Content uploaded by James Alm

Author content

All content in this area was uploaded by James Alm on Oct 12, 2017

Content may be subject to copyright.

Economics of Education Review 28 (2009) 728–738

Contents lists available at ScienceDirect

Economics of Education Review

journal homepage: www.elsevier.com/locate/econedurev

Distance and intrastate college student migration

James Alma,∗, John V. Wintersb

aDepartment of Economics, Andrew Young School of Policy Studies, Georgia State University, P. O. Box 3992, Atlanta, GA 30302-3992, USA

bDepartment of Economics, Auburn University at Montgomery, P.O. Box 244023, Montgomery, AL 36124-4023, USA

article info

Article history:

Received 29 November 2007

Accepted 3 June 2009

JEL classiﬁcation:

I200

I230

Keywords:

College education

Mobility

Gravity model

Intrastate migration

abstract

Most studies of student migration focus on interstate migration of college students, largely

because the aggregate data typically used are limited in geographic speciﬁcity to states.

However, interstate migration is only a small part of the total student migration. Public

institutions generally get most of their students from within their state; for example, 88

percent of ﬁrst-time freshmen who enrolled in University System of Georgia institutions

in 2002 graduated from Georgia schools. Such intrastate migration is seldom considered.

This paper examines intrastate college student migration, using data for Georgia. Aside from

such traditional measures of beneﬁts and costs like tuition, ﬁnancial aid, and school quality,

a crucial explanatory variable in our analysis is the distance from a student’s home to the

different Georgia state institutions. Our empirical results indicate that student intrastate

migration is strongly discouraged by greater distance, but with effects that differ across

types of higher education institutions.

© 2009 Elsevier Ltd. All rights reserved.

1. Introduction

Pursuing higher education involves a signiﬁcant migra-

tion decision by students, and most studies of student

migration focus on interstate migration of college stu-

dents (e.g., out-migration, in-migration, net migration),

largely because the aggregate data that are typically used

in these studies are limited in geographic speciﬁcity to

states. However, interstate migration is only a small part

of the total student migration that actually occurs. Pub-

lic institutions generally get most of their students from

within their state; for example, 88 percent of ﬁrst-time

freshmen who enrolled in University System of Georgia

(USG) institutions in 2002 graduated from Georgia schools

(Information Digest, 2002–2003), and many other states

exhibit a similar pattern. Such intrastate migration is sel-

dom considered. This paper examines intrastate student

migration using data for the USG, a statewide system of

∗Corresponding author. Tel.: +1 404 413 0093; fax: +1 404 413 4985.

E-mail addresses: jalm@gsu.edu (J. Alm),

jwinter3@aum.edu (J.V. Winters).

universities and colleges that includes all public institu-

tions of higher education in Georgia except for technical

colleges. Aside from such traditional measures of beneﬁts

and costs like tuition, ﬁnancial aid, and school quality, a

crucial explanatory variable in our analysis is the distance

from a student’s home to the different Georgia state institu-

tions. Our empirical results indicate that student intrastate

migration is strongly discouraged by greater distance, but

with effects that differ across types of higher education

institutions.

Indeed, evidence on USG enrollment rates by school dis-

trict indicates that distance may in fact affect enrollments

in USG institutions. Fig. 1 provides a map of Georgia public

school districts and the 33 institutions of the USG. School

districts are shaded according to their USG enrollment rate,

deﬁned as the number of ﬁrst-time freshmen enrolling in

any institution of the USG divided by the number of high

school graduates; location of a USG institution is indicated

by an open circle. The shading in Fig. 1 suggests that school

districts that contain or are near a USG institution gen-

erally have a higher USG enrollment rate. However, this

evidence is only suggestive, and more systematic analysis

is necessary.

0272-7757/$ – see front matter © 2009 Elsevier Ltd. All rights reserved.

doi:10.1016/j.econedurev.2009.06.008

J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738 729

Fig. 1. Map of member institutions of the University System of Georgia

and Georgia public school districts shaded by participation rates. Notes:

School districts are shaded according to their USG enrollment rate, deﬁned

as the number of ﬁrst-time freshmen enrolling in any institution of the

USG divided by the number of high school graduates. USG institutions

are indicated by open circles. The black random shapes around some

open circles represent the boundaries for independent city school dis-

tricts. (Most school districts are coterminous with county boundaries,

but a few counties have independent city school districts within them.)

Source: constructed by the authors based on data from the USG and the

CCD.

In-state students face two, sequential migration deci-

sions, both of which seem likely to be affected by distance.

First, a student must decide whether to pursue any type

of higher education. Second, conditional upon deciding to

pursue postsecondary education, the student must decide

which institution to attend. If there is no postsecondary

institution near a student’s home, then pursuing higher

education necessarily involves moving to a new place or

commuting long distances; even when there are higher

education options near the student’s home, the student still

has to decide if the best choice is a nearby institution or one

further away. If the student chooses an institution in his or

her home state, the institution may still be as far away as

two or three hundred miles in an average size state.

In this paper we examine these intrastate migration

decisions for ﬁrst-time freshmen who graduate from Geor-

gia public high schools and attend a USG institution.

Georgia public high school graduates account for 81 per-

cent of ﬁrst-time freshmen enrolled in the USG in 2002

(Information Digest, 2002–2003).1We are interested in

two related questions. First, what factors affect a student’s

probability of enrolling in a USG institution? Second, con-

1Graduates of Georgia private high schools account for an additional

seven percent.

ditional upon enrollment in a USG institution, what factors

affect a student’s choice among USG institutions? In both

cases, an important variable is the distance to the nearest

USG institution, and, along the lines of gravity models of

international trade (e.g., Bergstrand, 1985), we hypothesize

that the likelihood of attendance decreases as the distance

to the nearest USG institution increases. It seems likely that

individuals are more likely to enroll at all when they live

closer to an institution, and also that they are more likely to

enroll in institutions that are located closer to their home.2

We estimate the elasticity of higher education atten-

dance with respect to distance from home to college, using

a basic gravity model approach. We ﬁnd that the overall

probability of attending any USG institution is decreas-

ing with the distance to the nearest USG college, with an

enrollment-distance elasticity of −0.12. We also estimate

the distance elasticity for enrollment in individual insti-

tutions separately for each institution, and we ﬁnd that

demand for more prestigious institutions is less elastic with

respect to distance than less prestigious institutions.

In the next section we discuss previous research on stu-

dent migration. We then present in Section 3our approach

and our data. In Section 4we present our estimation results,

and we conclude in Section 5.

2. Previous work on student migration3

Much of the literature on student migration focuses on

the role of tuition and ﬁnancial aid policies in the stu-

dent migration process, and most of this work focuses

upon interstate student migration. For example, using data

on U.S. states Tuckman (1970) ﬁnds that higher aver-

age tuition and fees in a state increase out-migration

from the state, while levels of ﬁnancial aid do not affect

out-migration; Mixon (1992a, 1992b) also estimates that

higher in-state tuition repels students from their home

state. In general, most other studies conclude that stu-

dent enrollment responds in predictable ways to changes

in tuition and ﬁnancial aid policies.4However, there are

exceptions. Baryla and Dotterweich (2001) ﬁnd that higher

tuition increases the percentage of non-resident students

in the U.S. South and West.5Other variables that have been

examined include regional amenities such as climate and

participation in NCAA athletics (Mixon, 1992a; Mixon &

Hsing, 1994a, 1994b). Mixon and Hsing (1994a, 1994b) also

2Distance might also affect student retention and degree completion. If

the “costs” of attendance continue to be higher for students from further

away, we might expect distance to have a negative effect on both retention

and completion. A few previous studies support this notion. Card (1995)

ﬁnds that proximity to a college while in high school increases years of

completed education, and other research suggests that increases in col-

lege costs in general have a negative effect on college completion (Hoxby,

2004). However, an analysis of the effects of distance on student retention

and degree completion is beyond the scope of this paper.

3See Greenwood (1975) and Ghatak, Levine, and Price (1996) for gen-

eral surveys of the migration literature.

4For reviews of the literature, see Leslie and Brinkman (1987),Heller

(1997), and Hoxby (2004).

5One explanation offered by McHugh and Morgan (1984) and Baryla

and Dotterweich (2001) is that tuition levels are correlated with educa-

tional quality and that the quality effect may offset the price effect.

730 J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738

ﬁnd that the percentage of non-resident student enroll-

ment increases for institutions classiﬁed as historically

Black.

These interstate studies do not incorporate distance into

the framework. However, it seems obvious that distance

might affect student migration, and several other stud-

ies have incorporated distance into the student interstate

migration framework. The earliest work is by Gossman,

Nobbe, Patricelli, Schmid, and Steahr (1968), who ﬁnd that

distance has a strong deterrent effect on student inter-

state migration, with a distance elasticity of enrollment

between −1.5 and −2.0. Using state-to-state migration

ﬂows, McHugh and Morgan (1984) also ﬁnd a distance

deterrence effect with a distance elasticity of −1.3. Kyung

(1996) examines student in-migration to New York state,

and estimates a much smaller (but still negative) elasticity

of −0.5.6

These studies are concerned with U.S. interstate student

migration. However, intrastate migration is important as

well. A number of studies examine the importance of dis-

tance for intrastate migration, including McConnell (1965),

Kariel (1968),Ullis and Knowles (1975),Leppel (1993),

Ordovensky (1995),Desjardins, Dundar, and Hendel

(1999), and Ali (2003). All of these works (except Desjardins

et al. (1999)) ﬁnd a signiﬁcant deterrent effect. However,

only Ullis and Knowles (1975) and Ali (2003) compute

intrastate distance elasticities. Ullis and Knowles (1975)

estimate separate equations for four state colleges in Wash-

ington state, and estimate distance elasticities ranging from

−0.4 to −1.7. Ali (2003) examines interstate and intrastate

student migration to West Virginia University, and esti-

mates an interstate elasticity of −2.2 and an intrastate

elasticity of −0.7, suggesting that out-of-state students are

considerably more deterred by distance than are in-state

students.

Comparing these studies, we see that the distance elas-

ticity of enrollment demand is almost always signiﬁcantly

negative, but ranges signiﬁcantly in value. Reasons for the

variance are unclear, but it is certainly likely that enroll-

ment demand in different geographic areas and different

institutions is affected by distance in a non-uniform man-

ner, and these aspects have not been fully examined. Lowe

and Viterito (1989) assess the market penetration of private

colleges and universities in the U.S., and argue that individ-

ual schools have either a local, regional, or national base of

appeal. They also ﬁnd that market penetration is highly cor-

related with mean SAT scores indicating that more selective

schools tend to have larger bases of appeal. Of course, other

factors could explain the variability as well.

Despite the many insights from this work, there remain

important gaps. Most studies focus on interstate student

migration; of those that examine intrastate migration, the

focus is on ﬂows to only a small number of state institu-

tions. Further, the role of distance is not always considered

in the intrastate studies, and, when it is incorporated,

distance elasticities are not typically generated. Our con-

tributions are to examine the role of distance for all state

6In a more recent study, Sà et al. (2004) examine interregional student

migration in the Netherlands, and report a distance elasticity of −1.3.

institutions in a single state (Georgia), to estimate the

resulting distance elasticities, and to estimate institution-

speciﬁc intrastate distance elasticities.

3. Methods

3.1. A gravity model of student migration

We use a gravity model of student migration to estimate

the distance elasticity of enrollment demand for the Uni-

versity System of Georgia. Migration ﬂows Mij from school

district i to institution j can be written as:

Mij =AijP˛1

iP˛2

jdˇ

ij ,(1)

where Piis the student population of school district i (the

origin area), Pjis the student population of institution j (the

destination area), dij is the distance from i to j, Aij =˘k(zk

kij)

is a multiplicative shift term that allows for inclusion of

Kvariables measuring school district “push” factors and

institution “pull” factors zkij that inﬂuence migration, and

(k,˛1,˛2,ˇ) are parameters. The gravity model speciﬁes

migration ﬂows as a function of the origin area popula-

tion, the destination area population, the distance between

the origin and destination, and a number of push and pull

variables. Larger school districts are expected to send more

students to higher education institutions, and larger insti-

tutions are expected to receive more students, so both ˛1

and ˛2are expected to be positive. The distance between

a school district and a particular institution is expected to

deter the ﬂow of students between the two, so ˇis expected

to be negative. Distance is expected to deter student ﬂows

because greater distance creates both greater monetary

costs and greater “psychic” costs to migration; greater dis-

tance may also deter student ﬂows because it reduces the

amount of information (i.e. familiarity) a student receives

about a school. Examples of push factors include economic

and demographic characteristics of the school district. Pull

factors include characteristics of the institution.

Institutions are not evenly dispersed over space, so the

model also needs to account for the fact that some institu-

tions are geographically closer to other institutions while

some are quite distant from the nearest institution. One

way to do so that has been suggested by Fotheringham,

Nakaya, Yano, Openshaw, and Ishikawa (2001) and by Sà,

Florax, and Rietveld (2004) is to include a centrality index

that measures for each institution j the spatial competi-

tion between institutions. Deﬁning the centrality index to

be the ﬁrst of the Kvariables in Aij and suppressing the i

subscript (because the index is measured at the institution

level), the index for institution j is computed as:

z1j =˙m(Pm/dmj),(2)

where Pmis the student population of institution m and dmj

is the distance between institutions m and j. The centrality

index is a population-weighted average of the inverse dis-

tance between institution j and the other M−1 institutions

in the system. If there are many close institutions with large

student populations, then the centrality index for institu-

tion j would have a large value, indicating that institution j

is subject to greater spatial competition.

J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738 731

We linearize the gravity model by taking the natural

logarithm of both sides of Eq. (1), which yields our basic

estimation equation:

ln Mij =˙kk(ln zkij)+˛1ln(Pi)+˛2ln(Pj)+ˇln(dij ).(3)

The coefﬁcients can now be interpreted as elasticities of

enrollment demand. The transformed gravity model is usu-

ally estimated via Ordinary Least Squares (OLS) methods.

However, it is often the case that the ﬂows are left-censored

at zero, causing OLS to be inconsistent. Therefore in esti-

mation where ﬂows of zero exist, we add one to all ﬂows

before taking the log and then estimate equation (3) using

the standard Tobit model. Amemiya (1973) shows that the

Tobit estimator is consistent and asymptotically normal.

3.2. Data

Our data come from several sources. We obtain data on

ﬁrst-time freshman student ﬂows from 175 public school

districts in Georgia to the 33 member institutions of the

University System of Georgia for 2002 using the USG’s High

School Feedback Reports.7We merge these data with infor-

mation on school district characteristics from the National

Center for Educational Statistics (NCES) Common Core of

Data (CCD), School District Demographics System (SDDS),

and USG institutional characteristics from the NCES Inte-

grated Postsecondary Education Data System (IPEDS). The

year 2002 was chosen for the analysis largely because of

data limitations. The High School Feedback Reports are only

available beginning in 2000, and the additional data used in

the analysis are not available for more recent years. Because

of our focus on the year 2002, we are unable to determine

if the effects of distance are rising or falling over time. Our

dataset excludes Georgia private high school graduates,

non-resident students attending USG institutions, and stu-

dents not attending a USG institution. While the migration

patterns of these student groups are certainly interesting,

the focus of our study is on the intrastate college migration

of Georgia public high school graduates to schools within

the USG.

One of our key variables is distance. Our variable Dis-

tance (as well as its variants) is calculated using Geographic

Information Systems (GIS) via ArcMap, and is measured in

miles (before taking logs). We use Census boundary maps to

calculate geographic centroids for each school district. USG

institutions are then geocoded, and straight-line distances

from each institution to each of the 175 school district cen-

troids and between institutions are computed using the

great circle distance formula.8,9

7According to the National Center for Education Statistics (NCES), there

are 183 uniﬁed school districts in Georgia. However, only 175 of them go

through the 12th grade and produce high school graduates.

8One could also compute the “over-the-road” distance between school

districts and colleges, but, because the network of highways in Georgia is

fairly dense, these two measures are very highly correlated.

9In computing distance for institutions with multiple campuses, we

consider only the geographic location of the main campus. Unfortunately,

enrollment ﬂows from school districts to institutions are not broken down

by campus, so incorporating distance to satellite campuses would not be

helpful unless enrollment ﬂows to an institution can be allocated among

its multiple campuses. Given that most USG institutions have a single

As for other variables, the school district population Pi

is calculated as the total number of persons receiving high

school diplomas from a given school district in 2002 (HS

Diplomas). Similarly, the destination area population Pjis

deﬁned as the total number of freshmen who graduated

from a Georgia public high school in 2002 and chose to

attend a given institution minus the ﬂow from school dis-

trict i (GA Public HS Freshmen (not i)). The ﬂow from i is

subtracted to avoid having an explanatory variable that is

a linear function of the dependent variable. Distance and

population are also used to compute the centrality index

(Centrality).

Other variables include institution-speciﬁc pull factors

and origin-speciﬁc push factors, as well as some basic con-

trol variables. Pull variables include mean SAT scores (SAT),

the percentage of freshmen receiving the Georgia HOPE

scholarship (Percent HOPE), per pupil USG expenditures

(Expenditures USG), the estimated total cost of attendance

for resident students living off-campus (Cost), and the per-

centage of students who are Black in a given USG institution

(Percent Black USG). The ﬁrst three variables measure the

quality of the institution, and are expected to attract stu-

dents. The cost of attending a particular school is expected

to have a deterrent effect on student ﬂows. We use the

total cost of attendance instead of tuition for several rea-

sons. Tuition is set by the USG for each sector of the system

with virtually the only variation occurring across different

sectors and not within them. Also, 78 percent of ﬁrst-time

freshmen in the USG from Georgia high schools received

the Georgia HOPE scholarship in 2002, rendering tuition

even less important.10 The cost of attendance is also more

appropriate than tuition because Cost encompasses dif-

ferences in the cost of living at the different institutions.

Students on average may be either drawn to or repelled

by historically Black institutions, so the expected sign on

Percent Black USG is unclear. When appropriate, we also

include a dummy variable equal to one if an institution

is located in the Atlanta metropolitan area (Atlanta), and

dummy variables for the different academic sectors of USG

institutions (Research University,Regional University,State

University,State College).11

The push variables are the percentages of high school

graduates in a school district that are Black, Hispanic, and

Asian (Percent Diplomas Black,Percent Diplomas Hispanic,

Percent Diplomas Asian), as well as the median household

income in a school district (Median Income), the percent-

age of persons 25 and older in a school district with a

campus and that those with multiple campuses tend to have most of their

enrollment at their main campus, we do not believe that the existence of

satellite campuses is a major problem.

10 Note that some of these students lost the HOPE Scholarship in later

semesters due to failure to maintain the required grade point average.

11 The USG consists of 35 higher education institutions, with 33 of them

enrolling undergraduates in 2002. These institutions are classiﬁed into

ﬁve sectors, each with different missions. In 2002 the ﬁve sectors included

three research universities, two regional universities, thirteen state uni-

versities, two state colleges, and thirteen two-year colleges. Eight of these

are located in the Atlanta metropolitan area: Atlanta Metropolitan College,

Clayton State University, Georgia Institute of Technology, Georgia Perime-

ter College, Georgia State University, Kennesaw State University, Southern

Polytechnic State University, and the University of West Georgia.

732 J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738

Table 1

Summary statistics.

Variable Mean Standard deviation Minimum Maximum

ln ﬂow 0.497 1.027 0 6.913

ln (ﬂow + 1) 0.636 1.048 0 6.914

ln distance 4.667 0.708 −0.906 5.841

ln distance to closest USG institution 2.658 0.960 −0.9066 3.715

ln distance to closest college 3.137 0.779 0.664 4.418

ln distance to closest university 3.096 0.992 −0.906 4.367

ln HS diplomas 5.198 1.063 2.197 8.719

ln GA public HS freshmen 6.444 0.763 4.977 7.995

ln GA public HS freshmen (not i) 6.438 0.764 4.249 7.995

Percent diplomas Black 0.320 0.254 0 1

Percent diplomas Hispanic 0.016 0.030 0 0.272

Percent diplomas Asian 0.011 0.016 0 0.105

ln expenditures school district 9.003 0.171 8.674 9.994

Percent BA degree 0.147 0.0820 0.055 0.559

ln median income 10.425 0.240 9.902 11.175

ln expenditures USG 9.498 0.391 8.943 10.703

Percent HOPE 69.164 16.877 40.100 98.900

ln SAT 6.869 0.105 6.697 7.189

ln cost 9.328 0.213 8.721 9.733

ln centrality 5.886 0.619 5.110 7.505

Percent Black USG 30.539 25.252 2.100 94.400

bachelor’s degree or higher (Percent BA Degree), and expen-

ditures per pupil for each school district (Expenditures

School District).12 Demographic controls for school districts

are included because some ethnic minorities might have

lower rates of college enrollment than white students. Sim-

ilarly, students in wealthier districts and more educated

districts might be more likely to enroll in the USG. In some

estimations, we include four school district metropolitan

status indicator variables. Metro1 is a dummy variable for

the city of Atlanta, Metro2 represents central city school

districts in all other metropolitan areas in Georgia, Metro3

represents school districts in the Atlanta metropolitan

area excluding the city itself, and Metro4 represents non-

central city school districts in metropolitan areas other

than Atlanta. Non-metropolitan districts are the excluded

group when metropolitan dummies are employed.13 Sum-

mary statistics for all continuous variables are in Table 1.14

4. Empirical ﬁndings

4.1. Attending any USG institution?

Table 2 presents estimation results on how the like-

lihood of attending any USG institution varies with the

distance to the closest college or university, the number

of high school graduates in a district, and the various push

12 While we include the percent Black, Hispanic, and Asian of high school

graduates in a school district, we do not include percent Hispanic and

percent Asian for institutions in the USG. The rationale is that Hispanics

and Asians are disproportionately concentrated in certain school districts

but are not particularly concentrated within USG institutions. On the other

hand, Blacks tend to be highly concentrated in both school districts and

USG institutions.

13 Of the 175 school districts, one is classiﬁed as Metro1,14areMetro2,

25 are Metro3, and 11 are Metro4. The remaining 124 school districts are

not in metropolitan areas.

14 Note that ln Distance is negative between three school districts and

institutions because the computed distance is less than one mile. Recoding

the three negative values to zero (e.g., one mile) does not affect our results.

factors. This is not a standard gravity model because there is

only one destination, the USG, so we do not include institu-

tional pull factors.15 Table 2 presents various speciﬁcations.

The results in column (1) indicate that distance to the

closest USG institution has a signiﬁcantly negative effect

on enrollment in the USG. The distance elasticity of −0.075

suggests that a one percent decrease in the distance to the

nearest USG institution increases enrollment in the USG by

0.075 percent. Because this is the ﬁrst study to estimate the

elasticity of system enrollment with respect to the distance

to the nearest institution of a state university system, we

cannot compare this estimate to other distance elasticity

estimates in the literature. Even so, the conclusion is clear:

distance has an important deterrent effect on enrolling in

the USG. If public policies can decrease the distance Geor-

gia public high school graduates must travel to attend the

nearest USG institution, then more graduates will pursue

higher education; that is, the USG can enroll more students

by creating new institutions and new campuses for exist-

ing institutions that are closer to areas in which high school

graduates live.

The other variables in column (1) are school district

characteristics used as controls to account for heterogene-

ity in origin areas. The most important and signiﬁcant

variable is the number of persons receiving diplomas from

the school district in 2002. Its estimated coefﬁcient sug-

gests that a one percentage point increase in the number

of diploma recipients from a school district increases the

number of ﬁrst year USG enrollees from the district by

roughly one percent, or an elasticity of about unity.

The race/ethnicity variables are included because of a

concern that districts with a high percentage of minority

students would send fewer graduates to the USG. However,

the results in column (1) do not suggest that this is the case.

15 However, we did experiment with including characteristics of the

closest institution, closest university, and closest college. These variables

were generally insigniﬁcant.

J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738 733

Table 2

Attending any USG institution?

Estimation method Speciﬁcation

Variable (1) (2) (3) (4) (5) (6)

OLS OLS OLS OLS Tobit OLS

ln USG enrollment ln USG enrollment ln USG enrollment ln USG enrollment ln college enrollment ln university enrollment

ln distance closest −0.075*** [0.023]

ln distance closest college −0.122*** [0.023] −0.116*** [0.021] −0.647*** (0.061) 0.241*** [0.037]

ln distance closest university −0.017 [0.022] 0.018 [0.022] 0.131** (0.058) −0.088*** [0.030]

ln HS diplomas 1.033*** [0.035] 1.033*** [0.035] 1.034*** [0.036] 1.048*** [0.036] 0.930*** (0.071) 1.032*** [0.043]

Percent diplomas Black 0.101 [0.118] 0.09 [0.116] 0.097 [0.118] 0.101 [0.119] −0.156 (0.228) 0.438*** [0.155]

Percent diplomas Hispanic 1.964*** [0.681] 1.382** [0.666] 1.346** [0.648] 1.999*** [0.751] −0.049 (1.716) 3.232*** [1.210]

Percent diplomas Asian −1.509 [1.259] −1.615 [1.236] −1.564 [1.202] −1.513 [1.175] 0.641 (3.603) −1.87 [1.931]

ln expenditures school district −0.401** [0.195] −0.383** [0.186] −0.384** [0.188] −0.412** [0.202] −0.483 (0.342) −0.175 [0.207]

Percent BA degree 0.463 [0.371] 0.721*[0.380] 0.811** [0.340] 0.851** [0.375] −0.606 (0.869) 1.715*** [0.465]

ln median income 0.184 [0.138] 0.121 [0.132] 0.117 [0.131] 0.117 [0.137] −0.468 (0.330) 0.496** [0.203]

Metro1 −0.930*** [0.142] −1.005*** [0.157] −0.949*** [0.140] −0.690*** [0.144] −0.595 (0.652) −0.986*** [0.218]

Metro2 −0.072 [0.077] −0.037 [0.067] −0.023 [0.063] 0.021 [0.074] −0.418** (0.206) 0.09 [0.114]

Metro3 0.064 [0.045] 0.048 [0.046] 0.053 [0.047] 0.083*[0.047] 0.071 (0.156) 0.152** [0.071]

Metro4 0.064 [0.072] 0.085 [0.068] 0.084 [0.068] 0.065 [0.076] 0.139 (0.187) −0.005 [0.088]

Observations 175 175 175 175 175 175

Censored observations 00004 0

R-squared 0.95 0.96 0.96 0.95 0.93

Pseudo R-squared 0.50

Notes: Robust standard errors are in brackets, and Tobit standard errors are in parentheses.

*Signiﬁcant at 10%.

** Signiﬁcant at 5%.

*** Signiﬁcant at 1%.

734 J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738

Percent Diplomas Black and Percent Diplomas Hispanic have

positive coefﬁcients of 0.101 and 1.964, respectively, but

only the latter coefﬁcient is statistically signiﬁcant. Percent

Diplomas Asian is the only one with a negative coefﬁcient

at −1.509, but the coefﬁcient is not statistically signiﬁcant.

Somewhat surprisingly, the coefﬁcient on school district

expenditures is −0.401 and statistically signiﬁcant. This

result may suggest that higher spending school districts

have to spend more per pupil because their students come

from more disadvantaged backgrounds and are less pre-

pared for college. It may also be consistent with arguments

by Hanushek (2003) and others that increased spending

on education does not result in improved educational out-

comes.

As expected, the percent of adults in the district with a

college degree and the median income in the district have

a positive effect on USG enrollment, but neither coefﬁcient

is signiﬁcant. Of the four metro dummies, only Metro1, the

dummy for Atlanta City Schools, is statistically signiﬁcant,

with a coefﬁcient of −0.930. This result suggests that grad-

uates from Atlanta City Schools are disproportionately less

likely to attend a USG institution. Many of the best students

may opt out of the district by enrolling in private schools

or by moving to nearby districts to avoid the problems that

often plague central city school districts such as Atlanta.

Because we treat the entire USG as the destination area,

we lose information on the importance of institutional

characteristics. In order to recover some of these institu-

tional characteristics, we look at different classiﬁcations of

USG institutions. We combine the ﬁve sectors of the USG

into two groups, colleges and universities, and we test if

enrolling in the USG is differentially affected by distance to

the nearest college and distance to the nearest university.16

Column (2) shows the results. Distance to the closest col-

lege has a signiﬁcant deterrent effect on enrolling in the

USG, with a distance elasticity of −0.122. Distance to the

closest university has an elasticity of −0.017 but is not

statistically signiﬁcant. Columns (3) and (4) estimate a sim-

ilar speciﬁcation including only one distance variable each,

with distance to the closest college in column (3) and dis-

tance to the closest university in column (4), to test if there

is an adverse effect from including the two distance vari-

ables in the same regression. Distance to the closest college

is virtually unchanged in its effect, and distance to the clos-

est university becomes positive but is still insigniﬁcant.

These results conﬁrm that the state and two-year col-

leges of the USG play an important role in bringing students

into the system. Bringing colleges closer to students low-

ers the costs of pursuing higher education and encourages

them to enroll in the USG. Results for the other variables

in columns (2), (3), and (4) are virtually unchanged from

column (1), except that the percentage of adults with a

bachelor’s degree is signiﬁcantly positive in columns (2),

(3), and (4), suggesting that districts with a more educated

populace enroll more students in the USG.

16 The “colleges” category includes the two state colleges and the thir-

teen two-year colleges, while “universities” includes the three research

universities, two regional universities, and thirteen state universities. See

also Note 11.

We are also interested in how the choice between col-

leges and universities is affected by the distance to the

nearest college and distance to the nearest university. Col-

umn (5) looks at enrollment in colleges (and is estimated

with Tobit methods), and column (6) looks separately at

enrollment in universities. Column (5) shows that college

enrollment is signiﬁcantly deterred by distance to the near-

est college with an elasticity of −0.647 but that the distance

to the nearest university increases college enrollment with

an elasticity of 0.131. Column (6) shows parallel results

for university enrollment. Distance to the nearest univer-

sity decreases university enrollment with an elasticity of

−0.088, but distance to the nearest college increases uni-

versity enrollment with an elasticity of 0.241. Together,

these results suggest that, when students choose between

attending a college or a university, they again take distance

into consideration in assessing the relative costs and ben-

eﬁts. In particular, students who live nearer to colleges are

more likely to attend colleges, and students who live nearer

to universities are more likely to attend universities. This

result is quite similar to the ﬁndings of Ordovensky (1995)

for the NCES High School and Beyond longitudinal survey

respondents. He ﬁnds that the probability of attending a

two-year college is decreasing in the distance to the near-

est two-year college and increasing in the distance to the

nearest four-year school, and that the probability of attend-

ing a four-year school is increasing in the distance to the

nearest two-year college and decreasing in the distance to

the nearest four-year school.

These results have important implications for the USG

and other state systems of higher education. Decreas-

ing distance to the nearest university has little effect on

the overall likelihood of enrolling in any USG institution,

but increases the likelihood of enrolling in a university.

Decreasing distance to the nearest college increases the

likelihood students will enroll in the USG by increasing

the likelihood that they will enroll in colleges. However,

decreasing distance to the nearest college also has the

negative effect of decreasing the number of ﬁrst-time fresh-

men enrolling in universities. These results suggest that we

can think of students as choosing between three ordered

choices: attending a university, attending a college, and

not attending either. The distance elasticities suggest that

there are two types of students making marginal choices.

First, there are students choosing between attending a col-

lege and attending neither. Decreasing the distance to the

nearest college increases the likelihood the students will

choose to attend college. Second, there are students who

have decided to pursue higher education but must choose

between attending a college or a university. A decrease in

the distance to the nearest university increases the likeli-

hood the students will choose a university. However, the

closer is the nearest college the more likely it is that a stu-

dent will choose a college.

These ﬁndings are important because colleges, espe-

cially two-year colleges, are not perfect substitutes for

universities. The apparent hope is that students in aca-

demic tracks at two-year colleges will eventually transfer

to four-year schools and obtain a bachelor’s degree. How-

ever, there is some evidence that this is not necessarily

the case. Alfonso (2006) ﬁnds that enrolling in a two-

J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738 735

Table 3

Attending a speciﬁc USG institution? USG gravity model estimates.

Variable ln ﬂow

ln distance −1.345*** (0.027)

ln GA public HS freshman (not i) 0.625*** (0.042)

ln expenditures USG 0.963*** (0.152)

Percent HOPE 0.011*** (0.003)

ln SAT −4.170*** (0.915)

ln cost −0.514*** (0.137)

ln centrality −0.279*** (0.085)

Percent Black USG −0.008*** (0.003)

Atlanta −0.272*** (0.098)

Research university 0.679*** (0.241)

Regional university 1.296*** (0.182)

State university 0.360** (0.140)

State college −0.771*** (0.105)

Observations 5775

Censored observations 3547

Pseudo R-squared 0.31

Notes: Tobit standard errors are in parentheses.

** Signiﬁcant at 5%.

*** Signiﬁcant at 1%.

year college decreases a student’s probability of ultimately

obtaining a bachelor’s degree relative to enrolling in a four-

year institution. The costs of migrating from a two-year to

a four-year school might deter some students from trans-

ferring after their two years are complete. It may also be

the case that two-year colleges provide a lower quality

education, so that students who transfer from a two-year

college are less well-prepared for upper level classes than

students who enrolled in four-your schools as freshmen. If

this is true, then the USG faces a major dilemma concerning

the expansion of two-year colleges. Increasing accessibil-

ity to two-year colleges will bring more students into the

system, but will divert some students from universities to

colleges where they may be less likely to eventually obtain

a bachelor‘s degree.

4.2. Conditional upon attending any USG institution,

attending a speciﬁc USG institution?

Thus far we have been concerned with how enrollment

in any USG institution is affected by distance to the near-

est college or university. In this subsection we examine

how various factors (including distance) affect one’s choice

among the different USG institutions. We estimate a stan-

dard gravity model that includes the ﬂows from each of

the 175 school districts to the 33 institutions, for a total

of 5775 ﬂows. As suggested by Sà et al. (2004), we esti-

mate the model using school district ﬁxed effects so that

all of the push variables disappear. This is done to con-

trol for unobserved heterogeneity in origin areas that is

imperfectly captured by the push factors. Because we are

examining more than one destination, we now include the

centrality index, the destination population, and the vari-

ous pull factors. We also include the Atlanta metropolitan

area institutional dummy variable and dummy variables

for the sectors of the institutions. Given the number of zero

ﬂows, we use Tobit estimation.

Table 3 reports the results. All of the variables are sig-

niﬁcant at the 5 percent level or greater. Distance has

a very strong negative effect, with an estimated elastic-

ity of −1.345. The destination population has a positive

coefﬁcient of 0.625 that is signiﬁcantly less than one. Insti-

tutional expenditures per pupil has a positive pull on

enrollment ﬂows with a coefﬁcient of 0.963, as does the

percentage of ﬁrst-time freshmen who are Georgia res-

idents receiving the HOPE scholarship with a coefﬁcient

of 0.011. However, SAT scores has a negative coefﬁcient

of −4.17, so the evidence is mixed as to whether stu-

dents are drawn to higher quality institutions. Higher cost

of attendance and a higher percentage of Black students

deter enrollment ﬂows with coefﬁcients of −0.514 and

−0.008, respectively. The centrality index has a coefﬁcient

of −0.279, suggesting that greater spatial competition via

proximity to other institutions decreases enrollment in a

given institution.

The gravity model in Table 3 includes all 33 USG institu-

tions, and so implicitly assumes that the distance elasticity

and other coefﬁcients are the same for every institution.

However, as pointed out by Ullis and Knowles (1975),it

is likely that the distance elasticity of enrollment demand

varies by institution. In particular, we might expect that

more prestigious and higher quality institutions have a less

elastic demand, and students may be more willing to travel

greater distances to attend a research university than a

two-year college. There also might be variables other than

educational quality that affect the distance elasticity.

To test this hypothesis we estimate for each USG insti-

tution a modiﬁed gravity model via the Tobit estimator,

including only the distance from the school district to

the institution and the push variables included in the

estimations in Table 2. Estimated distance elasticities of

enrollment demand for each of the 33 USG institutions are

reported in Table 4, along with standard errors and the

institution’s sector classiﬁcation in 2002. Schools are listed

in descending order by the magnitude of their distance

elasticities.17

Looking ﬁrst at the bottom of Table 4, we see that

three schools have distance elasticities that are not signiﬁ-

cantly different than zero. Georgia Institute of Technology’s

distance elasticity is positive but insigniﬁcant, Georgia

State University’s is negative and insigniﬁcant, and Georgia

Perimeter College’s is negative and close to signiﬁcant.18

All other distance elasticities are negative and signiﬁcantly

different from zero at the 1 percent level, and institutions

of the USG have distance elasticities that range from 0.016

(Georgia Tech) to −3.248 (Waycross College).

Echoing the sentiments of Lowe and Viterito (1989),

some institutions have statewide appeal while others have

a much more local appeal. While the institution-speciﬁc

distance elasticities are not perfectly ordered by sector,

there is a general pattern. The three research universities

of the USG have the lowest (in magnitude) distance elas-

ticities, and two-year and state colleges generally have the

highest elasticities; regional and state universities are gen-

17 The USG includes four institutions that are historically Black. These

are Albany State University, Savannah State University, Fort Valley State

University, and Atlanta Metropolitan College.

18 Georgia Perimeter College is somewhat of an outlier perhaps because

it is one of the few USG institutions with relatively large satellite cam-

puses.

736 J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738

Table 4

Institution-speciﬁc distance elasticities.

Institution Sector Elasticity Standard error

Waycross College Two-year college −3.248*** 0.584

Bainbridge College Two-year college −2.584*** 0.350

Dalton State College State college −2.476*** 0.332

Darton College Two-year college −2.161*** 0.187

Coastal Georgia Community College Two-year college −2.104*** 0.281

Gainesville College Two-year college −2.088*** 0.283

Georgia Highlands College Two-year college −2.085*** 0.213

Macon State College State college −2.060*** 0.272

Augusta State University State university −2.027*** 0.222

South Georgia College Two-year college −1.863*** 0.197

Gordon College Two-year college −1.614*** 0.207

Middle Georgia College Two-year college −1.563*** 0.171

East Georgia College Two-year college −1.384*** 0.159

Kennesaw State University State university −1.374*** 0.216

Georgia Southwestern State University State university −1.349*** 0.168

Columbus State University State university −1.320*** 0.156

Armstrong Atlantic State University State university −1.274*** 0.154

Clayton State University State university −1.270*** 0.381

Georgia College & State University State university −1.269*** 0.154

Atlanta Metropolitan College Two-year college −1.227*** 0.405

North Georgia College & State University State university −1.074*** 0.132

Abraham Baldwin Agricultural College Two-year college −1.033*** 0.106

Valdosta State University Regional university −0.910*** 0.091

Georgia Southern University Regional university −0.878*** 0.077

University of West Georgia State university −0.847*** 0.102

Southern Polytechnic State University State university −0.819*** 0.142

Fort Valley State University State university −0.719*** 0.129

Savannah State University State university −0.624*** 0.136

Georgia Perimeter College Two-year college −0.400 0.268

Albany State University State university −0.399*** 0.144

University of Georgia Research university −0.303*** 0.072

Georgia State University Research university −0.086 0.139

Georgia Institute of Technology Research university 0.016 0.134

Notes: All parameters are estimated using Tobit methods.

*** Signiﬁcant at 1%.

erally in between. This ordering supports the notion that

higher quality institutions in the USG have a less elastic

demand.

Lowe and Viterito (1989) show that the market penetra-

tion index of a private college is correlated with its mean

SAT score, which is consistent with the notion that more

selective schools have wider market areas. To investigate

this notion, we use OLS to regress the estimated elastic-

ities in Table 4 on numerous institutional characteristics

to explain the variance in the distance elasticity across

institutions, using the various pull variables and the sector

dummy variables. These results are reported in Table 5.

Table 5

Determinants of the variation in distance elasticity.

Variable Distance elasticity

Percent Black USG 0.011*** (0.003)

ln GA public HS freshmen 0.450*** (0.150)

ln centrality 0.276** (0.123)

Research university 0.929*** (0.293)

Regional university 0.546** (0.262)

State university 0.467** (0.180)

State college −0.391** (0.141)

Observations 33

R-squared 0.78

Notes: Robust standard errors are in parentheses.

** Signiﬁcant at 5%.

*** Signiﬁcant at 1%.

These results indicate that 78 percent of the variance

in the distance elasticities can be explained by seven vari-

ables: the percentage Black, Georgia public high school

freshmen enrollment, the centrality index, and the dummy

variables for research universities, regional universities,

state universities and state colleges. All coefﬁcients are sig-

niﬁcant, and all but the state college dummy variable are

positive. A positive sign means that an increase in a variable

decreases the absolute magnitude of the distance elasticity

and so makes demand less elastic. The sector dummies are

ordered as we might expect, except that state colleges have

a more elastic demand than the omitted group, two-year

colleges. Research universities have the lowest demand

elasticity, followed by regional universities and then state

universities. The centrality index has a positive coefﬁcient

suggesting that institutions facing greater spatial compe-

tition have lower distance elasticities and are therefore

less likely to get all of their students from their immediate

vicinity and must bring in students from outside areas. The

positive coefﬁcient on percent Black indicates that Black

students value attending a historically African–American

institution, and are willing to migrate greater distances to

do so.19 Schools with higher enrollments of Georgia pub-

19 We also experimented with replacing the percent Black variable in

Table 5 with an indicator variable for a historically Black college or univer-

J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738 737

lic high school graduate entering freshmen also have less

elastic demands with respect to distance. Not surprisingly,

this suggests that within sectors large schools are better

able to bring in students from further away.

5. Summary and conclusions

This paper examines how accessibility affects enroll-

ment in and within the University System of Georgia. We

ﬁnd several important results. First, greater distance to

the nearest college decreases the likelihood of enrollment

in any USG institution with an elasticity of −0.12, while

greater distance to the nearest university has little or no

effect on the likelihood of enrollment in the USG. However,

once students decide to enroll in the USG, they must choose

between colleges and universities. Greater distance to the

nearest college decreases the likelihood that a student will

attend a college and therefore increases the likelihood

that he or she will attend a university. Greater distance

to the nearest university has a similar effect: it decreases

the likelihood of attending a university and increases the

likelihood of attending a college. If colleges offer a lower

quality education than universities, then this presents a

dilemma for policymakers. Making colleges more acces-

sible will both bring more students into the system and

draw students away from universities. Making universities

more accessible does not bring in more students to the sys-

tem, but does encourage more to go to universities rather

than colleges, which may mean that those in the system

receive a better education. Clearly, administrators for the

USG as well as other states’ higher education systems need

to understand the implications of improving accessibility

for both colleges and universities.

We also estimate traditional distance elasticities for the

USG as a whole and for each institution of the USG sepa-

rately. We ﬁnd that the intrastate distance elasticity for the

USG is −1.345, which is very similar to previous ﬁndings by

McHugh and Morgan (1984) and Sà et al. (2004). However,

the intrastate distance elasticity varies across institutions,

ranging from roughly 0 for Georgia Institute of Technology

to −3.248 for Waycross College. This variance can in part be

explained by the academic classiﬁcation of the institution,

the percentage of the institution’s students that are Black,

freshmen enrollment in the institution, and the degree

of spatial competition the institution faces. Knowledge of

these distance elasticities can help institution adminis-

trators predict how their institution-speciﬁc demand will

increase as the population around them grows, and can

also help USG administrators promote greater enrollment

in the system through the creation of new institutions and

campuses.

Though the analysis in this paper is for Georgia, the

results are also likely to have important implications for

other states. There are some peculiarities to Georgia such

sity. The coefﬁcient on this indicator variable was positive and signiﬁcant

with a coefﬁcient of 0.810. Given that the mean percent Black for HBCUs

and non-HBCUs is 92.4 percent and 22.0 percent, respectively, the mag-

nitude of the HBCU indicator variable is consistent with what would be

expected based on the coefﬁcient of the percent Black variable.

as the presence of the HOPE Scholarship, but for the most

part we expect that the results would be similar for most

other states. Distance is likely to play an important role in

whether students enroll in higher education and distance

is likely to be very important for students in determining

the speciﬁc institution they attend.

Acknowledgements

We are grateful to the editor and to two anonymous

referees for many helpful comments.

References

Alfonso, M. (2006). The impact of community college attendance on bac-

calaureate attainment. Research in Higher Education,47(8), 873–903.

Ali, M. K. (2003). Analysis of enrollment: A spatial-interaction model. The

Journal of Economics,29(2), 67–86.

Amemiya, T. (1973). Regression analysis when the dependent variable is

truncated normal. Econometrica,41(6), 997–1016.

Baryla, E. A., Jr., & Dotterweich, D. (2001). Student migration: Do signiﬁcant

factors vary by region? Education Economics,9(3), 269–280.

Bergstrand, J. H. (1985). The gravity equation in international trade: Some

microeconomic foundations and empirical evidence. The Review of

Economics and Statistics,67(3), 474–481.

Card, D. (1995). Using geographic variation in college proximity to esti-

mate the return to schooling. In L. Christoﬁdes, R. Swideinsky, & E. K.

Grant (Eds.), Aspects of labour market behaviour: Essays in honor of John

Vanderkamp (pp. 201–222). Toronto: University of Toronto Press.

Desjardins, S. L., Dundar, H., & Hendel, D. D. (1999). Modeling the college

application process in a land-grant university. Economics of Education

Review,18(1), 117–132.

Fotheringham, A. S., Nakaya, T., Yano, K., Openshaw, S., & Ishikawa, Y.

(2001). Hierarchical destination choice and spatial interaction mod-

eling: A simulation experiment. Environment and Planning A,33(5),

901–920.

Ghatak, S., Levine, P., & Price, S. W. (1996). Migration theories and evi-

dence: An assessment. Journal of Economic Surveys,10(2), 159–198.

Gossman, C. S., Nobbe, C. E., Patricelli, T. J., Schmid, C. F., & Steahr, T. E.

(1968). Migration of college and university students in the United States.

Seattle: The University of Washington Press.

Greenwood, M. J. (1975). Research on internal migration in the United

States: A survey. Journal of Economic Literature,13(2), 397–433.

Hanushek, E. A. (2003). The failure of input-based schooling policies. Eco-

nomic Journal,113(485), F64–F98.

Heller, D. E. (1997). Student price response in higher education: An update

to Leslie and Brinkman. Journal of Higher Education,68(6), 624–659.

Hoxby, C. M. (Ed.). (2004). College choices: The economics of where to go,

when to go, and how to pay for it. Chicago: University of Chicago Press

for NBER.

Information Digest, 2002–2003. University System of Georgia,

Ofﬁce of Strategic Research and Analysis. Available online at

http://www.usg.edu/usg stats/info digest/2002/.

Kariel, H. G. (1968). Student enrollment and spatial interaction. Annals of

Regional Science,2(1), 114–127.

Kyung, W. (1996). In-migration of college students to the state of New

York. Journal of Higher Education,67(3), 349–358.

Leppel, K. (1993). Logit estimation of a gravity model of the college enroll-

ment decision. Research in Higher Education,34(3), 387–398.

Leslie, L. L., & Brinkman, P. T. (1987). Student price response in higher

education. Journal of Higher Education,58(2), 181–204.

Lowe, J. C., & Viterito, A. (1989). Differential spatial attraction of private

colleges and universities in the United States. Economic Geography,

65(3), 208–215.

McConnell, H. (1965). Spatial variability of college enrollment as a function

of migration potential. The Professional Geographer,17(6), 29–37.

McHugh, R., & Morgan, J. N. (1984). The determinants of interstate student

migration: A place-to-place analysis. Economics of Education Review,

3(4), 269–278.

Mixon, F. G., Jr. (1992a). Factors affecting college student migration across

states. International Journal of Manpower,13(1), 25–32.

Mixon, F. G., Jr. (1992b). A public choice note on college student migration.

International Journal of Manpower,13(3), 63–68.

Mixon, F. G., Jr., & Hsing, Y. (1994a). College student migration and human

capital theory: A research note. Education Economics,2(1), 65–73.

738 J. Alm, J.V. Winters / Economics of Education Review 28 (2009) 728–738

Mixon, F. G., Jr., & Hsing, Y. (1994b). The determinants of out-of-state

enrollment in higher education: A tobit analysis. Economics of Edu-

cation Review,13(4), 329–335.

Ordovensky, J. F. (1995). Effects of institutional attributes on enrollment

choice: Implications for postsecondary vocational education. Eco-

nomics of Education Review,14(4), 335–350.

Sà, C., Florax, R. J. G. M., & Rietveld, P. (2004). Determinants of the regional

demand for higher education in the Netherlands: A gravity model

approach. Regional Studies,38(4), 375–392.

Tuckman, H. P. (1970). Determinants of college student migration. South-

ern Economic Journal,37(2), 184–189.

Ullis, J. J., & Knowles, P. L. (1975). A study of the intrastate migration of

Washington college freshmen: A further test of the gravity model.

Annals of Regional Science,9(1), 112–121.