the physical sci-
ence outcome, the coefficient at first attenuated
However, for physical science/engineering
from its initial value and then settled into a robust degrees, the result was quite different (see figure
value after model 3. This behavior is common in on page 1143, right panel). High mathematics
such analyses because variance accounted for by achievers were much more likely than low
initially entered variables is subsumed by succes- achievers to earn these degrees. For example, let
sive variables. We also checked for interactions us compare the estimated probabilities for two
between Career Expectations and the other inde- pairs of prototypical students with all other
pendent variables and did not find them to be sig- variables set to means. Suppose the first pair
nificant at the P = 0.05 level.
has average mathematics achievement scores
The odds ratios, calculated from the final (average math achievement score at eighth grade
model, were 1.9 for life sciences versus non- = 45, SD = 11). Here, the estimated probability
science and 3.4 for physical sciences/engineering of earning a physical science or engineering bac-
versus nonscience (table S7). This result suggests calaureate degree for the student who expected a
that, among the students who graduated with bac- science-related future career was 34%. In con-
calaureate degrees from 4-year colleges, those trast, for the student who expected a nonscience
who expected as eighth graders to have science- career, the estimated probability was 10%.
related careers at age 30 were 1.9 times more Suppose that for the second pair, we have high
likely to earn a life science baccalaureate degree mathematics achievers whose test scores were one
than those who did not expect a science-related standard deviation above average. Here, the esti-
career. Students with expectations for a science- mated probability of the student who expected
related career were 3.4 times more likely to earn a science-related future career was 51%,
physical science and engineering degrees than whereas the estimated probability of the student
students without similar expectations.
who expected a nonscience career was 19%. To
Next, we considered the estimated probabil- the extent that taking courses encourages expec-
ities of earning science baccalaureate degrees tations, this result supports the National Science
produced by the final model comparison (see Board's contention (8) that mathematics courses
figure on page 1143). For life sciences, esti- taken in grades 7 and 8 have an impact on the
mated probabilities nearly doubled for students physical sciences and engineering workforce.
who reported science-related career expecta-
There is an additional comparison across
tions compared with those who did not. For these pairs that should not go unnoted. An aver-
example, a prototypical student expecting a age mathematics achiever with a science-related
science-related career has an estimated proba- career expectation has a higher probability of
bility of obtaining a life science degree of 29% earning a baccalaureate degree in the physical
compared with 18% for a prototypical student sciences or engineering than a high mathematics
expecting a nonscience career, with all other achiever with a nonscience career expectation,
predictors set to the means. Eighth-grade math- 34% versus 19%. We make this comparison not
ematics achievement was not a significant pre- to minimize the importance of academic achieve-
dictor for life science degrees.
ment, but rather to highlight the importance of
career expectations for young adolescents. We analyzed (see figure at left) the propor- tion of students who earned the three types of baccalaureates degrees, according to eighthgrade expectations and math achievement scores. Most notable is the proportion of students who, in a sense, followed through on their eighth-grade science career choices--roughly half. In contrast, proportionally fewer students who reported nonscience career expectations switched into science--roughly a third. Much effort has been focused on raising test scores and promoting advanced courses at later ages; however, we should not overlook the likelihood that life experiences before eighth grade and in elementary school may have an important impact on future career plans. Although our current analysis does not provide proof of an uninterrupted causal chain of influence, our study does suggest that to attract students into the sciences and engineering, we should pay close attention to children's early exposure to science at the middle and even younger grades. Encouragement of interest and exposure to the sciences should not be ignored in favor of an emphasis on standardized test preparation (9). References and Notes 1. National Research Council, Rising Above the Gathering Storm: Energizing and Employing America for a Brighter Future (National Research Council, Washington, DC, 2005). 2. Data obtained through WebCASPAR ( 3. National Center for Educational Statistics, User's Manual: National Education Longitudinal Study of 1988 (NCES, Washington, DC, 2004). 4. D. A. Rock, J. M. Pollack, "Psychometric report for the NELS:88 base test battery" (Tech. Rep. NCES 91-468, National Center for Educational Statistics, Washington, DC, 1991). 5. N. Hativa, M. Marincovich, Disciplinary Differences in teaching and learning: Implications for Practice (JosseyBass, San Francisco, CA, 1995). 6. Occupational classifications are highly complex and, in this analysis, only limited data were available on parents' specific occupations. We paid special attention to parental occupation in this analysis, because conventional wisdom suggests that children whose parents have careers in science may be more likely to choose similar careers. However, the existing data related to parents' occupations were reported in categories that did not specify whether jobs were science related. As a result, we chose to use the broad categories of professional versus nonprofessional. 7. For a more detailed discussion of this technique, please see J. S. Long, J. Freese, Regression Models for Categorical Dependent Variables Using STATA® (STATA, College Station, TX, 2001). 8. National Science Board, An Emerging and Critical Problem of the Science and Engineering Labor Force: A Companion to the Science and Engineering Indicators 2004 (National Science Board, Washington, DC, 2004); available online ( nsb0407.pdf). 9. S. Dillon, "Schools cuts back subjects to push reading and math," The New York Times, 26 March 2006 ( 10. This study was funded in part through a grant from NSF, Directorate of Education and Human Resources, Division of Research, Evaluation, and Communication, Research on Learning and Education Program (REC 0440002, Project Crossover). We thank L. E. Suter (NSF), D. Herschbach (Harvard University), and D. R. Webb (Proctor and Gamble) for their comments. Supporting Online Material 10.1126/science.1128690
26 MAY 2006 VOL 312 SCIENCE Published by AAAS

RH Tai, CQ Liu, AV Maltese, X Fan

File: planning-early-for-careers-in-science.pdf
Title: Science Magazine
Author: RH Tai, CQ Liu, AV Maltese, X Fan
Author: American Association for the Advancement of Science
Published: Wed Oct 10 21:06:14 2007
Pages: 2
File size: 0.54 Mb

Noncoherent coded modulation, 12 pages, 1.15 Mb

, pages, 0 Mb

Class Schedule, 1 pages, 0.66 Mb

, pages, 0 Mb

Explorations in Australia, 292 pages, 1.09 Mb
Copyright © 2018