Vous êtes sur la page 1sur 11

Dale, Stacy Berg, Krueger, Alan B. November 2002.

Estimating the Payoff to Attending a


More Selective College: An Application of Selection on Observables and Unobservables.*
The Quarterly Journal of Economics vol. 117(4), 1491 1527.

One of the most important decisions of a young persons life is the decision of where to
attend college. Does one choose a more selective, prestigious University like Harvard or do they
choose to save money by a going to state school such as our University of Texas? One of the key
presumed differentiating factors that help students decide which type of University of decide is
cost and return on Investment. It is assumed that attending a more selective University leads to
higher salaries and success upon entrance into the jobs market, but this occurs at a high initial
cost. Students in this case must weigh the marginal utility of every extra dollar spent. However,
Berg Dale and Krueger in their work Estimating the Payoff to Attending a More Selective
College: An Application of Selection on Observables and Unobservables, contend students who
attended more selective colleges earned about the same as students of seemingly comparable
ability who attended less selective schools, (1491) seemingly shattering the myth that attending
a more selective school results in a higher lifetime payoff. They did note, however; that children
from low-income families earned more if they attended selective colleges (1491).
In building the argument that this paper presents, the authors first referred to other studies
that were undertaken that assessed similar ideas and concerns and explained possible
shortcomings in fully illuminating the relative benefits of attending a better school. Two new
approaches are proposed in order to account for the non-random selection of students (in terms
1

of acceptance) of more selective colleges. A stylized model is then created for college admissions
processes, an income form, and the subsequent earnings of these students. In order to analyze the
model, a large data set called College and Beyond was used, which contained information on
thousands of students who enrolled in a group of moderately to highly selective US colleges and
universities, together with survey information collected from the students at the time they took
the SAT (occurring approximately a year before college entry, on average), and salary
information collected in 1996, twenty years after the students had matriculated in college in the
fall of 1976. The authors contend utilized all this data in order to work around the concept that it
was hard to control for the lack of random assignation in college decisions, instead they decided
to work with the characteristics of colleges to which students applied and were admitted.
In order to build an argument that strengthened the position that their regression model
better ascertained the effect of colleges on students, one must consider similar studies and how
their selection criteria and analysis method impacted the resulting conclusions. Two studies that
were referenced regarding a correlation between higher average SAT scores or attendance at
higher tuition institutions suggested a return to higher earnings when they were observed in the
labor market. It was however noted, that the main issue with this conclusion is that students
who attend more elite colleges may have greater earnings capacity regardless of where they
attend school, (1492) suggesting an implicit selection bias by the Universities themselves. As
in, the attributes that lead to greater success in the labor market may be the same attributes that
selective universities search for during their admissions process.

Another issue that was raised is the consistent use of Ordinary Least Squares Regression
analysis in order to control for student differences, but the question arose that there could be
differences in students that are not observed, but have some sort of correlation with earnings and
the quality of college. However, college admissions decisions innately are based on
characteristics that the researchers are no privy to, which could overstate the payoff of attending
a selective school, relating to the issues mentioned in the preceding paragraph.
With respect to accurately assessing student ability and outcome, it is better to analyze data
from a different angle. Instead of identifying everything that could possibly matter for college
choice and earnings, it is better to assess with a key summary measure; the characteristics to
which students were applied and were admitted to. Using the data from the College and Beyond
survey to match 6,335 students who were accepted and rejected by a comparable set of college in
1976, it is more illuminating to compare labor market outcomes in 1995 amongst students who
had similar/same choices of Universities, but where some chose more selective schools than
other. The College and Beyond data set contains information on students who enrolled in a group
of moderately to highly selective U.S. college and universities, combined with survey
information collected from students at the time of taking the SAT, approximately a year before
college matriculation. In addition, utilizing the National Longitudinal Survey of the High School
Class of 1972 would be helpful to estimate the impact on a students subsequent earnings of the
average SAT scores of all the schools they applied to as well as the average SAT score of the
school they attended.

It is important to note that that College and Beyond included schools that were prestigious
private universities like Princeton and Yale; small private colleges like Williams and Oberlin;
and four public universities. There was a 350 (out of 1600) point range in the average SAT score
in the cohort of Universities. The 1976 Tuition rates ranged from a low of $540 to as high as
$3,850. Therefore, it can be concluded College and Beyonds survey is relatively expansive and
somewhat representative of the entire college population. College quality has been indirectly
assessed by the average SAT score of a school in conjunction with the net tuition.
As such, having all of the data pieces, it is easier to overcome selection bias by comparing
the outcomes of students who were accepted and rejected by similar cohorts of colleges, by
utilizing the observed data that is available to both economists and admissions counselors
(student SAT score) as an estimator to offset the impact of the unobserved variables which are
only available to the admissions counselors. This is due to the fact that the impact of these
unobserved variables is taken into account by the similar cohorts of acceptance and rejection of
Universities. Implicitly assumed is that students who are admitted and rejected to the same
cohort of schools have similar unobserved qualities, at least in the eyes of the college admissions
counselors.
College attendance is a process of three choices. A student must first decide where they
wish to apply for college, then colleges decide whether to accept the student into their institution,
and finally, the student decide which college the student will attend. College admissions consider
students on a variety of factors such as observable variables such as SAT scores and High School
grade point average and unobservable variables. College Admission, if placed in the context of

an econometric formula is described as Z ij = 1 X 1 i+ 2 X 2i +e ij > C j , then admit to college j,


otherwise reject applicant at college j. Z ij

is the quality of a student, as viewed by the

admission committee. e ij suggests the distinctive features that each schools admission board
has within their selection process that leads to weighting differing aspects of a students
application in a manner different than other schools.

X 1 i refers to the common observable

variables that are common amongst college applicants, such as SAT scores and high school GPA,
whereas

X 2 i is the valuation of the unobserved variables (in the point of the view of the

economist) that factors into a schools decision to accept or reject a candidate, with

1 and

2 referring to the weight school j places on each of these factors. If a students quality (
Z ij ), is greater than C j , the cut off value for admittance, then a student is admitted into a
school. This equation adequately explains the admission process for a college student from an
econometric perspective.
Where economists run into issues is translating this information into effects into career
income in the future. Linking wage effects to student attributes is done by assuming the

following,

j+ 2 X 1i + 3 X 2 i+ eij
j
. SAT is the average SAT score of the entire class that
lnW i= 0 + 1 SAT

enters along with student i. Similar to the formula utilized by admissions schools,

X 1 i refers

to the common observable variables that are common amongst college applicants, in this case the
individual student SAT scores, whereas

X 2 i is the valuation of the unobserved variables.

1 refers to the monetary payoff of attending more selective universities, negative or positive,

or even zero. Problems arrive because economists cannot approximate

3 X 2i

as these are

unobserved variables, it is almost impossible for adequate data to reflect those inputs, as such
economists are reduced to only assessing observable variables.
The result of these regressions is a biased and inconsistent output of monetary payoff of
attending more selective universities and observable variables, even if students selected which
school to attend on a purely random basis. The omitted variable bias from unobserved variables
is incredibly high as the unobserved variables are more likely to be admitted to and attend more
selective schools, and the positive effects of these unobserved variables would have a greater
impact on the other coefficients. Therefore, it is imperative find a way to measure impact of
school attendance, whilst discounting for unobserved variables such as effort, ambition, etc. and
the error. As such, the solution is to include an unrestricted set of dummy variables to indicate
groups of students who received the same admissions decisions from the same colleges. By
doing this, we limit the error term and even assuming some effect of university matriculation
decision originates from unobserved variables, some of the effect is absorbed. As such, assuming
that students apply to a large range of universities, the unobserved variables are controlled by the
accept and reject dummies, as the matched applicant pools would be unbiased as there would be

a range in values for unobserved variables within the pools matching student decision outcomes.
This model is known as the matched applicant model.
Another alternative solution to minimizing the impact of unobserved variables is to assume
that students themselves are aware of their academic potential, and display their ability by the
types of schools that students apply to. An example of this concept would be noted filmmaker
and film executive Stephen Spielberg who applied to film schools at prestigious institutions such
as the University of Southern California and the University of California Los Angeles, but
attended California State Long Beach. It should be noted that California State Long Beach does
not a film studies program that is near in stature as the aforementioned campuses of USC and
UCLA. This model, known as the self-revelation model focuses on the number of schools that an
applicant applied to, and the average SAT score of these schools as a measure to limit the impact
of unobserved variables. In this approach, dummy variables are used to indicate the number of
schools a student applied to, and the average SAT score of all of the schools the student sent
applications to.
It is important to note the difference between the matched applicant model and the selfrevelation model and how results can possibly imply different outcomes in each. The matched
applicant model would compare students who were accepted to the same schools and/or rejected
to the same set of schools and analyze the difference in earnings outcome based on the school
selection. It is entirely possible that the student selected a less selective school based on greater
earnings potential, an example being attending McCombs Business School at the University of
Texas as opposed to studying a less lucrative major at an Ivy League. In these situations,
estimates from the matched applicant model would be biased downward, depending on the

amount of students who chose the type of school they attended based on the aforementioned
situation. In contrast, in the self-revelation model the comparison is between students who
applied to the same subset of Universities, not taking into account acceptance decisions. It is
assumed that students who attended less selective institutions is assumed to be less promising, in
the view of the college admissions board than those who attended the more selective universities.
If it is assumed that students with higher unobserved variables are more likely to be accepted and
attend more selective schools, the self-revelation model would be biased upwards in favor of
showing income rate of return in relation to school selectivity.
Ultimately, school selectivity, measured by average SAT score of the students at a school
does not actually have a tangible benefit in terms of a higher income over time. Students who
were selected to go to more selective schools, but opted for less selective schools were not being
paid less in their career than students who went to the more selective schools. One possible
explanation for this result is that students who attended more selective schools graduated with
lower GPAs and class ranks than those who attended less selective schools in their cohort. There
is a direct negative observable variable outcome result in attending a more selective school (a
school with 100 point higher SAT score), approximately a 5-8% percentile drop. However, this is
offset by the fact that a 7% percentile increase only results in a 3% increase in earnings, resulting
in a marginal difference in income due to grade variation due school selectivity. Another
possibility is that students who attend more selective universities choose degrees that result in
less monetary income, but satisfaction through other means. Nevertheless, when including course
choice as a variable in the regression, school selectivity continues to have a marginal impact on
long-term salary.

Additionally, a better predictor of salary outcome was the average SAT of the schools that
the student applied to, but did not attend. As a result, it seems that the self-revelation model is
more effective in highlighting unobserved variables and their impacts on future income.
Applying to schools with higher average GPAs and selectivity status possibly highlight a
students motivation, ambition, etc. (the previously discussed unobserved variables). Therefore,
it can be assumed aforementioned unobserved variables are a stronger driving force in an
individuals salary capacity rather than the average SAT score of his or her college counterparts,
denoting selectivity of Universities. This outcome strengthens the concept of the self-revelation
model, as it assumed that students will attend the most selective school that they are accepted
into, suggesting that students are rational consumers that do not have non-random decision
making processes.
In this case, the matched-applicant model suggests minimal impact in attending more
selective schools, bringing into question the effectiveness of the matched-applicant model. The
main assumption of the matched applicant model is that students in cohorts would ultimately
choose schools randomly due individual selection factors. However, when testing observable
characteristics to predict whether students attended the most selective college that they were
admitted to, it is seen that students with higher SAT scores were more likely to attend the most
selective college that each gained admittance, leading to a conclusion that school selection is not
random as observed variables increase. If it is assumed that this would also be the case in
unobserved variables, then returns to school quality is biased upwards, further diminishing the
marginal return of attending a more selective school within your cohort. This gives further
credence to the self-revelation model, as it assumed that students will automatically choose to
attend the most selective school in which they are accepted.
9

Although school selectivity is not a major component in future salary, all other factors
being held equal, students who attended colleges with higher average tuition costs or spending
per student seemed to earn higher salaries upon entrance into the labor market. Which may seem
counterintuitive as more expensive schools are generally more selective in admission, this higher
salary seems to be expected due to a higher investment into the student by the schools
themselves through student costs. When controlling for selectivity, the rate of return in income
drops in respect to cost of attendance, but still remains large. This may be due to the fact that
schools with high tuition provide students with better or additional resources that are not
available to colleges with lower tuition. Although the regression attempted to control for family
income, it must be noted that student who attend higher cost schools generally have more wealth.
As a result, the impact of the cost of tuition may be upwardly biased by the impact of family
background in this instance.
It is interesting to note that the starkest returns to school in regards of income, whether it is
by the matching-applicant model, self-revelation model, or by Tuition are for students from
disadvantaged backgrounds. This serves to affirm the success of admission policies such as
Affirmative Action and financial aid packages that have a stated goal of encouraging qualified
lower income students to attend Universities. Results show that a continued practice would raise
national income, having many positive economic impacts; however, findings show that college
enrollment has only marginally increased for children from low-income families. This is
concerning, as it suggests that types of students that are best served by attending elite
universities, the high-achieving/low-income student are not adequately benefitting from their
opportunities.

10

11

Vous aimerez peut-être aussi