Full text of DOI:10.1111/0002-9092.00159

The Migration of Young Adults from
Non-Metropolitan Counties
Bradford Mills and Gautam Hazarika
Thisarticle examinesyoung adult migration from non-metropolitan countiesto either different
non-metropolitan countiesor to metropolitan area.sThe resultsshow that expected gainsin initial
earningsprovide young entrantsto the labor force with a marked incentive to migrate from
their non-metropolitan countiesof origin. Initial earningsgainsstem, in part, from higher returns
to schooling in both metropolitan areas and other non-metropolitan counties. The propensity to
migrate is also sensitive to the costs of migration, which, in turn, are correlated with paternal
education and the local presence of extended family.
Key words: earnings, migration, non-metropolitan young adults, returns to schooling.
Net migration flowsbetween non-metro-
politan and metropolitan areasof the United
Stateshave reveresd three timesover the
past thirty years (Fuguitt and Beale). Net out-
migration from non-metropolitan areasprior
to the 1970sand during the 1980sled to calls
for policiesto fotser the retention of rural
populations, while concern largely focused
on the preservation of “rural America” and
the management of growth during periodsof
net in-migration. Non-metropolitan counties
have consistently been concerned with retain-
ing productive labor, given high migration
propensities among educated young adults
and the aging of retained populations. Suc-
cessful rural development initiatives aimed
at retaining young adults must be based
on a thorough understanding of the rea-
sonsfor young adult out-migration. Thisarti-
cle exploresthe economic factorsunderlying
such out-migration from non-metropolitan
counties.
the underlying componentsof the migration
decisions of non-metropolitan area residents
in the United Stateshave either focuesd
exclusively on persistent non-metropolitan–
metropolitan area earningsgapsor on
the monetary and psychic costs of migra-
tion (Hoch; Deaton, Morgan, and Anschel;
Broomhall and Johnson). A related body
of literature hasexamined wage differen-
tialsfaced by workersemployed in farming
to explain labor flowsfrom the esctor (e.g.,
Gisser and Davila). This article builds on
these efforts by developing a formal model
of the migration of young adultsfrom non-
metropolitan countiesin the United States
either to other non-metropolitan countiesor
to metropolitan areas. The model incorpo-
ratesgainsin initial hourly earningsand
broadly defined migration costs encompass-
ing the financial, psychic, and employment
attainment costs of migration.
The empirical tsrategy employed isasfol-
low.sInitial hourly earningsfunctionsfac-
ing young migrantsand non-migrantsare
estimated using a sample of 1476 individu-
alsfrom the National Longitudinal Survey
of Youth (NLYS), who resided in non-
metropolitan areasat age 14. Since earn-
ingsupon migration are only obesrved for
migrantsand earningsupon staying are only
observed for non-migrants, the maximum
likelihood method of type 2 tobit (Amemiya)
isemployed to recover eslectivity-corrected
etsimatesof an initial hourly earningsfunc-
tion for each group. For every individual in
the sample, migrant or non-migrant, these
Frameworksfor identifying the benefits
and the monetary and non-monetary costs
of migration may be traced back to the
seminal paper by Sjaastad. Empirical meth-
ods to consistently estimate the benefit and
cost components of migration decisions have
previously been applied (e.g., Robinson and
Tomes). However, earlier attempts to identify
Bradford F. Mills is an assistant professor in the Department of
Agricultural and Applied Economicsat Virginia Tech. Gautam
Hazarika is a visiting assistant professor in the Department of
Agricultural and Applied Economicsat Virginia Tech.
The authorsthank Jeff Alwang, Spiro Stefanou, and two
anonymousreviewersfor helpful comments.
Amer. J. Agr. Econ. 83(2) (May 2001): 329–340
Copyright 2001 American Agricultural Economics Association

330
May 2001
Amer. J. Agr. Econ.
consistent estimates are used to predict ini- work experience in standard Mincer earnings
tial hourly earningsupon migration and upon functions(e.g., Willisand Roesn). It follows
remaining in the county of origin. Next, con- that
sistent estimates of the influence of individ-
M t
(1) Wiꢀ tꢀ M = Wiꢀ 0ꢀ M eg
and
ual and county characteristics on the cost
component of the migration decision are
recovered through probit estimation of a
structural equation that includes the log of
N t
(2) Wiꢀ tꢀ N = Wiꢀ 0ꢀ N eg
the ratio of predicted initial hourly earnings
upon migration to predicted initial hourly
where Wiꢀ tꢀ M and Wiꢀ tꢀ N are, respectively,
earningsupon remaining in the county of ori-
worker i’searningsfrom migrating and from
gin asan explanatory variable.
remaining in the non-metropolitan county
The results highlight the important roles
of origin at time t after commencement of
that factors associated with initial earn-
employment.
ingsdifferencesand migration cotsplay
The transition from school to work is not
frictionless. Assume that, upon completing
in explaining the migration behavior of
young adults. Unlike typical models of
cshool, workersreceive job offersgenerated
rural to urban migration, the earnings
by stochastic processes. Worker i expectsto
and cost components of the migration
receive a job offer with probability λiꢀ M if
decision induce migration to other non-
she migrates upon completing school. On the
metropolitan areasaswell asmetropoli-
other hand, the worker expectsto receive a
tan areas. Notably high returns to schooling
job offer with probability λiꢀ N in the local
make other non-metropolitan countiesa par-
non-metropolitan labor market. These prob-
ticularly attractive migration destination for
abilitiesare functionsof labor market condi-
educated non-metropolitan youth. The find-
tionsand worker characteristics.
ingsimply that non-metropolitan areasare
The worker’splanning horizon isa period
in aggregate able to provide remunerative
of duration T . Denote the subjective rate
employment opportunitiesto well educated
of discount as ri, letting ri ꢀ gM ꢀ gN . Then,
non-metropolitan youth. However, aslabor
given (1), the expected present value of
marketsfor high ksill groupsare often
the worker’searningsupon migration over
relatively thin in specific non-metropolitan
time T is
areas, educated young adults may not find
ꢀ
T
Wiꢀ 0ꢀ M e−(r −g )t dt
these remunerative positions in their non-
i
M
(3) Viꢀ M = λiꢀ M
metropolitan county of origin.
0
ꢁ
ꢂ
These findings are developed in the
remainder of the article. The next section
presents a simple model of migration. The
empirical specification, estimation method,
and data are then discussed, followed by
empirical results and simulations. The article
concludeswith implicationsfor rural devel-
opment policy.
Wiꢀ 0ꢀ M
iꢀ M ri − gM
M )T
1 − e−(r −g
ꢁ
i
= λ
Similarly, given (2), the expected present
value of her earningsover time T in the local
non-metropolitan area is
ꢀ
T
Wiꢀ 0ꢀ N e−(r −g )t dt
i
N
(4) Viꢀ N = λiꢀ N
0
ꢁ
ꢂ
Wiꢀ 0ꢀ N
iꢀ N ri − gN
N )T
1 − e−(r −g
ꢁ
i
= λ
A Model of Young Adult Migration
With negligible costs of migration, worker i
Individual i, resident of a non-metropolitan
area, ison the verge of completing hisor
her desired level of schooling and enter-
ing the labor force. Initial hourly earning
would migrate if
Viꢀ M
ꢀ 1ꢁ
Viꢀ N
iꢀ 0ꢀ M . The
upon migration isdenoted asW
growth rate or earningsupon migration is
denoted as gM . Similarly, initial earningsand
the growth rate of earningsin the local non-
metropolitan labor market are denoted as
Wiꢀ 0ꢀ N and gN , respectively. These growth
ratesmay be interpreted asthe coefficientsof
On the other hand, if the costs of migration
are substantial, the condition for migration is
modified as
Viꢀ M
(5)
ꢀ Ctꢁ
Viꢀ N

Mills and Hazarika
Migration of Young Adults from Non-Metropolitan Counties
331
where Ci isan index of the financial and
psychic costs of migration, Ci ꢀ 1. Hence,
taking logsof both sidesof (5) and using (3)
and (4), worker i would migrate only if
Assume that
(8)
ln Wiꢀ 0ꢀ M = Xi1B1 + ei1
(9)
ln Wiꢀ 0ꢀ N = Xi2B2 + ei2
ln λiꢀ M − ln λiꢀ N
ꢃ
ꢄ
ln(λiꢀ M ) + ln(Wiꢀ 0ꢀ M ) − ln(ri − gM )
(10)
ꢁ
ꢂ
M )T
+ ln 1 − e−(r −g
− ln(λiꢀ N
)
i
+ α1ri − ln Ci = Xi3B3 + ei3
− ln(Wiꢀ 0ꢀ N ) + ln(ri − gN )
where the Xi are exogenous regressors and
ꢁ
ꢂ
the ei are error terms. The term (ln Wiꢀ 0ꢀ M
ln Wiꢀ 0ꢀ N ) in (7) isdeigsnated the initial
hourly earningscomponent of individual
i’smigration deciison, whereasthe portion
−
N )T
− ln 1 − e−(r −g
ꢀ ln Ciꢁ
i
Let
ꢃ
ꢄ
ln λiꢀ M − ln λiꢀ N + α1ri − ln Ci istaken to
(6) Ii∗ = ln(λiꢀ M ) − ln(λiꢀ N ) + ln(Wiꢀ 0ꢀ M
− ln(Wiꢀ 0ꢀ N ) − ln(ri − gM )
+ ln(ri − gN )
)
be the cost component.1 The term (ln λiꢀ M
−
ln λiꢀ N ), the log of the ratio of the prob-
ability of finding a job upon migration to
the probability of finding a job in the
local non-metropolitan labor market, may
be considered related to the net employ-
ment attainment cost of migration. Substitut-
ing (8)–(10) in (7) yields
M )T
+ ln 1 − e−(r −g
i
[
]
N )T
− ln 1 − e−(r −g
− ln C ꢁ
i
[
]
i
Ii∗ may be interpreted asthe worker’slatent
(11) Ii∗ = Ziꢀ + vi
tendency to migrate. Worker i would migrate
if Ii∗ ꢀ 0 and choose to remain in the county
of origin if Ii∗ ≤ 0.
where Zi is the union of the regressors Xi1 to
Xi3, and vi consists of an error from approxi-
mation and a linear combination of the errors
ei1 to ei3. While I∗ isunobserved, the migrant
status of worker i is known. Assuming vi is
normally distributed, the equation
Empirical Specification and Estimation
By a Taylor series approximation to the non-
linear termsin (6) around the mean value
of riꢀ r¯,
Ii∗
σv
ꢀ
i σv
vi
(12)
= Z
+
σv
(7) Ii∗ ꢁ α0 + (ln Wiꢀ 0ꢀ M − ln Wiꢀ 0ꢀ N
)
σv = Var(vi), may be estimated by probit ML.
Initial hourly earningsof workerswho
completed schooling and entered the labor
force before 1979, the first year of the NLSY,
are unavailable. Hence, hourly earningsequa-
tionsare estimated as
+ (ln λiꢀ M − ln λiꢀ N ) + α1ri
− ln Ci
where
1
1
α1 =
−
(8.1) ln Wiꢀ 21ꢀ M = Xi1B1 + gU t21 + ei1
(9.1) ln Wiꢀ 21ꢀ N = Xi2B2 + gRt21 + ei2
r¯ − gN
r¯ − gM
Te−(r¯−g
Te−(r¯−g
M )T
N )T
+
−
1 − e−(r¯−g
1 − e−(r¯−g
M )T
N )T
α0 = ln(r¯ − gN ) − ln(r¯ − gM )
+ ln[1 − e−(r¯−g
]
M )T
where the Wiꢀ 21 are worker i’shourly earn-
ingsat age 21 or the earliets year there-
− ln[1 − e−(r¯−g
r¯
]
N )T
r¯
−
−
+
1 The individual’s subjective rate of discount ri isarguably not
part of the cost component, but because it is naturally unobserv-
able it must be treated as a function of the same set of exogenous
regressors.
r¯ − gN
r¯ − gM
T r¯e−(r¯−g
T r¯e−(r¯−g
M )T
N )T
+
)T ꢁ
1 − e−(r¯−g
1 − e−(r¯−g
M )T
N

332
May 2001
Amer. J. Agr. Econ.
after when observed out of school, t21 iswork at age 21 or the earliest year thereafter when
experience at thistime, and the g are, aspre-
viously defined, the growth rates of earnings
upon migration and remaining in the local
non-metropolitan area.2
observed out of school were not available for
420 of these individuals. County of residence
wasunavailable for three additional individ-
uals, leaving a study sample of 1476 per-
sons. An individual is considered a migrant
if, by age 21 or the earliest year thereafter
when observed out of school, he or she is
found residing in a county other than the
non-metropolitan county of residence at 14.
Migrantsto metropolitan areasand to other
non-metropolitan areascomprise 30.8% and
15.7%, respectively, of the total sample.3
Supplemental data on county unemploy-
ment ratesfor the years1975 and 1979 were
obtained from the Bureau of Labor Statis-
tics. Information on 1975 and 1980 county per
capita income, 1980 population per square
mile, and 1980 agricultural and manufac-
turing employment shares were obtained
from the 1983 County and City Data Book,
U.S. Department of Commerce, Bureau
The log of worker i’sobserved hourly earn-
ing isln Wiꢀ 21ꢀ M = Xi1B1 + gU t21 + ei1 if I ∗ =
Ziꢀ + vi > 0 and ln Wiꢀ 21ꢀ N = Xi2B2 + gRt21 +
ei2 if I∗ = Ziꢀ + vi ≤ 0. Since cov(ei1ꢀ vi)
ꢂ =0
and cov(ei2ꢀ vi)
ꢂ =0, estimating (8.1) and (9.1)
by OLS applied, separately, to samples of
migrantsand non-migrantsmay yield incon-
sistent estimates for the full sample. Assum-
ing ei1 and vi, and ei2 and vi, are bivariate
normally distributed, (8.1) and (12), and then
(9.1) and (12), may be consistently estimated
by the maximum-likelihood method of type 2
tobit (Amemiya).
With consistent estimates of B1 and B2,
ꢅ
ꢅ
namely B1 and B2, worker i’sexpected ini-
tial hourly earningswith no work experience
ꢀ
upon migration ln Wiꢀ 0ꢀ M and upon remain-
ꢀ
ing in the county of origin ln Wiꢀ 0ꢀ N may be of the Census. County-level employment-
ꢅ
ꢅ
growth data for the years1975 to 1979
were obtained from the 1996 Regional Eco-
nomic Information System, U.S. Bureau of
Economic Analysis.
generated as Xi1B1 and Xi2B2, respectively.
It follows that a consistent estimate of B3
up to a factor of proportionality may be
obtained by applying probit ML to the struc-
tural equation
Log hourly earning in the local non-
metropolitan area, ln Wiꢀ 21ꢀ N , ipssecified
asa function of individual human capital
attributes(GRADE, AFQT80, EXPER),
gender (MALE), race (BLACK), ethnicity
(HISPANIC), local economic conditions
(AGSHARE, MANUSHAR, COUNEMP,
EMPGR, YPERCAP, POSQML), and
dummy indicatorsof region (NC14, SO14,
WE14). Log hourly earningsupon migra-
tion, ln Wiꢀ 0ꢀ M , are specified as a function
only of individual human capital attributes,
gender, race, and ethnicity, since destination
area characteristics are not observed for
non-migrantsfor the purpoes of generating
predicted initial earningsfrom migration.
Yearsof cshooling, the Armed Forces
Qualifying Test measure of basic skills, and
yearsof experience are all expected to be
positively related to individual stocks of
human capital and, therefore, to hourly earn-
ings. Gender, race, and ethnicity character-
istics are included because females, as well
asracial and ethnic minoritie,shave been
shown to earn less in both metropolitan
ꢃ
ꢄ
∗
ꢀ
ꢀ
(13) Ii = α0 + ln Wiꢀ 0ꢀ M − ln Wiꢀ 0ꢀ N
+ Xi3B3 + εi
with εi denoting the regression error.
The structural equation (13) decomposes
worker i’smigration propenisty
Ii∗ into its
ꢃ
ꢄ
ꢀ
ꢀ
earningscomponent, ln Wiꢀ 0ꢀ M − ln Wiꢀ 0ꢀ N
,
and itscots component,
Xi3B3. Note that
identification of (13) requiresthat Xi1 or Xi2
containsat leats one variable that isnot
included in Xi3.
Data and Variables
The primary data source for the study is
the NLSY, a unique panel of 12,686 indi-
viduals14 to 21 yearsof age in 1979 that
hasbeen reusrveyed annually. Of the full
sample, 1899 individuals resided in a non-
metropolitan county at age 14. Earningsdata
2 Mincer earningsfunctionsoften include a squared experience
term to account for growing depreciation of human capital are
workersage. The qsuared experience term isdropped in this
study because all individuals are close to the date of first employ-
ment. Others(e.g. Lazear) have decomposed returnson experi-
ence into aging and on-the-job training components.
3 Most movesto other non-metropolitan areaswere not to an
adjacent county, as72.4% of the migrantsmoved to a different
multi-county labor market area asdefined by Tolbert and Sizer.

Mills and Hazarika
Migration of Young Adults from Non-Metropolitan Counties
333
and non-metropolitan labor markets, con- hourly earnings, but it is unlikely that they
trolling for levelsof human capital (ese, are observed beforehand by employers and
among others, Darity and Mason). Non- are, thus, taken not to influence the proba-
metropolitan countieswith high agricultural bilitiesof a job offer. Neither isthere reason
shares of employment may have low hourly to believe that these basic skills influence the
earnings, given relatively low wages in the financial and psychic costs of migration or the
sector (Mills). Similarly, since manufactur- rate of discount.
ing wagesare relatively high, hourly earn-
ingsmay be relatively high in countieswith
Education may influence the cost com-
ponent of the migration decision through
high manufacturing shares. The effect of the relative costs of employment attainment
county unemployment rateson hourly earn-
ingsisan open empirical quetsion. High
county unemployment ratesare often indica-
upon migration and upon forgoing migration.
However, thiseffect may be muted if dif-
ferencesin ratesof return to cshooling of
tive of a local surplus of unskilled labor migrantsand non-migrantsadjust to compen-
with low earningspotential (Cheisre). How-
ever, the earningsof other ksill groupsmay
sate migrants for migration costs entailed in
meeting regional demandsfor psecific types
not be similarly influenced. High unemploy- of skilled labor. Gender, race, and ethnic-
ment ratesmay alos be indicative of high
ratesof employment turnover and a preva-
ity may also influence migration costs if
discrimination affects the relative costs of
lence of vacancies, resulting in upward pres- employment attainment or if groupsexperi-
sure on local earnings. If employment growth ence different psychic costs to migration. For
stems from an increase in labor demand, it example, young men may feel more pressure
will also be associated with increased hourly than women to remain in their countiesof
earnings. However, the influence of employ- origin in order to tend to family financial
ment growth may, like the influence of unem- concerns.
ployment rates, be concentrated in specific
Strong local economic conditionsare gen-
sectors or skill groups. County income per- erally expected to decrease the costs of
capita and population per square mile may attaining employment locally and, thus, to
be positively associated with hourly earn- deter migration. However, aswith initial
ings because of their positive relationships earnings, the strength of the influence of
with the cost-of-living. Population density has unemployment ratesand employment growth
also been shown to be positively associ- ratesmay depend on whether unemploy-
ated with productivity and, therefore, wages ment and employment growth are concen-
(Ciccone and Hall). Dummy indicatorsof
three of the four U.S. censusregionsare also
included in the earningsspecification to con-
trol for regional differencesin cots-of-living
and labor productivity.
trated in specific sector and skill groups
or are broad based. As mentioned, high
unemployment ratesmay tsem from rapid
employment turnover and an abundance
of vacancies. Long-term declines in non-
As discussed, log initial hourly earnings metropolitan area agricultural and manufac-
upon migration and upon remaining in the turing sector employment shares suggest that
local non-metropolitan area are predicted individualsin countieswith high employ-
using type-2 tobit estimates of the wage ment shares in these sectors may face
equations(8.1) and (9.1) with EXPER est
relatively high costs of employment attain-
to zero since individuals have no work expe- ment. Individualsfrom countieswith high
rience at the time of entry to the labor shares of agricultural employment may also
force. These predictions create the variable have stronger ties to the community that
ln Wiꢀ 0ꢀ M − ln Wiꢀ 0ꢀ N , termed DIFFEARN, increase the psychic cost of migration and
in the structural equation (13). Consistent mitigate the influence of agricultural employ-
etismatesof factorsinfluencing the cots
ment share on the propensity to migrate. Pop-
component of the migration decision are ulation density will generally lower the costs
recovered by including all variablesfrom
the earningsequations(8) and (9) in the
of local employment attainment because
densely populated areas provide a larger
structural equation (13) except AFQT80. The array of employment opportunitiescompat-
AFQT80 variable isexcluded to meet identi- ible with an individual’s specific skills and
fication conditions. The basic skills measured occupational expertise (Kim).
by the AFQT score influence worker pro-
Three variables(PAGRADE, MAGRADE,
ductivity and quickly manifest themselves in NOATTACH) not included in the earnings

334
May 2001
Amer. J. Agr. Econ.
equationsappear in (13) to capture addi-
hourly earningsfrom migration. The etsi-
tional financial and psychic costs of migra- mated ratesof return to schooling and expe-
tion. More educated parents, particularly rience upon migration are 6.3% and 5.8%,
fathersisnce they are more likely to partic-
ipate in the labor force, can provide their
children with better information on employ-
ment opportunitiesoutisde the local labor
market. Thu,sfather’seducation isexpected
to lower employment attainment costs in the
destination area labor market. The variable
NOATTACH, taking the value 1 if the indi-
vidual’smother wasborn outside the individ-
ual’s state of residence at 14 and the value 0
otherwise, indicates a local absence of matri-
lineal extended family that may reduce the
psychic cost of migration.
Descriptive statistics of all variables in the
analysis are presented by migration status
in table 1. Migrantshsow a higher level
of educational attainment, have more edu-
cated parents, score higher on the Armed
ForcesQualification Test, and receive higher
hourly earnings. Migrants also tend to come
from countieswith higher agricultural sector
shares of total employment, lower manufac-
turing shares of total employment, and lower
population densities. Migrants are also more
likely to have mothersnot born in the state.
respectively. Both parameter estimates are
statistically different from zero at the p =
0ꢁ01 level.
To ascertain whether rates of return to
schooling and experience are uniformly low
in non-metropolitan countiesrelative to
metropolitan counties, hourly earnings equa-
tionsfor an expanded tri-tsate migration
model incorporating the decisions to migrate
to a metropolitan area, to migrate to another
non-metropolitan county, or to remain in the
county of residence at 14 are also estimated.
Consistent estimates of the three earnings
equationsin thismodel are obtained by
including the selectivity correcting inverse
Millsratiosgenerated from multinomial logit
estimation in the corresponding OLS earn-
ings regressions (Lee). Structural parameter
etsimatesof the earningsand cots compo-
nents of these tri-state migration decisions
cannot be recovered, however. Complete
identification requiresthat each earnings
equation contain an explanatory variable not
included in the others. While the earnings
equation for the non-metropolitan county of
origin containscounty characteritsicsomit-
ted from the other two earningsequa-
tions, no identifying variables with respect to
metropolitan migrant and non-metropolitan
migrant earningsequationsare plauibsly
justified. The estimates, presented in table
3, reveal that ratesof return to cshool-
ing and experience are very high, 9.3%
and 8.3%, upon migration to another non-
metropolitan county. In fact these rates of
return are higher than returnsto cshool-
ing and experience aupon migration to a
metropolitan area.
Results and Simulations
Consistent estimates of the log hourly earn-
ingsequations(8.1) and (9.1) are reported
in table 2. Being male, AFQT score, county
employment growth, and residence in the
Western region of the United States all show
positive and statistically significant relation-
shipswith hourly earningsupon remaining in
the non-metropolitan county of origin. The
estimated returns to an additional year of
schooling are also positive, 3.3%, but the
parameter estimate is not statistically differ-
ent from zero at conventional levels. Simi-
larly, the estimated rate of return to work
experience is2.3%, but not isgnificantly dif-
ferent from zero. Further, the error term
in the county of origin log hourly earnings
equation iscorrelated with the error term
in the reduced form migration equation (12).
Thus, OLS parameter estimates of the earn-
ingsequation for individualsforgoing migra-
tion would provide biased estimates of the
potential earningsfor all individual.sBeing
Probit estimates of the structural
equation (13) are presented in table 4.
The probability of migration from a non-
metropolitan county isfound to significantly
increase in DIFFEARN, the log of the ratio
of predicted starting hourly earnings upon
migration to predicted hourly earningsin
the county of origin. After controlling for
these initial earnings differences, the other
parameter estimates represent the influence
of variableson the cots component of the
migration decision. For example, being male
is negatively associated with the propensity
male and AFQT scores also show positive to migrate at the p = 0ꢁ10 level, i.e., pos-
and statistically significant relationships with itively associated with migration costs that

Table 1.Descriptive Statistics
County StayersMtoigOratnhtesr Counties
Variable
Description
Mean
St. Dev.
Mean
St. Dev.
GRADE21∗∗
Years of schooling at age 21 or at completion of schooling
Male = 1
12ꢁ347
0ꢁ529
0ꢁ169
0ꢁ059
9ꢁ939
10ꢁ729
0ꢁ110
0ꢁ227
0ꢁ078
1ꢁ817
0ꢁ499
0ꢁ375
0ꢁ235
3ꢁ576
2ꢁ711
0ꢁ081
0ꢁ123
0ꢁ025
13ꢁ295
0ꢁ474
0ꢁ141
0ꢁ071
11ꢁ138
11ꢁ265
0ꢁ125
0ꢁ202
0ꢁ076
2ꢁ612
0ꢁ500
0ꢁ348
0ꢁ257
3ꢁ847
2ꢁ895
0ꢁ090
0ꢁ116
0ꢁ030
MALE∗∗
BLACK
Black = 1
HISPANIC
Hispanic origin = 1
PAGRADE∗∗
MAGRADE∗∗
AGSHARE∗∗
MANUSHAR∗∗
COUNEMP∗
Father’syearsof schooling
Mother’syearsof schooling
Agriculture sector share of employment
Manufacturing sector share of employment
Average 1975 to 1979 rate of unemployment,
county of origin
EMPGR
Average 1975 to 1979 rate of employment
growth, county of origin
1980 population (000) per sq. mile, county of origin
Average 1975 and 1980 per capita income ($000),
county of origin
0ꢁ021
0ꢁ020
0ꢁ020
0ꢁ021
POSQML∗∗
YPERCAP
0ꢁ663
6ꢁ174
0ꢁ539
1ꢁ322
0ꢁ520
6ꢁ159
0ꢁ446
1ꢁ273
NOATTACH∗∗
NC14∗∗
Mother not born in state = 1
0ꢁ217
0ꢁ332
0ꢁ528
0ꢁ081
40ꢁ697
2ꢁ939
1ꢁ477
0ꢁ412
0ꢁ471
0ꢁ500
0ꢁ273
27ꢁ657
1ꢁ512
0ꢁ487
0ꢁ357
0ꢁ412
0ꢁ437
0ꢁ108
51ꢁ618
2ꢁ607
1ꢁ625
0ꢁ479
0ꢁ493
0ꢁ496
0ꢁ311
29ꢁ977
1ꢁ840
0ꢁ524
North Central region of origin = 1 (Northeast = 0)
South region origin = 1
SO14∗∗
WE14∗
West region origin = 1
AFQT80∗∗
EXPER∗∗
LNEARN∗∗
% score on 1980 Armed Forces Qualification Test
Yearsin labor force after completion of schooling
Logarithm of hourly earningsin 1979 dollars
Number of observations
801
675
∗
Note: ∗∗ and indicate difference in county non-migrant and county migrant samples at P = 0ꢁ05 level and P = 0ꢁ10 level, respectively.
a t T h e U n i v e s i t y o f C a l g a r y o h n t t M p : a / / y a j 2 a 7 e , . o 2 x 0 f 1 o 5 r d j o D u r o n w a l n s l . o o a r d g e / d f r o m

336
May 2001
Amer. J. Agr. Econ.
Table 2.Log Hourly Earnings Equations
Out-of-County
In-County
Coefficient
ASE
Coefficient
ASE
CONSTANT
GRADE21
MALE
0ꢁ212
0ꢁ063
0ꢁ240
0ꢁ286
0ꢁ033
0ꢁ204
0ꢁ032
0ꢁ043
0ꢁ003
0ꢁ023
−0ꢁ052
0ꢁ211
−0ꢁ222
2ꢁ695
0ꢁ017
0ꢁ077
0ꢁ108
0ꢁ117
0ꢁ284
0ꢁ375
0ꢁ414
0ꢁ014∗∗∗
0ꢁ042∗∗∗
0ꢁ065
0ꢁ024
0ꢁ195
0ꢁ038∗∗∗
0ꢁ068
BLACK
−0ꢁ9E−3
HISPANIC
AFQT80
EXPER
0ꢁ012
0ꢁ076∗∗∗
0ꢁ090
0ꢁ005
0ꢁ001
0ꢁ001∗∗∗
0ꢁ027
0ꢁ058
0ꢁ019∗∗∗
AGSHARE
MANUSHAR
COUNEMP
EMPGR
YPERCAP
POSQML
NC14
0ꢁ381
0ꢁ246
0ꢁ940
1ꢁ111∗∗
0ꢁ020
0ꢁ050
0ꢁ097
SO14
0ꢁ098∗∗
WE14
0ꢁ119
Rho
0ꢁ203
0ꢁ186
0ꢁ136∗∗∗
∗Indicatessignificance in a two-tailed t-test at the P = 0ꢁ10 level.
∗∗Indicatessignificance in a two-tailed t-tes t at the P= 0ꢁ05 level.
∗∗∗Indicatessignificance in a two-tailed t-tes t at the P= 0ꢁ01 level.
Table 3.Selectivity Corrected Log Hourly Earnings Equations in Three State Model
Migration Status
Metropolitan
Coefficient ASE
Other Non-Metropolitan
Non-Migrant
Coefficient
ASE
Coefficient
ASE
CONSTANT
GRADE21
MALE
0ꢁ519
0ꢁ048
0ꢁ175
0ꢁ010
0ꢁ109
0ꢁ005
0ꢁ059
0ꢁ372
−0ꢁ203
0ꢁ093
0ꢁ485
0ꢁ292
0ꢁ031
0ꢁ218
0ꢁ028
0ꢁ047
0ꢁ003
0ꢁ018
−0ꢁ120
0ꢁ235
−0ꢁ303
2ꢁ672
0ꢁ019
0ꢁ080
0ꢁ120
0ꢁ117
0ꢁ286
0ꢁ215
0ꢁ342
0ꢁ019∗∗
0ꢁ048∗∗∗
0ꢁ075
0ꢁ028∗∗∗
0ꢁ065∗∗∗
0ꢁ114∗
0ꢁ122∗∗∗
0ꢁ001
0ꢁ021
0ꢁ213
0ꢁ034∗∗∗
0ꢁ054
BLACK
−0ꢁ034
−0ꢁ208
0ꢁ004
HISPANIC
AFQT80
EXPER
0ꢁ092∗∗∗
0ꢁ073
0ꢁ001
0ꢁ001∗∗∗
0ꢁ023
0ꢁ024∗∗
0ꢁ083
0ꢁ036∗∗
AGSHARE
MANUSHAR
COUNEMP
EMPGR
YPERCAP
POSQML
NC14
0ꢁ348
0ꢁ199
0ꢁ759
1ꢁ044∗∗
0ꢁ017
0ꢁ047∗
0ꢁ082
SO14
0ꢁ081∗∗∗
WE14
0ꢁ101
Rho
−0ꢁ013
0ꢁ125
0ꢁ041
0ꢁ176
0ꢁ140∗∗∗
∗Indicatessignificance in a two-tailed t-test at the P = 0ꢁ10 level.
∗∗Indicatessignificance in a two-tailed t-tes t at the P= 0ꢁ05 level.
∗∗∗Indicatessignificance in a two-tailed t-tes t at the P= 0ꢁ01 level.
inhibit migration. Similarly, father’seduca-
providing the young adult with better infor-
tion and mother’s birth outside of the state mation on potential employment opportuni-
are positively related to the propensity to ties, while absence of a matrilineal extended
migrate. As discussed, father’s education is family in the area reduces the psychic costs
likely to decrease the cost of employment of migration. Individual schooling levels and
attainment in destination labor markets by local economic conditionsdo not hsow a

Mills and Hazarika
Migration of Young Adults from Non-Metropolitan Counties
337
Table 4.Structural Probit Equations
“typical” ratio of initial hourly earningsupon
migration to initial hourly earningsin the
county of origin will result in a 7.9 percentage
point increase in the probability of migration.
Since the earningsgap increasesin schooling,
raising schooling will increase the probabil-
ity of migration via the earningscomponent
of the migration decision. The probit estimate
of the structural equation (13) also reveals
that schooling does not have a statistically
significant effect on the cost component of
the migration decision. Combining the earn-
ingsand cots componentsin the probit etsi-
mate of the reduced form equation (12), the
net marginal effect of an additional year of
schooling on the probability of migration is a
2.8 percentage point increase in the probabil-
ity of migration.
Coefficient
ASE
CONSTANT
GRADE21
MALE
−1ꢁ178
0ꢁ018
−0ꢁ121
0ꢁ202
0ꢁ152
0ꢁ035
−0ꢁ019
0ꢁ029
0ꢁ350
−0ꢁ040
4ꢁ351
0ꢁ038
−0ꢁ193
0ꢁ381
0ꢁ270
0ꢁ207
0ꢁ429
1ꢁ887
0ꢁ446∗∗∗
0ꢁ043
0ꢁ069∗
0ꢁ125
0ꢁ153
0ꢁ012∗∗∗
0ꢁ016
0ꢁ636
0ꢁ462
1ꢁ40
3ꢁ231
0ꢁ035
0ꢁ120
0ꢁ078∗∗∗
0ꢁ213
0ꢁ219
0ꢁ365
0ꢁ956∗∗
BLACK
HISPANIC
PAGRADE
MAGRADE
AGSHARE
MANUSHAR
COUNEMP
EMPGR
YPERCAP
POSQML
NOATTACH
NC14
SO14
WE14
DIFFEARN
∗Indicatessignificance in a two-tailed t-tes t at the P= 0ꢁ10 level.
∗∗Indicatessignificance in a two-tailed t-tes t at the P= 0ꢁ05 level.
∗∗∗Indicatessignificance in a two-tailed t-tes t at the P= 0ꢁ01 level.
Discussion and Policy Implications
Three salient findings summarize the results.
First, the migration of non-metropolitan
young adultsto other non-metropolitan
countiesisa viable and frequently
statistically significant influence on the cost
component of the migration decision, but the
population density variable has a negative
coefficient that just fails the test for statistical
significance at the p = 0ꢁ10 level.
employed alternative to “non-metropolitan
to metropolitan” area migration. Second,
the probability of migration from the non-
metropolitan county of origin increases in
schooling because of a higher rate of return
to schooling upon migration. In particu-
lar, other non-metropolitan countiesare an
attractive alternate destination for educated
non-metropolitan youth because of notably
high returnsto cshooling. Third, expected
gainsin initial earningscontinue to pro-
vide young adultswith tsrong incentivesto
migrate from their non-metropolitan coun-
tiesof origin. The propenisty to migrate is
also sensitive to several indicators of migra-
tion costs, specifically paternal education and
a measure of extended family in the area.
The reusltspoint to areasof opportunity
and areasof concern for non-metropolitan
countie.sNon-metropolitan areasasa whole
are able to retain a proportion of educated
young adultsby offering them high returnsto
schooling. There are, no doubt, other attrac-
tionsto remaining in non-metropolitan coun-
tiessuch asarea-specific amenity valuesand
lower costs of living, which explain why
an average-earningsgap between metropoli-
tan and non-metropolitan areas can persist
over time (see, among others, Topel). On
Based on earnings equation estimates, a
“typical” individual in the sample (male,
white, non-Hispanic, living in the south, and
possessing mean levels of other characteris-
tics) expects to earn $4.22 per hour upon
migration and $4.00 in hiscounty of origin
immediately after leaving school, an earnings
gap of 5.5%. Asshown in figure 1, the ratio of
simulated initial hourly earnings upon migra-
tion to simulated initial hourly earnings in
the county of residence at age 14 increases
in work experience, and the earningsgap is
26% after five yearsof experience. 4 Similarly,
the ratio of simulated initial hourly earnings
upon migration to simulated initial hourly
earningsin the county of origin increaessin
schooling. The initial earnings gap for migra-
tion relative to remaining in the county of
origin is14% with a high cshool education
and 28% with a college education.
Marginal effectscalculated from the probit
estimate of the structural equation (13) indi-
cate that a 10 percentage point increase in the
4 The linear rate of return on experience isnot extrapolated
beyond five years because estimates are based on individuals
at age 21 or the earliest year thereafter when observed out of
school.

338
May 2001
Amer. J. Agr. Econ.
1.40
Education
Experience
1.30
1.20
1.10
1.00
0
8
1
9
2
3
4
5
10
11
12
13
14
15
16
17
18
Row 1: Years of Experience
Row 2: Years of Schooling
Figure 1.Migrant to non-migrant earnings ratio by education and experience
the other hand, the results suggest that indi- origin. Thin labor marketsin many non-
vidual non-metropolitan countiesface com-
metropolitan areasfor individualswith at
petition for their educated youth not just least a college education motivate a plausible
from metropolitan countiesbut alos from
explanation. Employment opportunitiesfor
other non-metropolitan counties. The low highly educated young adultsare relatively
estimated rates of return to schooling and skill-specific. Few local employment opportu-
experience in the non-metropolitan counties nitiesfor highly educated young adultsmake
of origin are, in thiscontext, of particu-
lar concern. Low local returnsto cshooling
those preferring to stay in their counties of
origin more likely to take jobsthat poorly
create a disincentive to invest in education match their specific skills and, thus, pay a
and contribute to persistently lower levels of lower wage than could be attained upon
educational attainment in non-metropolitan migration to an area where their skills better
areas. This disincentive may be particularly match employer needs. Earnings of individu-
strong among those individuals with strong alswith low levelsof education are relatively
preferencesto remain in the county of origin. unaffected by skill mismatches because low-
However, a test of the influence of predicted skilled positions are relatively abundant in
migration propensity on GRADE21 provides non-metropolitan areas. By the same token,
no evidence that schooling isendogenousin
ratesof return to cshooling may be high
the reduced form equation (12). Thus, the dis- when migrating to other non-metropolitan
incentive to invest in education among those areasbecaues individualsmuts be compen-
wishing to remain in the county of origin does sated for migration costs to attain the better
not appear to be disproportionately large.
The findingsmotivate the quetison of
why young adultsare able to attain high
ratesof return to cshooling upon migra-
skill matches. Diamond and Simon provide
evidence of compensating wage differentials
in specialized labor markets due to high
geographic mobility costs between markets.
tion to another non-metropolitan county, but Similarly, returnsto cshooling upon migra-
not within their non-metropolitan county of tion to other non-metropolitan countiesmay

Mills and Hazarika
Migration of Young Adults from Non-Metropolitan Counties
339
References
plausibly be greater than those found in
metropolitan labor marketsto compenaste
skilled workers for the risk of moving to
labor marketswith few alternative employ-
ment opportunities.
Amemiya, T. Advanced Econometrics. Cambridge
MA: Harvard University Press, 1985.
Bartik,T.J.WhoBenefitsfromStateandLocalDevel-
opment Policies. Kalamazoo: W. E. Upjohn
Institute for Employment Research, 1991.
Broomhall, D.E., and T.G. Johnson. “Economic
Factorsthat Influence Educational Perfor-
mance in Rural Schools.” Amer. J. Agr. Econ.
76(August 1994):557–67.
Cheirse, P.C. “Inner AreasasSpatial Labour
Markets: A Critique of Inner Area Studies.”
Urban Stud. 16(February 1979):29–43.
Ciccone, A., and R.E. Hall. “Productivity and the
Density of Economic Activity.” Amer. Econ.
Rev. 86(March 1996):54–70.
Darity, W.A., and P.L. Mason. “Evidence on
Discrimination in Employment: Codes of
Color, Codesof Gender.” J. Econ. Perspect.
12(November 1998):63–90.
Deaton, B.J., L.C. Morgan, and K.R. Anschel. “The
Influence of Psychic Costs on Rural–Urban
Migration.” Amer. J. Agr. Econ. 64(May 1982):
177–187.
Diamond, C., and C. Simon, “Industrial Special-
ization and the Returnsto Labor.” J. Labor
Econ. 8(April 1990):175–201.
Fuguitt, G.V., and C.L. Beale. “Recent Trendsin
Non-metropolitan Migration: Towardsa New
Turnaround?” Growth and Change 27(Spring
1996):156–74.
Gisser, M., and A. Davila. “Do Farm Workers Earn
Less?” Amer. J. Agr. Econ. 80(November
1998):669–682.
Migration flowsof educated labor may
also influence the propensity for further
migration. The results show that employment
growth positively influences hourly earnings
in the local non-metropolitan labor mar-
ket. Thusout-migration isretarded in high
employment growth areasand accelerated
in low employment growth areasvia the
earningscomponent of the migration deci-
sion. Self-enforcing effects may arise from
the deisre of firmsto locate in areaswith
existing work forces well matched to their
labor requirements(Bartik). Asa result, firms
requiring skilled workers are more likely to
locate in areaswith exitsing concentrations
of skilled workers. Under such conditions,
two stable development paths are possible
(Murphy, Shleifer, and Vishny). Initial firm
concentration may be sufficient to attract
skilled workers to the area. Additional skilled
workerscreate a deisrable labor market for
the location of additional firms, which in turn
attractsadditional ksilled worker.sAlterna-
tively, the existing concentration of skilled
workersmay be inusfficient to attract new
firmsor even maintain current firmsin the
area. Worker migration then further reduces
the skill base of the local labor force, mak-
ing the area even less attractive for existing
firms.
Significant state and local development
resources are devoted to firm attraction and
retention programs. Most of the resources
are used to provide firms with informa-
tion on local opportunitiesand labor mar-
ket charactertiis,csaswell asincentives
Hoch, I. “Settlement Size, Real Income, and
the Rural Turnaround.” Amer. J. Agr. Econ.
61(December 1979):953–59.
Isserman, A.W. “State Economic Policy and Prac-
tice in the United States: A Survey Article.”
Int. Regional Sci. Rev. 16(1994):49–100.
to relocate (Isserman). Successful business Kim, S. “Diversity in Urban Labor Markets and
attraction and retention effortsincreeas
labor demand and create employment. How-
ever, non-metropolitan areasexperiencing
a co-dependent rise in out-migration of
ksilled workersand firmsmay be rela-
tively unresponsive to traditional state and
local economic development efforts. Coordi-
nated effortsto develop local labor force
capacity and to attract firmscompatible
with that capacity may be necessary, if not
sufficient, to retain skilled young adults in
such circumstances.
Agglomeration Economics.” Pap. Rgnl. Sci.
Assoc. 62(1987):57–70.
Lazear, E. “Age, Experience, and Wage Growth.”
Amer. Econ. Rev. 66(September 1976):548–58.
Lee, L.F. “Generalized Econometric Modelswith
Selectivity.” Econometrica 51(March 1983):
507–12.
Mills, E.S. “The Location of Economic Activity in
Rural and Nonmetropolitan United States.”
The Changing American Countryside. Emory
N. Castle, ed. Lawrence KS: University of
Kansas Press, 1995.
Murphy, K., A. Shleifer, and R. Vishny. “Industri-
alization and the Big Push.” J. Polit. Econ.
97(October 1989):1003–1026.
[Received April 1999;
accepted August 2000.]

340
May 2001
Amer. J. Agr. Econ.
Robinson, C., and N. Tomes. “Self-selection and
Interprovincial Migration in Canada.” Can. J.
Econ. 15(August 1982):474–502.
and Labor Market Areas. U.S. Department of
Agriculture, E.R.S. Staff Paper 9614, 1996.
Topel, R.H. “Local Labor Markets.” J. Polit. Econ.
94(June 1986):111–43.
Sjaastad, L.A. “The Costs and Returns of Human
Migration.” J. Polit. Econ. 70(October 1962): Willis, R.J., and S. Rosen. “Education and Self-
80–93.
Selection.” J. Polit. Econ. 87(October 1979):
Tolbert, C.M., and M. Sizer. U.S. Commuting Zones
S7–S36.