http://www.jstor.org
Power Couples: Changes in the Locational Choice of the College Educated, 1940-1990
Author(s): Dora L. Costa and Matthew E. Kahn
Source: The Quarterly Journal of Economics, Vol. 115, No. 4, (Nov., 2000), pp. 1287-1315
Published by: The MIT Press
Stable URL: http://www.jstor.org/stable/2586925
Accessed: 16/08/2008 00:25
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=mitpress.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.
POWER COUPLES: CHANGES IN THE LOCATIONAL
CHOICE OF THE COLLEGE EDUCATED, 1940-1990*
DORA L. COSTA AND MATTHEW E. KAHN
College educated couples are increasingly located in large metropolitan areas.
These areas were home to 32 percent of all college educated couples in 1940, 39
percent in 1970, and 50 percent in 1990. We investigate whether this trend can be
explained by increasing urbanization of the college educated or the growth of dual
career households and the resulting severity of the coloration problem. We argue
that the latter explanation is the primary one. Smaller cities may therefore
experience reduced inflows of human capital relative to the past and thus become
poorer.
I. INTRODUCTION
Couples in which both husband and wife have at least a
college education are increasingly, and disproportionately, located
in large metropolitan areas. In 1940 32 percent of all college
educated couples were located in metropolitan areas of at least 2
million. By 1970 this figure was 39 percent and by 1990 50
percent. In contrast, there has been little change in the proportion
of couples in which neither spouse has a college education in large
cities. In 1940 27 percent of these couples were located in large
metropolitan areas, and in 1970 30 percent were. By 1990 only 34
percent were.1
Two main factors may account for the disproportionate increase
in college educated couples' propensity to be in a large city.
The college educated, regardless of marital status, may be becoming
more urbanized, perhaps because the returns to education in
large cities are rising relative to those in small cities or because
such urban amenities as access to cultural activities are normal
goods.2 In addition, more of the college educated may now be
* We have benefited from the comments of Edward Glaeser, Claudia Goldin,
Lawrence Katz, Michael Kremer, Steven Levitt, Erzo Luttmer, Sendhil Mullainathan,
Jorn-Steffen Pischke, James Poterba, Peter Temin, and two anonymous
referees and of workshop participants at the University of Chicago, Boston
College, Princeton University, Stanford University, and the April 1999 NBER
Labor Studies meeting. Dora Costa gratefully acknowledges the support of NIH
grant AG12658.
1. Estimated from the integrated public use census samples [Ruggles and
Sobeck 1997].
2. Of course, the cost of living in a large metropolitan area is higher rents and,
for those living in central cities, such urban disamenities as crime and pollution.
However, since 1970 these disamenity costs have been falling [Glaeser 1998; Kahn
19971.
e 2000 by the President and Fellows of Harvard College and the Massachusetts Institute of
Technology.
The Quarterly Journal of Economnics,November 2000
1287
1288 QUARTERLYJOURNALOF ECONOMICS
meeting and marrying in large cities because fewer of their
marriages are now formed in high school or college and because
more of the single college educated are moving to large cities. If
they remain in large cities, the proportion of college educated
couples in large cities will increase.
Second, as more households become dual career households,
more of them face a colocation problem. All dual career households
are more likely to be joint decision makers, and they face the
difficulty of finding two jobs commensurate with the skills of each
spouse within a reasonable commute from home. In 1940, among
couples in which the husband was 25 to 39 and the wife 23 to 37
and in which both had at least a four-year college education, only
20 percent of wives worked. By 1990 73 percent of them were in
the labor force.3 The colocation problem is likely to be particularly
severe for the college educated. As documented by Goldin [1997],
by 1990 college educated women aspired to careers, not to the jobs
held by their counterparts in the 1970s. The resulting colocation
problem should lead to the greater concentration of college
educated couples in large metropolitan areas because such areas
offer many more potential job matches. Furthermore, if the
returns to education are greater in larger cities because large
cities permit specialization, the financial sacrifice of living in a
small relative to a large metropolitan area would be greater for a
dual career family than for other household types.
This paper documents trends in locational choice since 1940
between large, midsize, and small cities by household type. We
examine whether the long-run trends are most consistent with a
worsening of the colocation problem or with the growing urbanization
of the college educated. We predict that if colocation is a
plausible explanation then dual career households will be disproportionately
represented in large cities relative to other couple
types. In contrast, if urbanization were the primary explanation,
then we should not observe a disproportionate increase in college
educated couples in large metropolitan areas relative to college
educated singles. We find that colocation is the most likely
explanation for the observed trend in locational choice.
Our findings have implications for city growth and for family
economics. As skilled professionals are increasingly bundled with
an equally skilled spouse, their demand for large cities will
3. Estimated from the integrated public use census samples [Ruggles and
Sobeck 19971.
POWERCOUPLES 1289
increase. The presence of large numbers of highly skilled workers
within a concentrated geographic area may in turn provide
positive growth externalities.4 Smaller cities, particularly those
located in low amenity areas, may experience reduced inflows of
human capital relative to the past and therefore become poorer.
We present some suggestive evidence on the ability of firms in
small cities to attract highly skilled workers by examining how
the relationship between city size and the quality of university
graduate programs has changed since 1970. Colocation increases
inequality not only across city size but also among households.
Because large cities facilitate the matching of the highly educated
and because they enable professional couples to solve their joint
location problem, large cities magnify household income inequality.
Large cities may also preserve the marriages of dual career
households because living in large metropolitan areas with diversified
labor markets reduces the degree to which both husband
and wife must compromise their individual gains from marriage
[Mincer 1978].
II. TRENDS
We view all households as one of five types: "power"couples in
which both spouses have a college education, "part-power"couples
in which only one spouse has a college education, "low-power"
couples in which neither spouse has a college education, and
single households of the college educated and the noncollege
educated. We define the college educated as those who have
completed four or more years of college in 1940, 1970, and 1980
and those who hold a bachelor's or graduate degree in 1990 (see
the Appendix for details). We distinguish between college and
noncollege rather than postcollege (more than four years of college
completed in 1940, 1970, and 1980 and a graduate degree in
1990), college, and noncollege because the postcollege represent a
small proportion of the population and because college educated
women experienced greater career growth.5 We define two individu-
4. This insight has been incorporated in theoretical models [Lucas 1988;
Acemoglu 1996] and is supported by empirical investigations of the relationship
between wages and average human capital [Heckman, Layne-Farrar, and Todd
1996; Rauch 1993] and local growth and average human capital within a city
[Simon 1998; Glaeser, Shleifer, and Scheinkman 19951.
5. The postcollege educated were more likely to have careers in 1970 whereas
the college educated were more likely to have "jobs."In 1970 and 1990 couples in
which both husband and wife had a postcollege education were more likely to be in
a larger rather than a smaller metropolitan area. The growth in power couple
1290 QUARTERLYJOURNALOF ECONOMICS
TABLE I
PERCENTAGEOF MARRIAGESBY COUPLE TYPE
1940 1960 1970 1980 1990
Low-power 90.3 82.5 76.8 67.6 66.2
Part-power 7.3 12.3 14.6 18.4 18.5
Power 2.4 5.2 8.6 14.0 15.3
A power couple is defined as one in which both husband and wife are college graduates, a part-power
couple as one in which only one spouse is a college graduate, and a low-power couple as one in which neither
spouse is a college graduate. All numbers are estimated from the integrated public use census samples
[Ruggles and Sobek 19971 and are for households in which the husband was age 25 to 39 and the wife 23 to 37.
als to be a couple if they are married to each other and are both in
the same household. Singles can be either never married, divorced,
or widowed. Both couples and singles may be in multifamily
households. We examine couples in which the husband was 25
to 39 years of age and the wife 23 to 37. This age restriction allows
us to examine couples in the early stages of their careers and also
allows us to create a comparable group of singles. We impose the
same age restrictions on singles.
I. 1. Trends: Power Couple Formation
The proportion of married couples in which both husband and
wife have at least a college education has increased from 2 percent
in 1940 to 15 percent in 1990 (see Table I). The percentage
increase in the proportion of couples in which only one spouse has
a college education has been smaller, rising from 7 percent in 1940
to 18 percent in 1990.6 These increases in the relative proportion
of power couples arose largely from greater college attendance
rates, which in turn were spurred both by the growth of public
universities and of high schools and by the rising economic
returns to college.7
Rising wives' labor force participation rates have turned
couples into true dual career households and increased the
fraction of couples with a colocation problem. This is particularly
true for power couples because increases in wives' labor force
concentration in large metropolitan areas was slightly more pronounced among
the college educated compared with the postcollege educated.
6. Among part-power couples in 1940, the husband was the college educated
spouse in 76 percent of all cases. In 1990 he was the college educated spouse in 63
percent of all cases.
7. Although prime-aged women in 1970 gained little direct economic return
from their degrees, their indirect gains were considerable because for them college
was a marriage market [Goldin 19921.
POWERCOUPLES 1291
TABLEII
EMPLOYMENT AND FERTILITY TRENDS BY EDUCATION OF COUPLE
1940 1960 1970 1980 1990
Wife works (%)
Low-power 18.1 27.0 36.6 52.3 64.2
Part-power 18.3 22.7 34.5 57.4 70.4
Power 20.1 29.6 43.4 64.8 73.3
Have child (%)
Low-power 73.1 89.9 90.2 85.9 83.4
Part-power 60.9 86.8 82.7 74.6 72.6
Power 58.0 81.3 72.9 64.4 62.7
Wife works and works full-time (%)
Low-power 73.6 68.0 64.8 66.8 68.5
Part-power 73.4 65.7 60.5 65.6 68.7
Power 68.1 62.7 58.1 68.1 70.2
Wife works and in traditionally
female job (%)
Power 71.5 71.4 73.2 59.4 42.7
Afull-time job is defined as 35 hours or more per week. Atraditionally female occupation is defined as one
in which women were overrepresented relative to men in 1970; that is, one in which more than 50 percent of all
employees age 18 to 64 were women in 1970. All couples are restricted to those in which the husband was 25 to
39 years of age and the wife 23 to 37. All numbers are estimated from the integrated public use census samples
[Ruggles and Sobek 19971. A power couple is defined as one in which both husband and wife are college
graduates, a part-power couple as one in which only one spouse is a college graduate, and a low-power couple
as one in which neither spouse is a college graduate.
participation rates have been larger among power than among
low-power couples (see Table II).8 The labor force participation
rate of power couple wives rose from 20 to 73 percent between
1940 and 1990, whereas the increase for low-power wives was
from 18 to 64 percent. Since 1970 the increase in full-time work
was greatest among power wives.
Other fundamental economic changes have also increased the
costs to a household of picking a city size in which the wife earns
relatively little. The proportion of working power couple wives in
such traditional female occupations as schoolteacher, nurse, librarian,
or social worker fell from 72 to 43 percent between 1940 and
1990. The percentage with at least one child rose to 81 percent in
1960 from 58 percent in 1940, but by 1980 had fallen to 64 percent.
Prime-aged college educated women in 1940 generally became
schoolteachers upon graduation, leaving the labor force because of
8. Some of the growth in wives' labor force participation rates may arise from
the increased propensity of women who aspire to careers to marry [Goldin 1997].
1292 QUARTERLYJOURNALOF ECONOMICS
marriage bars. Goldin [1997] describes their experience as "first
jobs then family." In contrast the experience of their 1970 counterparts
was "first family then jobs." The majority majored in such
fields as education and nursing where few men got degrees and
upon graduation had been tracked into traditionally female
sectors, regardless of their majors. They left the labor force when
their first child was born and only reentered when all children
were in school. By 1990, and to a lesser extent by 1980, college
educated women aspired to "career then family" or "career and
family" [Goldin 1997]. Their college majors were similar to men's,
and in terms of labor supply parameters they began to resemble
men as well, with small wage and income elasticities [Goldin
1990, pp 119-158].
11.2. Trends: Locational Choice
We study trends in household locational choice conditional on
marital status and on the education of both spouses. That is, for
every year, we estimate
(1) Prob (lives in big city Ihousehold type).
Our data come from the 1940 and 1970-1990 integrated public
use census samples. We classify the suburbs of central cities as
part of the labor market of the central city and create three city
size categories: large metropolitan areas (those with populations
of at least 2 million), midsize metropolitan areas (those with
populations of between 2 million and 250,000), and small and
nonmetropolitan areas (metropolitan areas with populations of
less than 250,000 and nonmetropolitan areas). Although we
document trends from 1940 to 1990, our discussion will emphasize
the 1970 to 1990 trends because this was the period of growth in
wives' careers. An additional advantage of focusing on this period
is that the definition of a metropolitan area was most comparable
from 1970 to 1990 and because there is likely to be less measurement
error in our classifications of households by educational
levels.9 Details about variable construction and comparability are
provided in the Appendix.
Table III illustrates trends in locational choice between large
and midsize metropolitan areas and small localities. Note that the
9. College graduation rates are overstated in the 1940 census in part because
individuals who went to the preparatory department within a college were
enumerated as having gone to college. We thank Claudia Goldin for pointing this
out to us.
POWERCOUPLES 1293
TABLE III
PROBABILITYOF LOCATIONALCHOICEBY HOUSEHOLDTYPE
1940 1970 1980 1990
Conditional on power couple
Large metropolitan area 0.321 0.391 0.414 0.495
Midsize metropolitan area 0.254 0.313 0.325 0.295
Small and nonmetropolitan area 0.426 0.296 0.261 0.210
Conditional on part-power couple
Large metropolitan area 0.319 0.362 0.371 0.421
Midsize metropolitan area 0.268 0.326 0.334 0.308
Small and nonmetropolitan area 0.413 0.312 0.295 0.271
Conditional on low-power couple
Large metropolitan area 0.266 0.301 0.308 0.339
Midsize metropolitan area 0.240 0.299 0.312 0.292
Small and nonmetropolitan area 0.494 0.399 0.380 0.369
Conditional on single, power man
Large metropolitan area 0.383 0.523 0.512 0.569
Midsize metropolitan area 0.258 0.291 0.295 0.266
Small and nonmetropolitan area 0.358 0.186 0.193 0.165
Conditional on single, power woman
Large metropolitan area 0.286 0.507 0.499 0.555
Midsize metropolitan area 0.223 0.309 0.308 0.281
Small and nonmetropolitan area 0.491 0.184 0.193 0.164
Conditional on single, low-power man
Large metropolitan area 0.299 0.442 0.415 0.441
Midsize metropolitan area 0.225 0.297 0.305 0.278
Small and nonmetropolitan area 0.476 0.260 0.280 0.281
Conditional on single, low-power woman
Large metropolitan area 0.307 0.455 0.430 0.444
Midsize metropolitan area 0.245 0.305 0.313 0.297
Small and nonmetropolitan area 0.448 0.240 0.257 0.260
A power couple is defined as one in which both husband and wife are college graduates, a part-power
couple as one in which only one spouse is a college graduate, and a low-power couple as one in which neither
spouse is a college graduate. couples were restricted to those in which the husband was 25 to 39 years of age
and the wife 23 to 37. All singles were also in that age range. A power single is a college graduate, while a
low-power single is not. Large metropolitan areas are those in which the population was at least 2 million,
midsize metropolitan areas as those between 2 million and 250,000, and small as those less than 250,000.
Within each year for each household type, the probabilities should sum to one. Estimated from the integrated
public use census samples [Ruggles and Sobeck 1997].
probability of a power couple being in a large city rose by 0.174
between 1940 and 1990 and by 0.104 between 1970 and 1990. In
contrast, the increase in these probabilities for part-power couples
was only 0.102 and 0.059, respectively. For college educated
singles the probability of being in a large city rose sharply
1294 QUARTERLYJOURNALOF ECONOMICS
between 1940 and 1970, but only changed by 0.040 for men and
0.048 for women between 1970 and 1990.
III. HYPOTHESESANDEMPIRICALMETHODOLOGY
Are power couples urbanizing because they are college educated
or because they have a colocation problem? We begin by
considering the benefits of being in a large city for different couple
types within a single year and whether changes in these factors
might account for the trend.
Consider first the advantages of urbanization by couple type.
If large metropolitan areas offer higher returns to education and if
urban amenities are normal goods, then large metropolitan areas
will appeal to the college educated, regardless of marital status. If
urban areas offer such singles' amenities as marriage markets,
then singles will be more concentrated in large metropolitan areas
than couples of the same educational level. The concentration of
power couples in large cities will increase if the returns to city size
for the college educated are rising, if urban amenities continue to
be normal goods, or if more college educated singles are moving to
large cities because cities are becoming better marriage markets
and are staying there upon marriage. We will run wage regressions
to test whether the returns to city size have been rising for
the college educated. We find that most of the increasing urbanization
of the college educated is explained by increasing returns to
city size and not because urban amenities are normal goods. We
cannot directly test whether large cities are attracting more
college educated singles because they are becoming better marriage
markets for the college educated. But, we can provide some
evidence on the marriage propensities of the college educated
relative to the noncollege educated.
Second, consider the colocation problem. Large metropolitan
areas offer both spouses the opportunity to pursue careers. The
diversified labor markets of larger cities also insure against
health and unemployment shocks to one spouse. All couples face a
colocation problem. Because the college educated are more versatile
(and can take jobs for which they are overqualified), the
colocation problem may be less severe for the college educated.
But, because the college educated may be more career oriented or
may have more specialized skills, the colocation problem may be
POWERCOUPLES 1295
severer for the college educated.10 Large cities offer more potential
job matches, so the probability of drawing a good initial match is
higher. The probability of drawing good subsequent matches is
also higher, and this increased job mobility will lead to greater
lifetime wage growth [Topel and Ward 1992]. Furthermore, if the
initial match was a poor one, then the probability of drawing a
good match on the second try will be higher than in a smaller city.
A spouse who knows that the other spouse has these options in a
large city can make firm-specific career investments. The long-run
financial sacrifice to being in a large city is therefore likely to be
smaller. A worsening colocation problem should therefore increase
the proportion of couples, particularly power couples, in large
cities, but not affect the proportion of singles.
We test whether the increase in the proportion of power
couples in large cities is determined by colocation by comparing
power couples with other couple types and to singles. Our
comparisons with singles are based upon "coincidental" couples.
We refer to a coincidental couple as two individuals (one male and
one female) coincidentally living in a large metropolitan area.
Trends in coincidental couple concentration allow us to analyze
trends in the urbanization of the college educated independent of
colocation. Suppose that there are 100 single power men and
women and that 40 of the men are in large cities and 60 are in
small cities and that 60 of the women are in large cities and 40 are
in small cities. At most 80 "marriages" could form-40 in large
cities and 40 in small cities. The probability of a coincidental
couple being in a large city is therefore 0.5. We therefore take our
estimates of the probability of a single power man living in city
size s(pM'P) and our estimates of the probability of a single power
woman living in city size s(p'P) and take the minimum of these
probabilities (min (pSP pSP)) We then estimate the probability
that a coincidental couple will be living in a given city size (min
(pMsP,pFP)'yNs mm (pM,P FP)).11
Table IV provides a schematic illustration of our strategy to
identify the colocation problem of power and low-power couples
and the differential colocation problem of power relative to
10. Baumgardner [19881 documents the greater specialization among physicians
found in large cities. See Kim [1989] for a model of labor specialization and
the size of the labor market.
11. Note that this method assumes that the numbers of single power men and
single power women are the same. Although this is a reasonable assumption in
most years, in 1940 there were more single unmarried power women because of the
low marriage propensities of college educated women.
1296 QUARTERLYJOURNALOF ECONOMICS
TABLE IV
CHANGE IN BENEFITS OF LIVING IN A LARGECITY BY COUPLE TYPE
Couple type Benefits of living in a large city
1 Power couple coloration power
urbanization power (urban amenities,
education)
2 Coincidental power couple urbanization power
singles' amenities (e.g., marriage markets)
3 Double difference (1-2) coloration power, singles' amenities
4 Low-power couple coloration low-power
urbanization low power (urban amenities)
5 Coincidental low-power couple urbanization low power
singles' amenities
6 Double difference (4-5) coloration low-power, singles' amenities
7 Triple difference (3-6) colocation power relative to low-power
A power couple is defined as one in which both husband and wife are college graduates, and a low-power
couple as one in which neither spouse is a college graduate. A coincidental power couple consists of two single
college educated individuals (one male and one female) coincidentally living in the same city size. A
coincidental low-power couple consists of two single noncollege educated individuals (one male and one
female) coincidentally living in the same city size.
low-power couples. The increase in power couple concentration in
large metropolitan areas could be driven by a worsening colocation
problem and the growing urbanization of the college educated.
The increase in the concentration of coincidental couples
could be driven by the urbanization of power couples or the
increased value of singles' amenities (including changes in the
value of large urban areas as marriage markets). If the value of
singles' amenities has not changed and if urban amenities are of
the same value to singles as they are to married couples, then the
double difference (the increase in power couple concentration
minus the increase in coincidental couples concentration) measures
the differential effect of coloration. Note that any differences
in the valuation of urban amenities of married couples and singles
will not be differenced out. Therefore, an alternative explanation
for the increase in power couple relative to power single concentration
in large metropolitan areas is the increased amenity value of
metropolitan areas to couples. Now consider the case of low-power
couples. The increased concentration of low-power couples in
large metropolitan areas could arise either from growing urbanization
or from colocation. The growing concentration of low-power
singles could arise either from growing urbanization or from the
growing value of such singles' amenities as a marriage market. If
POWERCOUPLES 1297
the value of singles' amenities has not changed and if urban
amenities are of the same value to singles as they are to married
couples, then the double difference (the increase in low-power
couple concentration minus the increase in coincidental lowpower
couple concentration) yields the effect of colocation for
low-power couples. Again, differences in the valuation of urban
amenities to married couples and to singles will not be differenced
out. The triple difference yields the differential effect of coloration
for power relative to low-power couples, assuming either that
there are no differences in the valuation of urban amenities
between married couples and singles or, if there are, that these
differences are the same for the college educated as for the
noncollege educated.
The triple difference estimate of the differential effect of
colocation for power relative to low-power couples can be further
refined. Because labor force participation rates of single women
are higher than those of married women and because only couples
in which the wife works have a colocation problem, we also
compare power and low-power couples in which the wife works
with coincidental couples in which the woman works and calculate
the triple difference. Larger double and triple differences for
couples in which the wife works than for all couples would suggest
that colocation, not differences in amenity values, determines the
concentration of power couples in large metropolitan areas.
IV. BASIC RESULTS
Testing the theories requires us to examine differential
trends in locational choice by couple type. We present these trends
in this section, standardized on race and age. That is, we estimate
(2) Prob (lives in city size s and wife works household type)
(3) Prob (lives in city size s and wife does not work household type)
for couples and
(4) Prob (lives in city size s and works household type)
(5) Prob (lives in city size s and does not work household type)
for singles. For couples these probabilities can be thought of as the
outcome of a joint decision between city size and wife's labor force
participation. Because of data limitations we cannot unravel
whether the low labor force participation rate of wives in smaller
1298 QUARTERLYJOURNALOFECONOMICS
cities arises from the choice of these areas by wives with little
attachment to the labor force or whether it arises from wife's wage
effects of being in a small city.
We estimate the probabilities in equations (2) through (5) by
means of a multinomial logit choice model. We use information on
what city size category a couple chooses to live in and whether the
wife works to create six groups, one each for wife's labor force
participation status and the city size category. We then estimate a
multinomial logit of the choice of wife's labor force participation
and city size as a function of the husband's age and race and the
educational attainment of the husband and wife (dummies for less
than high school graduate, high school graduate, college graduate,
and postcollege graduate with less than high school graduate
as the omitted dummy). Thus, defining Psww as the probability of
being in one of three city sizes s and ww as an indicator equal to
one if the wife works, we estimate
PSW
(6) log (sww) X
for s = 1 to 2 and ww = O 1 and s = 0 and ww = 0. We calculate
robust standard errors clustering on metropolitan areas (or on the
state in the case of nonmetropolitan areas). Finally, we predict the
choice of location and wife's labor force participation, PSWW,for a
white household in which the husband is 35 years of age and the
wife 33 conditional on being a power, part-power, or low-power
couple. For single individuals we estimate similar multinomial
logit specifications (separately for men and women) except that
labor force status is own labor force status and we control only for
own characteristics. We then predict locational choice for single,
white men age 35 and single, white women age 33 conditional on
being a power or a low-power individual and estimate locational
choice for coincidental couples.12 Recall that a coincidental couple
consists of two single individuals (one male and one female)
coincidentally living in a large metropolitan area.
Table V shows the predicted probabilities, conditional on
being a power, part-power, or low-power couple, of locational
choice across city sizes and the wife's labor force participation
status for a white couple in which the husband was 35 years old
and the wife 33. Power couples are leaving nonmetropolitan areas
12. The base group is small or nonmetropolitan area and wife out of the labor
force. Our results are robust to the choice of base group.
POWERCOUPLES 1299
TABLE V
PREDICTEDPROBABILITIESOF LOCATIONALCHOICEAND WIFE'S LABORFORCE
PARTICIPATION(LFP) STATUS CONDITIONALON HOUSEHOLDTYPE
1940 1970 1980 1990
Conditional on power
Large metropolitan area, LFP = 1 0.083 0.146 0.255 0.348
Large metropolitan area, LFP = 0 0.264 0.240 0.157 0.141
Midsize metropolitan area, LFP= 1 0.045 0.136 0.206 0.218
Midsize metropolitan area, LFP = 0 0.217 0.183 0.121 0.080
Small and nonmetropolitan area, LFP= 1 0.067 0.142 0.179 0.165
Small and nonmetropolitan area, LFP= 0 0.325 0.152 0.082 0.049
Conditional on part-power
Large metropolitan area, LFP = 1 0.058 0.092 0.187 0.262
Large metropolitan area, LFP = 0 0.276 0.258 0.179 0.142
Midsize metropolitan area, LFP = 1 0.043 0.101 0.176 0.210
Midsize metropolitan area, LFP = 0 0.226 0.222 0.152 0.097
Small and nonmetropolitan area, LFP= 1 0.064 0.116 0.174 0.205
Small and nonmetropolitan area, LFP = 0 0.333 0.210 0.132 0.084
Conditional on low-power
Large metropolitan area, LFP = 1 0.048 0.083 0.145 0.200
Large metropolitan area, LFP = 0 0.246 0.205 0.157 0.132
Midsize metropolitan area, LFP = 1 0.044 0.106 0.159 0.191
Midsize metropolitan area, LFP = 0 0.200 0.196 0.152 0.100
Small and nonmetropolitan area, LFP= 1 0.070 0.153 0.196 0.245
Small and nonmetropolitan area, LFP = 0 0.392 0.258 0.190 0.133
All predictions are from a multinomial logit model in which the outcome variables were city size and labor
force participation rate. The independent variables were husband's age, age squared, and race; wife's age and
age squared, and dummy variables for educational levels (less than high school, high school, college, college
plus) of the husband and of the wife. The predictions are for a white couple in which the husband was 35 years
old and the wife 33. Robust standard errors ranged from 0.001 to 0.006. Conditional on couple type, the rows
should sum to one. Estimated from the integrated public use census samples.
and moving to the largest metropolitan areas. In 1940 39 percent
were in nonmetropolitan areas whereas by 1970 the figure was 29
percent and by 1990 21 percent. In 1940 35 percent were in the
largest metropolitan areas, and in 1970 39 percent were; whereas
in 1990 49 percent were. Power wives' labor force participation
rates are lower in larger metropolitan areas in 1970, 1980, and
1990, suggesting that higher returns to city size for college
educated husbands are inducing them to leave the labor force.
Power couples are increasingly concentrating in large metropolitan
areas relative to part-power or low-power couples. Between
1970 and 1990 the predicted proportion of low-power
couples located in nonmetropolitan areas fell only slightly from 41
to 38 percent, and their share in the largest metropolitan areas
1300 QUARTERLYJOURNALOFECONOMICS
increased from 29 to 33 percent. The predicted percentage of
part-power couples in the largest metropolitan areas rose from 35
to 40 percent.
Table VI presents predictions of locational choice from multinomial
logit models for unmarried men and women conditional on
their education. Note that between 1940 and 1970 the probability
of singles being in a large metropolitan area increased substantially,
rising from 29 to 49 percent among power women. Between
TABLE VI
PREDICTED PROBABILITIES OF LOCATIONAL CHOICE, UNMARRIED MEN AND WOMEN,
CONDITIONAL ON EDUCATION
1940 1970 1980 1990
Single, power man
Large metropolitan area 0.394 0.526 0.518 0.542
Midsize metropolitan area 0.267 0.294 0.291 0.273
Small and nonmetropolitan area 0.339 0.180 0.191 0.185
Single, power woman
Large metropolitan area 0.289 0.489 0.501 0.534
Midsize metropolitan area 0.230 0.319 0.305 0.291
Small and nonmetropolitan area 0.482 0.192 0.193 0.175
Coincidental power couple
Large metropolitan area 0.337 0.508 0.510 0.544
Midsize metropolitan area 0.268 0.305 0.296 0.278
Small and nonmetropolitan area 0.395 0.187 0.194 0.178
Single, low-power man
Large metropolitan area 0.313 0.432 0.406 0.411
Midsize metropolitan area 0.232 0.304 0.305 0.282
Small and nonmetropolitan area 0.454 0.264 0.290 0.308
Single, low-power woman
Large metropolitan area 0.323 0.420 0.415 0.413
Midsize metropolitan area 0.250 0.316 0.317 0.304
Small and nonmetropolitan area 0.428 0.264 0.269 0.283
Coincidental low-power couple
Large metropolitan area 0.322 0.425 0.414 0.421
Midsize metropolitan area 0.238 0.308 0.311 0.289
Small and nonmetropolitan area 0.440 0.267 0.274 0.290
All predictions are derived from a multinomial logit model and are for white 35 year old men and white 33
year old women. Within each year for each group the predicted probabilities for singles and coincidental
couples should sum to one. The independent variables used in the multinomial logit were age, age squared,
race, and educational level. The outcome variables were city size and labor force participation. With the
exception of 1940, relatively few individuals were out of the labor force. The results that are presented are the
predicted probabilities summed over city size. Robust standard errors ranged from 0.002 to 0.006. Estimated
from the integrated public use census samples [Ruggles and Sobeck 19971.
POWERCOUPLES 1301
1970 and 1990 changes were much more modest. Among power
men the probability of being in a large city rose from 53 to 54
percent. The increase among coincidental couples was from 51 to
54 percent. The probability of low-power men and women being in
a large city decreased somewhat, with a decline from 43 to 42
percent among coincidental couples.
We summarize standardized trends in locational choice by
couple type, including coincidental couples (see Table VII). Between
1970 and 1990 the probability of power couples being in a
TABLE VII
TRENDS IN PROPENSITY TO LIvE IN GIVEN CITYSIZE, 1970-1990,
BY COUPLE TYPE (BASED UPON PREDICTED PROBABILITIES)
City size
Large Midsize Small
Differences, 1990-1970
Power couples (AP) 0.103 -0.022 -0.081
(0.007) (0.006) (0.006)
Part-power couples (APP) 0.054 -0.016 -0.038
(0.006) (0.006) (0.006)
Low-power couples (ALP) 0.043 -0.011 -0.033
(0.003) (0.002) (0.004)
Coincidental power couple (ACP) 0.036 -0.027 -0.009
(0.005) (0.010) (0.005)
Coincidental low-power couple (ACLP) -0.004 -0.019 0.023
(0.005) (0.004) (0.004)
Power, wife works (APW) 0.202 0.082 0.023
(0.004) (0.004) (0.005)
Part-power, wife works (APPW) 0.170 0.109 0.089
(0.004) (0.004) (0.004)
Low-power, wife works (ALPW) 0.117 0.085 0.092
(0.002) (0.002) (0.003)
Power, wife does not work (AP'W) -0.099 -0.103 -0.103
(0.005) (0.005) (0.006)
Part-power, wife does not work (APP-W) -0.116 -0.125 -0.126
(0.004) (0.004) (0.004)
Low-power wife does not work (ALP-W) -0.073 -0.096 -0.125
(0.003) (0.003) (0.004)
Coincidental power, woman works (ACPW) 0.037 -0.031 -0.006
(0.007) (0.007) (0.006)
Coincidental low-power, woman works (ACLPW) -0.011 -0.029 0.040
(0.004) (0.005) (0.004)
Differences are in probability units. Probabilities are calculated from Tables V and VI, except for
coincidental couples in which the woman works. These were calculated by using the multinomial logit
predictions for working women only.
1302 QUARTERLYJOURNALOFECONOMICS
large metropolitan area rose by 0.103, whereas that of part-power,
low-power, and coincidental power couples being in a large city
rose by 0.054, 0.043, and 0.036, respectively. The increase in
coincidental couple concentration suggests that at most 35 percent
of the increase in power couple concentration in large cities is
attributable to the growing urbanization of the college educated.
Among power couples in which the wife works, the probability of
being in a large city rose by 0.202; whereas among power couples
in which the wife does not work, this probability fell by 0.099.
Among coincidental power couples the probability of being in a
large city rose by only 0.036, and among low-power coincidental
couples it fell by -0.004. Urbanization of the noncollege educated
therefore cannot explain the rise in low-power concentration in
large cities.
The differential trend for power couples' probability of being
in a large metropolitan area was 0.049 relative to part-power
couples, 0.060 relative to low-power couples, and 0.067 relative to
coincidental couples (see Table VIII). The latter figure implies
that at least 65 percent of the 0.103 rise in power couple
concentration in large cities is attributable to colocation. The
differential trend of 0.047 in low-power relative to coincidental
low-power couple concentration implies that all of the 0.043
increase in low-power couple concentration in large cities is
attributable to coloration. The triple difference shows that 0.020
(19 percent) of the increase in power couple concentration is
accounted for by the unique coloration problems of the college
educated relative to the noncollege educated.
Recall that we can further refine our triple difference estimate
by restricting ourselves to couples (including coincidental
couples) in which the woman works. We would only expect a
colocation problem among working couples. Among couples in
which the woman works, the differential trend in power couples'
probability of being in a large city was 0.032 relative to part-power
couples, 0.085 relative to low-power couples, and 0.165 relative to
coincidental couples. The comparison with coincidental couples
implies that 87 percent of the rise in working power couple
concentration can be explained by colocation. The differential
trend in low-power couples' probability of being in a large city
relative to coincidental couples was 0.128, conditional on the
woman being in the labor force. This figure implies that all of the
increase in working low-power couple concentration in large cities
can be explained by coloration. The triple difference estimate of
POWERCOUPLES 1303
TABLE VIII
DIFFERENTIALTRENDS IN PROPENSITYTO LIvE IN GIVEN CITY SIZE, 1970-1990,
BY COUPLE TYPE (BASED UPON PREDICTEDPROBABILITIES)
City size
Large Midsize Small
Double differences, 1990-1970
AP- APP 0.049 -0.006 -0.043
(0.009) (0.008) (0.008)
AP - ALP 0.060 -0.011 -0.048
(0.008) (0.006) (0.007)
AP- ACP 0.067 0.005 -0.072
(0.009) (0.007) (0.008)
ALP AACLP 0.047 0.008 -0.056
(0.006) (0.004) (0.006)
AP,W- ACPW 0.165 0.113 0.029
(0.008) (0.008) (.007)
ALPW - ACLP,W 0.128 0.114 0.052
(0.004) (0.005) (0.005)
Triple differences, 1990-1970
[AP - ACP] - [ALP - ACLP] 0.020 -0.003 -0.016
(0.010) (0.008) (0.010)
[APW - ACP,W]] - [ALP,W - A CLP,W] 0.037 -0.001 -0.023
(0.009) (0.009) (0.009)
AP, APP, ALP, and ACP represent the change from 1970 to 1990 of the probability of being in a given sized
metropolitan area for power, part-power, low-power, and coincidental power couples, respectively. APw,ALPW.
ACPw,and ACLPWrepresent the probability of being in a given sized metropolitan area for power and low-power
couples in which the wife works and this probability for coincidental power and low-power couples in the
woman works, respectively. Probabilities are calculated from Tables V and VI.
0.037 suggests that 36 percent of the increase in power couple
concentration is accounted for by the unique coloration problems
of the college relative to the noncollege educated.
V. URBANIZATION AND THE COLLEGE EDUCATED
Two factors could account for the increasing urbanization of
the college educated: increasing returns to city size by education
and the increasing value of urban amenities to the college
educated (including marriage markets). To test the labor market
incentives of moving to large metropolitan areas, we estimate the
returns to city size by education in every decade. We therefore
estimate wage regressions of the form
(7) ln(w) = Po + ln(d) + MAX+u
1304 QUARTERLYJOURNALOF ECONOMICS
for white, full-time, and full-year wage and salary workers for
every decade and every education level, where w is the hourly
wage in 1997 dollars, d is metropolitan area population, X is a
vector consisting of age and age squared, and u is an error term.
We include both married and unmarried men and women in our
sample.13 Because we cannot observe population in area of
residence for individuals living in nonmetropolitan areas, we omit
these individuals from our regressions. For ease of exposition we
estimate separate equations by educational attainment. When
income is topcoded, we multiply the topcode by 1.45. The 1970
census reported weeks and hours of work only as intervals. We
take the midpoint of the interval. We drop all individuals whose
hourly earnings are less than half of the minimum wage. We also
drop individuals earnings more than $500 per hour.
Table IX shows the returns to city size by education for
individuals living in a metropolitan area (pj). Note that returns to
city size are greater for the highly educated. For each increase in
the logarithm of population in 1970 the logarithm of a man's wage
increases by 0.065 for those with a graduate degree and by 0.060
for those with a college degree in 1970 whereas for those with a
high school education it only increases by 0.049 and for those with
less than a high school education by 0.034. Only the returns to city
size for those with a graduate school education are statistically
significantly different from the returns for those with less than a
high school education (at the 5 percent level). By 1990 the
logarithm of the husband's wage increases by 0.101 for those with
a graduate education, by 0.078 for those with a college education,
by 0.076 for those with a high school education, and by 0.023 for
those with less than a high school education. The returns to city
size for those with a graduate, college, and high school education
are statistically significantly different from the returns to those
with less than a high school education, at the 1, 10, and 1 percent
level, respectively. Returns to city size by education barely
changed between 1970 and 1980, a period characterized by falling
returns to education [Katz and Murphy 1992].
Women's returns to city size by educational attainment
exhibit similar patterns as those of men, but only in 1990 are the
returns to city size for those with a graduate education statistically
significantly different from the returns to those with less
than a high school education. One possible explanation is that
13. The results are very similar when we restrict the sample to the married.
POWERCOUPLES 1305
TABLE IX
RETURNS TO CITYSIZE BY EDUCATION
1940 1970 1980 1990
Men
Less than high school 0.045 0.034 0.007 0.023
(0.008) (0.014) (0.019) (0.015)
High school 0.053 0.049 0.041 0.076
(0.005) (0.006) (0.010) (0.006)
College 0.044 0.060 0.051 0.078
(0.010) (0.007) (0.005) (0.010)
Graduate degree -0.002 0.065 0.061 0.101
(0.019) (0.005) (0.007) (0.014)
Women
Less than high school 0.065 0.062 0.040 0.065
(0.007) (0.011) (0.012) (0.014)
High school 0.060 0.066 0.058 0.075
(0.012) (0.006) (0.007) (0.009)
College 0.085 0.069 0.055 0.071
(0.018) (0.007) (0.006) (0.009)
Graduate degree 0.072 0.038 0.056 0.091
(0.028) (0.011) (0.010) (0.013)
Estimated returns are for individuals living in a metropolitan area only. We estimated regressions by
educational attainment in which the logarithm of the hourly wage was the dependent variable and the
independent variables were the logarithm of metropolitan area population and age and age squared. Robust
standard errors are in parentheses. Estimated from the integrated public use census samples.
many women work in the nonprofit sector and in this sector wage
differences across city size may be smaller and may have grown
more so over time. In the case of schoolteachers this may have
been spurred by the move away from direct local funding to state
and federal funding and the move toward collective bargaining
agreements [Flyer and Rosen 1997]. In fact, when we exclude
schoolteachers, those in government jobs, or those in traditionally
female jobs (a job in which women were overrepresented relative
to men in 1970), we find that in 1990 the returns to city size were
0.058, 0.093, 0.089, 0.123 for those with less than a high school
education, a high school education, a college education, and a
graduate education, respectively. The returns to city size for those
with a high school education or greater were statistically significantly
different from the returns for those with less than a high
school education.
Table IX shows that the returns to city size for the college
educated rose between 1970 and 1990, with most of the increase
taking place between 1980 and 1990. We predict that when the
1306 QUARTERLYJOURNALOF ECONOMICS
returns to city size for the college educated are rising then the
college educated should move to large cities and that they should
leave large cities when these returns are falling. Between 1970
and 1980, when the returns to city size by education were
stagnant or falling, coincidental couples migrated to large cities at
a slower rate (0.002) than between 1980 and 1990, when the
returns to city size by education were rising (0.034). These figures
suggest that most of the 0.036 increase (0.032 = 0.034 - 0.002) in
coincidental power concentration in large cities between 1970 and
1990 can be explained by increasing returns to city size for the
college educated. Note also that between 1970 and 1980, when the
returns to city size were stagnant, power couples still migrated to
large cities at a faster rate (0.026) than coincidental couples
(0.002), implying that the immediate pecuniary gains to city size
are not the sole motivation for power couple migration.14
Our final hypothesis for the urbanization of the college
educated was that power singles may be more likely to migrate to
large cities because cities may be becoming better marriage
markets. Most power singles are in a large metropolitan area (see
Table VI) and, because fewer marriages are now formed in high
school or college, the value to being in an area with a large
concentration of singles may be rising. It is also conceivable that a
forward looking highly educated single will foresee a future
coloration problem and will therefore prefer to meet someone who
already has invested in a large city. But, if there is value to being
in an area with a large concentration of singles, this has not
increased the probability of marriage. Power singles in large cities
are now less likely to marry relative to low-power singles.15
We therefore conclude that 35 percent of the increased
concentration of power couples in large cities can be explained by
the increasing urbanization of the college educated and that
virtually all of this 35 percent was due to increasing returns to city
size for the college educated.
14. Such couples, however, may recognize long-term pecuniary gains. Glaeser
and Mare [forthcoming] show that there are returns to city experience. Large cities
allow for greater job mobility and hence for greater lifetime wage growth and for
insurance against job loss.
15. Among individuals living in a large city 5 years ago and less than 30 years
of age, marriage rates among power men fell from 55 to 54 percent between 1980
and 1990, whereas those of low-power men rose from 58 to 63 percent. Among
power women marriage rates fell from 48 to 47 percent and among low-power
women they fell from 66 to 65 percent. Because we do not know metropolitan area
of residence from the 1970 census five years ago, we only present results for 1980 to
1990.
POWERCOUPLES 1307
VI. COLOCATION
The major pieces of evidence suggesting that coloration is an
important explanation for the rise in power couple concentration
are the double and triple differences based upon coincidental
power couple comparisons. The double differences suggest that
colocation explains up to 65 percent of the trend in power couple
concentration in large cities and all of the trend in low-power
couple concentration in large cities. The triple difference suggests
that 19 to 36 percent of the trend in power couple concentration is
attributable to the unique colocation problems faced by the college
educated and 29 to 46 percent to the coloration problems faced by
all couples.
Second, consider the prediction that power couples should be
concentrating in large metropolitan areas faster than part- or
low-power couples and that the increase in power couple concentration
in large cities should mainly be among couples in which
the wife works. As previously discussed, Table VIII shows that the
differential trend in power couples being in large cities compared
with low- and part-power couples was 0.060 and 0.049. Table VII
shows that between 1970 and 1990 the predicted proportion of
working power couples located in large cities rose by 0.202,
whereas the proportion of power couples in which the wife does
not work fell by 0.099. Note that the predicted proportion of
working power couples in major cities grew relative to working
part-power and low-power couples. The respective differential
trends were 0.032 and 0.085 (see Table VII). In contrast, when the
wife does not work, the differential trend in the predicted locational
choice of power couples and part-power and low-power
couples is smaller (0.017 and -0.026, respectively).
Finally, coloration also predicted that the trend in power
couple concentration in large cities should be most pronounced
among those couples in which the wife is working full-time or is in
a nontraditional occupation and among power couples in which
the husband and wife were in different professional occupations.16
In the largest metropolitan areas the predicted proportion of
16. Although the choice of city size is likely to affect hours of work, full-time
work may be an indicator of attachment to the labor force, particularly among the
college educated. City size could affect occupation as well, but a wife who has at
least a college education and is in an occupation that was traditionally male, say
law or medicine, is more likely to need the diversified labor market of a large city
than a wife who is in such a traditionally female occupation as that of schoolteacher.
Husbands and wives in the same occupation may not need the diversified
labor market offered by a large city.
1308 QUARTERLYJOURNALOF ECONOMICS
power couples in which the wife was in a nontraditional occupation
rose by 0.167 between 1970 and 1990. The increase among
power couples in which the wife was in a traditional occupation
was only 0.055. The predicted proportion of power couples in
which the wife works full-time and who were in a large metropolitan
area rose by 0.166 between these years, whereas the increase
among couples in which the wife works part-time was only 0.053.
Power couples in which the wife is in the same occupation
increased their concentration in metropolitan areas by only 0.025
between 1970 and 1990, whereas those in which the wife is in a
different occupation increased their concentration by 0.197.
The in-migration of power couples to large cities could reduce
power couples' incentive to stay in or move to large cities. But,
Table IX shows that wages are providing a continued incentive for
power couples to continue to move to large cities. Increasing
returns to scale models (such as Acemoglu [1996]) provide a
microfoundation for why the demand for highly skilled workers
increases with their numbers.17 As returns to education increase
the financial sacrifice of being in a small city rises, particularly for
dual career households.
While wages provide an incentive for power couples to move
to large cities, rents are a disincentive.18 Because power, partpower,
and low-power couples face the same rents but the wage
incentives to power couples of being in a large city have increased,
our results suggest that small cities have become relatively less
attractive to power than to part-power or low-power couples.
VII. IMPLICATIONS
Our finding that since 1940 power couples have been increasingly
likely to locate in large metropolitan areas and that the
growth of the coloration problem accounts for a large proportion of
this trend has implications for city growth. As the probability that
power couples choose a large metropolitan area rises, mean
educational levels in the city will rise. Educational levels in a city
are in turn positively related to city wages [Rauch 1993] and city
growth [Glaeser, Scheinkman, and Shleifer 1995].
17. Acemoglu's [1996] search model shows that ex ante investment and
bilateral search in the labor market will make the rate of return on human capital
increasing in the average human capital of the workforce. In Becker and Murphy's
[1992] model the greater density of urban areas lowers the costs of coordinating
specialists.
18. For evidence, see Rauch [1993] or Gyourko and Tracy [1991].
POWERCOUPLES 1309
As the share of power couples among all the college educated
rises, firms in smaller cities may find that it is becoming harder
and harder to attract highly skilled individuals. We illustrate the
difficulties faced by firms in small cities with suggestive evidence
on a particular type of firm-the university. One advantage of
examining universities is that their capital-to-labor ratio is fairly
fixed. Another is that universities, unlike firms, rarely move. We
therefore examine whether the relative quality of graduate research
doctorate programs in small cities in the United States has
fallen since 1970 to learn whether a firm that employs highly
skilled workers is now less likely to locate to a small city.
We use the National Research Council's data set, Research
Doctorate Programs in the United States, to obtain rankings of
1142 graduate programs of 100 universities in 1993, 1983, and
1970, and we link these data to metropolitan area population. (We
use the same metropolitan areas as in our previous empirical
work.) We classify all programs into quintiles in every year:
distinguished, strong, good, adequate, and marginal. The relative
proportion of programs in each category is therefore constant
across years, but between 1970 and 1993 58 percent of programs
changed categories. We then estimate an ordered probit model in
which the dependent variable consists of a categorical variable for
our five groups and in which the independent variables are the
logarithm of metropolitan area population, dummy variables for
broad program field (arts and humanities, biological sciences,
engineering, physical sciences and mathematics, and social and
behavioral sciences), and a dummy variable equal to one if the
university was a public institution.
Table X presents the derivatives with respect to the logarithm
of metropolitan area population from the ordered probit model.19
Note that the relationship between graduate program ranking
and population size is stronger in 1993 than in 1970. In 1970 an
increase of one in the logarithm of population increased the
probability that a program would be ranked distinguished by
0.007 and decreased the probability that it would be ranked
marginal by 0.010. Both the increase and decrease in these
respective probabilities for 1993 was 0.027. The relative decline in
quality of universities in small cities suggests that, if there are
spillover effects from universities, larger rather than smaller
cities are more likely to reap these benefits.
19. The derivatives for the ordered probit model were calculated by taking the
derivative for every observation and then taking the mean over all derivatives.
1310 QUARTERLYJOURNALOFECONOMICS
TABLE X
DERIVATIVES OF PROBABILITY THAT GRADUATE SCHOOL PROGRAM
Is DISTINGUISHED STRONG, GOOD, ADEQUATE, AND MARGINAL WITH RESPECT
TO LOGARITHM OF CITY POPULATION, 1993, 1983, AND 1970
Derivative wrt logarithm of
city population
1970 1983 1993
Mean of logarithm of population 7.075 7.547 7.673
Probability
Distinguished (top quintile) 0.007 0.024 0.027
(0.007) (0.008) (0.009)
Strong 0.007 0.013 0.007
(0.006) (0.004) (0.003)
Good -0.000 -0.006 -0.011
(0.001) (0.003) (0.004)
Adequate -0.005 -0.016 -0.017
(0.005) (0.006) (0.006)
Marginal (bottom quintile) -0.010 -0.027 -0.027
(0.013) (0.019) (0.020)
1142 observations in all years. Population is in the 1000s. The categorical outcome variable for the ordered
probit model is the quintile ranking, and the independent variables are the logarithm of population, dummy
variables for broad program field, and a dummy variable equal to one if the university was a public institution.
Derivatives are mean derivatives. Standard errors are in parentheses. We rejected the hypothesis that 1970
and 1993 and 1970 and 1983 should be pooled. Estimated from Research Doctorate Programs in the United
States.
VIII. CONCLUSION
This paper has documented the rising concentration of power
couples in larger over smaller metropolitan areas and over
nonmetropolitan areas relative to other household types and to
that which would have been predicted for two observationally
identical single individuals. We argued that up to 65 percent of the
increased concentration of power couples in larger metropolitan
areas could be explained by the coloration problem. Of this 65
percent, 19 to 36 percent was accounted for by the unique
coloration problems faced by the college educated and 29 to 46
percent by the coloration problems faced by all couples. The
remaining 35 percent could be explained by the increasing
urbanization of the college educated because of rising returns to
city size by education.
The increased concentration of power couples in large metropolitan
areas may have implications for the dynamics of city
growth. Economic growth depends upon the ability to absorb
POWERCOUPLES 1311
existing knowledge and to create new knowledge, both of which
are directly related to the existing stock of human capital. Smaller
markets have always exported the highly skilled to larger markets
and, as this paper has documented, this phenomenon has
been magnified by the increased bundling of the highly skilled
with other highly skilled spouses. Cities, especially low amenity
cities, may face a greater net "brain drain" than in the absence of
power couple bundling. The coloration problem may also affect a
small city's adjustment to local labor market shocks. Regional
adjustment to local labor market shocks is primarily driven by
labor mobility [Blanchard and Katz 1992]. But, because of bundling
a small city may only slowly attract high skilled talent
despite a local boom. Foreseeing this, firms may be unwilling to
locate in smaller cities. Universities provide suggestive evidence.
We have shown that although the quality of graduate doctoral
programs was positively related to city size in both 1970 and in
1993, the relationship between program ranking and city size has
become stronger.
This paper has sketched a 50-year trend in power couple
locational choice, but will information technology affect future
locational choice? It is possible that the growth of information
technology that permits highly skilled workers to telecommute
may solve the coloration problem for some couples by permitting
at least one spouse to live far from where their employer is
located. This could allow smaller cities to attract a highly skilled
couple. But, as more couples become true dual career households,
an increasing proportion will be faced with a coloration problem.
Both the 50-year trend and the increase in returns to city size by
education suggest that power couples will turn to large cities.
Furthermore, information technologies may be a complement, not
a substitute, for living in a large city if they facilitate making new
business contacts [Gaspar and Glaeser 1998]. Power couples are
likely to have a comparative advantage in making new contacts
relative to two highly educated single people. Although this paper
has not explicitly explored the couple "synergies" of two highly
educated people being married, a power couple may work as a
"team" to maximize household income. They will therefore both
seek potential business contacts who can work with them or their
spouse, and such contacts are more likely to be found in large
cities.
1312 QUARTERLYJOURNALOFECONOMICS
DATA APPENDIX
We use the 1940 and 1970-1990 censuses of population and
housing.20 For each person we observe marital status, age, sex,
race, education, labor force status, occupation, and metropolitan
area. Our couples are all married and living in the same household,
and our singles are either never married, divorced, or
widowed. We treat cohabitating couples (whom we can only
identify in 1990) as singles. All estimates use the population
weights for 1940 and 1990. With the exception of 1970 we observe
metropolitan area five years ago.
We construct the following variables.
1. Metropolitan area size classifications.
Although the concept of a metropolitan area has remained
essentially the same throughout the years, the boundaries
of metropolitan areas have grown, new metropolitan areas
have emerged, and there were slight variations in how the
concept was defined from census to census and in the
confidentiality criteria that had to be met for a metropolitan
area identified. We use metropolitan area size classifications
that allow for comparability across all census
years.
Our first step in creating metropolitan area size classifications
is to classify the suburbs of central cities as part of
the labor market of the central city (e.g., Westchester
county is classified as part of the New York City labor
market). We allow for the expansion of metropolitan areas
into farmland at the periphery. For example, the 1940 and
1970 censuses did not include Santa Rosa as part of the
San Francisco Bay Area, whereas those of 1980 and 1990
did. Our definition of the San Francisco Bay Area excluded
Santa Rosa in 1970 (when it was still a small, rural town)
but included it in 1980 and 1990 (when it had become a
suburb). We have experimented with consistent definitions
of metropolitan area across census years and our
results are not affected by the definition that we used.21
Our final step is to create three city size categories: large
20. We use the integrated public use micro samples available at http://
www.ipums.umn.edu/. Earlier censuses did not identify education. We cannot use
the 1950 census because education is known only for the sample line person. We
cannot use the 1960 census because metropolitan area is not identified.
21. Jaeger et al. [1998] show how to construct consistent definitions of
metropolitan areas from 1970 to 1990.
POWERCOUPLES 1313
metropolitan areas (those with populations of at least 2
million), midsize metropolitan areas (those with populations
of between 2 million and 250,000), and small and
nonmetropolitan areas (metropolitan areas with a population
of less than 250,000 and nonmetropolitan areas). The
1940 census identified metropolitan areas if the population
in these areas was at least 100,000 in 1980 and the
1980 and 1990 censuses identified metropolitan areas
with populations of at least 100,000 in the census year.
The 1970 census identified metropolitan areas with populations
of at least 250,000 in 1970. Our definition of small
and nonmetropolitan areas is therefore consistent across
time.
2. Educational level
The definition of education differed across census years.
We use the highest grade of school or year of college
completed for 1940, 1970, and 1980. Education is overstated
in 1940 [Goldin 1998]. In 1990 the education
variable gives the respondent's highest grade of school
completed through the eleventh grade, but classifies high
school graduates according to their highest diploma or
degree earned. We therefore define our categories of less
than high school as grade 11 or less, high school as grade
12 in 1940, 1970, and 1980 and as twelfth grade, high
school diploma, or GED in 1990, college as four or more
years in 1940, 1970, and 1980 and as a bachelor's or
graduate degree in 1990. We classify those who did not
complete college (less than four years in 1940, 1970, and
1980 and some college but no degree or occupational!
academic associate degree in 1990) together with the high
school graduates.
3. Hourly Wage
Our hourly wage variable is constructed from annual wage
and salary divided by annual hours worked (current
weekly hours multiplied by weeks worked in the past
year). We examine wage and salary income only because
self-employment income is unavailable in 1940. We adjust
all income numbers to 1997 dollars. We multiply the
topcode by 1.45. In 1980 a larger proportion of the population
was covered by topcoding than in 1970 or 1990.
However, our results remain unchanged when we impose a
new topcode in 1970 and in 1990.
1314 QUARTERLYJOURNALOFECONOMICS
The 1940, 1980, and 1990 censuses provide actual hours
worked during the reference week and actual weeks
worked in the past year. The 1970 census only provides
intervals. We therefore take the midpoint of these
intervals.
MASSACHUSETTSINSTITUTEOFTECHNOLOGYANDNATIONALBUREAUOFECONOMIC
RESEARCH
COLUMBIAUNIVERSITY
REFERENCES
Acemoglu, Daron, "A Microfoundation for Social Increasing Returns in Human
Capital Accumulation," Quarterly Journal of Economics, CXI (1996), 779-804.
Baumgardner, James R., "Physicians' Services and the Division of Labor across
Markets," Journal of Political Economy, XCVI (1988), 948-982.
Becker, Gary S., and Kevin M. Murphy, "TheDivision of Labor, Coordination Costs,
and Knowledge," Quarterly Journal of Economics, CVII (1992), 1137-1160.
Blanchard, Olivier Jean, and Lawrence F. Katz, "Regional Evolutions," Brookings
Papers on Economic Activity (1992), 1-61.
Flyer, Frederick, and Sherwin Rosen, "The New Economics of Teachers and
Education," Journal of Labor Economics, XV (1997, Part 2), S104-139.
Gaspar, Jess, and Edward L. Glaeser, "Information Technology and the Future of
Cities," Journal of Urban Economics, XLIII (1998), 136-156.
Glaeser, Edward L., "Are Cities Dying?" Journal of Economic Perspectives, XII
(1998), 139-160.
Glaeser, Edward L., and David C. Mare, "Cities and Skills," Journal of Labor
Economics, forthcoming.
Glaeser, Edward L., Jose A. Scheinkman, and Andrei Shleifer, "Economic Growth
in a Cross-Section of Cities," Journal of Monetary Economics, XXXVI (1995),
117-143.
Goldin, Claudia, Understanding the Gender Gap: An Economic History of American
Women (New York, Oxford: Oxford University Press, 1990).
_ "The Meaning of College in the Lives of American Women: The Past
One-Hundred Years," National Bureau of Economic Research Working Paper
No. 4099, June 1992.
_ "Career and Family: College Women Look to the Past," in Gender and Family
Issues in the Workplace, Francine Blau and Ronald Ehrenberg, eds. (New
York:Russell Sage, 1997).
_ "America's Graduation from High School: The Evolution and Spread of
Secondary Schooling in the Twentieth Century." Journal of Economic History,
LVIII (1998), 345-374.
Gyourko, Joseph, and Joseph Tracy, "The Structure of Local Public Finance and
the Quality of Life,"Journal of Political Economy, XCIX (1991), 774-806.
Heckman, James, Anne Layne-Farrar, and Petra Todd, "Human Capital Pricing
Equations with an Application to Estimating the Effect of Schooling Quality
on Earnings,"Review of Economics and Statistics, LXXVIII (1996), 562-610.
Jaeger, David A., Susanna Loeb, Sarah E. Turner, and John Bound, "Coding
Geographic Areas across Census Years: Creating Consistent Definitions of
Metropolitan Areas," National Bureau of Economic Research Working Paper
No. 6772, October 1998.
Kahn, Matthew E., "Particulate Pollution Trends in the United States," Regional
Science and Urban Economics, XXVII (1997), 87-107.
Katz, Lawrence F., and Kevin M. Murphy, "Change in Relative Wages, 1963-1987:
Supply and Demand Factors," Quarterly Journal of Economics, CVII (1992),
35-78.
Kim, Sunwoong, "Labor Specialization and the Extent of the Market," Journal of
Political Economy, XCVII (1989), 692-705.
POWER COUPLES 1315
Lucas, Robert E., Jr., "On the Mechanics of Economic Development," Journal of
Monetary Economics, XXII (1988), 3-42.
Mare, Robert D., "Five Decades of Educational Assortative Mating," American
Sociological Review, LVI (1991), 15-32.
Mincer, Jacob, "Family Migration Decisions," Journal of Political Economy,
LXXXVI (1978), 749-773.
Rauch, James E., "Productivity Gains from Geographic Concentration of Human
Capital: Evidence from the Cities," Journal of Urban Economics, XXXIV
(1993), 380-400.
Ruggles, Steven, and Matthew Sobek, Integrated Public Use Microdata Series,
IPUMS-98 Version2.0 (Social History Research Laboratory. Minneapolis, MN:
Department of History, University of Minnesota, 1997), http://www.ipums-
.umn.edu/.
Simon, Curtis, "Human Capital and Metropolitan Employment Growth," Journal
of Urban Economics, XLIII (1998), 223-243.
Topel, Robert H., and Michael P. Ward, "Job Mobility and the Careers of Young
Men," Quarterly Journal of Economics, CVII (1992), 439-479.