THE UNIVERSITY OF CHICAGO PRESS JOURNALS 'the-. Snrifty of l.nljnr F,rtidnmints NORC at the University of Chicago Why Are Power Couples Increasingly Concentrated in Large Metropolitan Areas? Author(s): Janice Compton and Robert A. Pollak Source: Journal of Labor Economics, Vol. 25, No. 3 (July 2007), pp. 475-512 Published by: The University of Chicago Press on behalf of the Society of Labor Economists and the NORC at the University of Chicago Stable URL: http://www.jstor.org/stable/10.1086/512706 Accessed: 19-03-2018 14:41 UTC JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://about.jstor.org/terms NORC at the University of Chicago, Society of Labor Economists, The University of Chicago Press are collaborating with JSTOR to digitize, preserve and extend access to J'ournal of Labor Economics JSTOR This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Why Are Power Couples Increasingly Concentrated in Large Metropolitan Areas? Janice Compton, University of Manitoba Robert A. Pollak, Washington University in St. Louis Using the Panel Study of Income Dynamics (PSID), we test Costa and Kahn's colocation hypothesis, which predicts that power couples—couples in which both spouses have college degrees—are more likely to migrate to the largest cities than part-power couples or power singles. We find no support for this hypothesis. Instead, regression analyses suggest that only the education of the husband and not the joint education profile of the couple affects the propensity to migrate to large metropolitan areas. The observed location trends are better explained by higher rates of power couple formation in larger metropolitan areas. I. Introduction Couples in which both husband and wife have college degrees are increasingly likely to be located in large metropolitan areas. In 1970, 39% of these couples—called "power couples" by Costa and Kahn (2000)— lived in metropolitan areas with a population of at least 2 million. In 1990, this number had jumped to 50%. By comparison, among couples in which We would like to thank Bart Hamilton, Emek Basker, Scott Drewianka, and Courtney Lafountain for their helpful comments and suggestions and Michael Orlando for programming assistance. Pollak thanks the John D. and Catherine T. MacArthur Foundation for their support. The usual disclaimer applies. Contact the corresponding author, Janice Compton, at comptonj@cc.umanitoba.ca. [Journal of Labor Economics, 2007, vol. 25, no. 3] © 2007 by The University of Chicago. All rights reserved. 0734-306X/2007/2503-0001$10.00 475 This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 476 C ompton/Pollak only one spouse has a college education—"part-power couples"—the proportion living in large metropolitan areas increased more slowly, growing from 36% to 42% over the 2 decades. Couples in which neither spouse has a college degree—"low-power couples"—have the lowest probability of living in a large city and the lowest rate of increase, growing from 30% to 34% in 20 years (Costa and Kahn 2000). Costa and Kahn consider two main explanations for the increasing concentration of power couples in large metropolitan statistical areas (MSAs). First, large metropolitan areas may be increasingly attractive to the college educated, regardless of their marital status. The increase in urbanization of the college educated may reflect higher returns to education in larger MSAs or urban amenities more commonly found there. Thus, the observed trends may be the result of college-educated singles moving to large urban areas and then marrying. Second, the labor markets found in large metropolitan areas may hold particular attraction for power couples. All married couples potentially face the possibility that their location preferences will differ. This "colocation problem" may arise from different preferences for amenities or proximity to family, but economists emphasize job opportunities. If husband and wife work in the paid labor market and if they live together, then they must both find acceptable employment in the same location. With the rise in married women's labor force participation, the potential for colocation conflict has increased for all couples but may be most acute for the college educated because they tend to have more specialized careers. Solving this type of colocation problem may be easier in the labor markets of large metropolitan areas, and thus these areas should be magnets for power couples. Costa and Kahn conclude that the ability of large MSAs to solve the colocation problem explains most of the observed increase in the concentration of power couples in large MSAs. Although the data used in their analysis are cross-sectional, they suggest that the migration of power couples is the principal mechanism underlying the changes in observed location patterns. In this article we investigate the relationship between migration and the education profiles of couples. We ask two questions. First, is there evidence that the colocation pressure faced by highly educated couples is a determinant of migration? That is, compared to other couples, are power couples more likely to migrate into, and less likely to leave, large MSAs? Second, do the migration patterns of power couples differ from those of other couples? Positive responses to these questions would support Costa and Kahn's colocation explanation for the concentration of power couples in large urban areas. We do not, however, find evidence supporting the colocation explanation. Using data from the Panel Study of Income Dynamics (PSID), we find that, holding all else equal, power couples are more likely to migrate and, This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 477 when they migrate, are more likely to migrate to large urban areas than are low-power couples. We find, however, that the factor determining couples' migration is not their joint education profile but the husband's education. More specifically, we find that the migration behavior of part-power couples in which only the husband is a college graduate is indistinguishable from that of power couples, while the migration behavior of part-power couples in which only the wife is a college graduate is indistinguishable from that of low-power couples. Our conclusion that migration is affected by the husband's education rather than the couple's joint education profile is unaltered when we include controls for postgraduate education and occupation to identify those couples most likely to have specialized careers. Logic suggests that the observed location of power couples must be the result of four dynamic processes: migration, marriage, divorce, and human capital accumulation. Looking at the stock of power couples living in large metropolitan areas, we can conclude only that each spouse must have earned a college degree, married a spouse who earns/earned a college degree, remained married, and migrated to and/or chosen to remain in a large metropolitan area. We argue that the migration of power couples into large urban areas is unlikely to be the primary mechanism behind the concentration of power couples in large MSAs. The formation of power couples in large MSAs provides a more plausible explanation. Although the differences between groups are not precisely estimated, the patterns suggest that differences in the migration rates of power singles, in assortative mating patterns, and in college enrollment rates by city size may be responsible for the increasing concentration of power couples in large MSAs. This suggestion is strengthened by the 2000 census data, data unavailable to Costa and Kahn. Between 1990 and 2000 the proportion of power couples living in large MSAs fell, while the proportion of low-power couples in large MSAs increased. This reversal of the earlier trend is difficult to explain in terms of power couple migration and colocation pressure. The article proceeds as follows. Section II provides a brief review of the literature on couple migration. In Section III we describe the data, the econometric model, and the variables we use to investigate the relationship between the joint education profile of spouses and the probability of migration. Regression results are presented in Section IV. Section V provides an overview of couple migration patterns as well as the other mechanisms that may explain the location trends—migration of singles, assortative mating, marriage and divorce trends, and educational attainment by city size. Section VI reports on changes in power couple concentration in large urban areas using data from the 2000 census. Section VII concludes. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 478 C ompton/Pollak II. Migration of Couples Economists frame individuals' migration behavior as human capital decisions (Sjaastad 1962; Mincer 1978). A person will migrate if the expected benefit of living in another location less migration costs exceeds the expected benefit of remaining in the current location. The main statistical trends of migration are consistent with this model—younger individuals are more mobile than older individuals, more educated persons are more mobile than those with less education, rural to urban migration is more common than the reverse, and the longer individuals have resided in an area the less likely they are to migrate. (For a review of the migration literature, see Greenwood [1997]). Polachek and Horvath (1977), Sandell (1977), and Mincer (1978) were among the first to frame the migration decision explicitly in terms of family net gain rather than individual net gain. These researchers abstract from decision making within the family and assume that families migrate if and only if the sum of the benefits of migrating is greater than the sum of the costs, where the sums are taken over all family members. Family migration is less likely than individual migration since the costs of migrating increase with the number of family members while the benefits may not. The model suggests that one spouse, typically the husband, is often a "tied stayer" while the wife is often a "tied mover." Tied stayers forgo moves that would result in positive net returns for the individual but negative net returns for the family. Tied movers participate in moves that result in a loss for themselves but positive net returns for the family. Empirical work confirms the importance of husbands' career opportunities in migration decisions. Using a sample of college graduates, Duncan and Perrucci (1976) show that the higher the occupational prestige and migration opportunities connected to the husband's occupation, the greater the probability that the couple will migrate. They find, however, that the occupational prestige and migration opportunities of the wife's career did not affect migration probabilities. Using data also from the 1970s, Bielby and Bielby (1992) link the migration behavior of couples to gender-role beliefs. Couples who subscribe to a more traditional "provider role" for husbands will not be deterred from moves that harm the wife's employment opportunities but will be deterred from moves that harm the husband's career opportunities. More recent studies continue to find similar gendered determinants of couple migration (Gardner, Pierre, and Oswald 2001; Nivalainen 2004). Unlike these large-scale statistical studies, Green (1997) uses a case-study approach to investigate the migration/location decisions of dual-career households. On the basis of interviews with 30 such couples, she finds that the leader in migration decisions is most often the spouse with (a) the more highly paid career, (b) the more location-constrained career, This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 479 and/or (c) the more insecure career. In the large majority of cases, the spouse with these job characteristics was the husband. Green notes that although the respondents appreciate the greater career opportunities in large urban centers, many are unwilling to incur the costs associated with such large areas—high costs of living, congestion, and so forth (Green 1997). Because career choice is endogenous to beliefs about migration and marriage, Frank (1978) argues that expectations of migration based on the maximization of net family welfare help explain the male-female wage gap. Women are less likely to invest in their own human capital when their opportunity set is limited by their husbands' location choices. Similarly, if employers believe that the location decisions of women are determined largely by their husbands' careers and not their own, employers may be less likely to invest in firm-specific human capital for women. As a result, women may tend to enter more mobile careers—careers that build largely on general rather than specific human capital and careers in which opportunities are not concentrated in large urban areas. Studies of couple migration by Spitze (1984), Bird and Bird (1985), Morrison and Lichter (1988), Shihadeh (1991), Cooke and Bailey (1996), and Jacobsen and Levin (1997) are broadly consistent with Frank's argument. Robst and McGoldrick (1996) investigate the impact of destination size on outcomes for tied movers. They began by hypothesizing that while a trailing spouse has a job search limited to the location that the leading spouse has chosen, if the location is a large MSA and this in turn reflects higher job vacancies, then the trailing spouse may find a wage offer near the maximum that could be obtained in a nationwide search. This hypothesis was not supported by the data. The authors found that the size of the local labor market had no significant effect on the likelihood of women being "overeducated," that is, having a higher level of education than the average within their occupation. They suggested that it is not the market's size that is important, but its job composition: a small local labor market with a relatively large concentration of white-collar employment is as beneficial to women as a large labor market without such a concentration. III. Econometric Model and Data Description Our econometric analysis addresses the following question: compared to part-power and low-power couples, are power couples more likely to migrate to and less likely to migrate from large cities? To answer this question, we consider the determinants of couple migration, focusing on couples' joint education profiles. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 480 C ompton/Pollak A. Data The regression analyses presented in the following sections use data from the Panel Study of Income Dynamics (University of Michigan Institute for Social Research 2002). The PSID is a longitudinal survey of a representative sample of American individuals and families. The original sample included members of 4,800 families. All individuals from the original families are followed as they move or form new family units. For this study we have included all married couples who live together and all unmarried heterosexual couples who have lived together for at least 1 year.1 In all that follows, we define "power couples" as those in which both husband and wife have a bachelor's degree; "part-power couples" as those in which only one spouse holds a bachelor's degree; and "low-power couples" as those in which neither spouse holds a bachelor's degree.2 When we refer to an individual as "college educated" we mean that he or she holds a bachelor's degree. The sample is limited to men aged 25-39 and women aged 23-37. This restriction allows us to focus on those individuals and couples whose migration decisions would be most influenced by career considerations.3 The Geocode Match Supplement to the PSID provides detailed geographic identifiers necessary for a migration study. Building up from the county of residence, we sort observations into Metropolitan Statistical Areas (MSAs). The U.S. Census Bureau defines MSAs as agglomerations of counties that include a large population nucleus and adjacent communities that share a high degree of social and economic integration. We define migration as a move between MSAs, between an MSA and a non-MSA county, or between two non-MSA counties. Thus, moves within an MSA are not treated as migration. The changing definition of MSAs poses 1 Dropping couples who were not legally married did not alter the results in preliminary analyses. 2 Some imputation was required for the education variable. Education was asked of respondents when they entered the household but not at each subsequent round. In 1985, the education questions were re-asked of all individuals in the sample. Since year of graduation was asked, it was possible to work backward from 1985 to code the education attainment for previous years. For years after 1985, only new heads/spouses were asked the education questions. For those already in the sample, if they were enrolled in school in 1985, they were assumed to have finished their program and so a degree was imputed to them in following years. If individuals were not enrolled in school in 1985, their education level observed in 1985 was simply carried forward. The resulting imputations look reasonable in comparison to average education of the individual's occupation. A robustness test was performed in which education was raised to be at least the minimum level of education that individuals who reported the same occupation deemed necessary to perform the job. The results are not substantially different. 3 These age categories are the focus of Costa and Kahn (2000). Similar trends in location are found with expanded age categories. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 481 a problem for migration analysis. For example, according to the 1990 MSA definitions, Baltimore, MD, and Washington, DC, were distinct MSAs, but according to the 2000 definitions they are part of a single MSA. To prevent changing MSA definitions from creating the appearance of changes in migration rates, we apply the 1990 MSA definitions consistently throughout the sample.4 In the analyses that follow, we define "large" metropolitan areas as MSAs with populations greater than 2 million, "midsize" metropolitan areas as MSAs with populations between 250,000 and 2 million, and "small" metropolitan areas as nonmetropolitan areas and MSAs with populations less than 250,000. We choose these categories to be consistent with those used by Costa and Kahn. Appendix A lists the large MSAs for the 1970-2000 period. In our regression analyses we use data for the period 1980-93. We limit our analysis to these years because some of the relevant variables—most importantly, the education variables—were not available for any longer period. Since it is during this period that we observe the greatest increase in power couple concentration in large MSAs, we would expect the co-location effect to be strongest during this period. The panel data are pooled into 12 2-year cross-sections. For each couple-year observation, the dependent variable captures the migration behavior observed in year 2 and all independent variables measured in year 1. Our sample includes 21,955 couple-year observations with 638 observed migrations (3%). MSA attributes are dervied from the 1988 and 1993 County and City Data Books (U.S. Census Bureau 1988, 1993). B. Econometric Model We estimate two econometric models of migration. In the first we estimate the probability of migrating without regard to the destination. The dependent variable is binary, indicating whether or not the couple migrates. In the second model we distinguish migrants by the size of destination. Here, the couple chooses among four alternatives: staying in the current location, migrating to a small MSA, migrating to a midsize MSA, and migrating to a large MSA. The decision to migrate and the size of destination are determined simultaneously. 1. Sample Selection Issues Two selection issues require attention. The first is attrition: the requirement that migration variables be defined in year 2 results in the loss of observations. The second is change in marital status: because our observations are at the household level—unmarried individual heads of 4 Costa and Kahn (2000), whose focus is not dynamic, use the current year MSA definition for each census year that they consider, but they note that their results do not change substantially under alternative (stable) definitions. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 482 C ompton/Pollak households and married couples—and are defined over a 2-year period, the marital status must be constant over the 2 years. Hence, the location in year 2 is defined only for those couples that remain married. As a result, our samples are doubly selected to include only those observations that (i) have not been lost to attrition and (ii) have the same marital status in year 2 as in year 1. Beginning with a sample of 26,324 couple-year observations, we lose 2,264 observations to attrition and 1,354 observations to divorce. We dropped 751 additional observations due to missing values for key variables or death. The model for the binary choice decision is a three-equation latent dependent variable model where equation (1) estimates the probability of remaining in the sample in year 2 (i.e., no attrition), equation (2) estimates the probability of remaining married in year 2, and equation (3) is the equation of interest, estimating the probability of migration: qf = V,5 + v„ i = 1 ...Nu (1) zf = \% + co„ i = 1 ...N2, N2,, cc,-, and u{ are identically and independently normally distributed zero mean error terms. Following Vella (1992, 1993, 1998) and Eklof and Karlsson (1999), we test for sample selection in the binary choice model. Rewrite equation (1) as yf = Xfi + Xjcc, + X2u; + I/,-, i = 1 ... Nt, (4) where Xt and X2 are equal to oHJal and oHJal, respectively. The term ?7, is a white noise error term. Replacing cc, and i>, with the estimated generalized residuals from the probit models—which by Vella (1993) are equal to their respective inverse Mills ratios, we estimate yf = Xfi + Xt [0(2/7)1 + \ [<%'S)1 ~ A 2 mq'M ■i/,., i = l...Nv (5) Under the null hypothesis, Xt and X2 are equal to zero and 17, is equal to ux. The t-tests on Xt and X2 are tests of selection bias since under the null This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 483 hypothesis the estimates are consistent. In general, however, under the alternative hypothesis, the estimates are not consistent. The test results are presented in table 1. Shown are some key variables as well as the test statistics.5 Column A presents the binary choice probit regression (eq. [1]) without adjusting for sample selection. Column B presents the results from equation (5), including the two test statistics. As indicated by the t-tests, neither Xt nor \2 are significant. In the regressions presented in columns C and D we alter the modeling of sample selection. In column C the two selection criteria are combined. In this model the first step equation estimates the probability of being observed in year 2 (i.e., no divorce and no attrition). In column D the observations that have been lost due to attrition are simply dropped, and we model only the sample selection arising from divorce.6 In both variations, the t-test indicates insignificant test variables. Column E estimates the abridged model of column D under a maximum likelihood probit estimation procedure outlined by van de Ven and van Praag (1981) and implemented through STATA using the HECKPROB command.7 The estimates in this model are consistent and asymptotically efficient even in the presence of selection bias as it allows for the estimation of the correlation between the error terms and ur The likelihood ratio test of independent equations—testing the null hypothesis that p = 0— indicates again that sample selection bias is not a concern. Overall, the t-tests on the inverse Mills ratios added to the regressions in columns B-D and the likelihood ratio test in column E indicate that sample selection bias is not significant in this model. In the analyses that 1 Full regression results are available upon request. ' More formally, in col. C we present a variation of eq. (5): yf = Xß + \, mi). » = 1, ... JV„ where zf estimates the probability that individual i is observed and married in year 2. Likewise, in col. D we present the following variation of eq. (5): Xß + X, *&7). ■ Vi, 1, ... N2, z? = Wfl + oii, i = 1, where observations lost to attrition are dropped from the sample and zf estimates the probability that individual i is married in year 2. 7 Maximum likelihood estimates in which the missing observations are dropped and selection is solely due to marital stability are similar. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Table 1 Effect of Power Status on Probability of Migrating: Testing Sample Selection Probit: Maximum ML Estimation Without Sample „ , ■ T £ c 1 c 1 ■ Likelihood without Past _ , . _ r , Probit: lests tor sample selection . . . , .. selection Controls _-_ Estimation Behavior Variables (A) (B) (C) (D) (E) (F) Power: both spouses have college degrees .378 (.060)* .363 (.085)* Half power: only one spouse has a college degree .245 (.000)* .249 (.080)* Origin—midsize MSA -.018 (.045) -.016 (.045) Origin—large MSA -.165 (.065)* -.162 (.066)* Currently residing in home state of at least one spouse -.543 (.040)** -.542 (.040)* Head has previously moved for a job .092 (.044)* .094 (.045)* x, -.511 (1.129) K 1.157 (1.163) x3 p Wald test of independent equations (p = 0): Test statistic: x2(l) Prob > x2 .396 (.076)* .346 (.084)* .383 (.059)* .199 (.177) .266 (.078)* .018 (.045) .165 (.065)* .218 (.073)* -.017 (.045) -.163 (.065)* .254 (.053)* -.019 (.044) -.163 (.064)* .095 (.122) -.020 (.030) -.094 (.058): .543 (.040)* -.543 (.040)* -.531 (.043)* .093 (.044)* .092 (.045)* -.645 (1.114) .090 (.044)* .389 (1.086) .478 (.511) -.891 (.151) .62 3.76 .432 .052 Note.—Coefficients and standard errors presented. Standard errors are adjusted for clustering on household identifier. ;:" Coefficients are significant at the 5% level. Power Couples 485 follow, we present the binary choice estimation results from the maximum likelihood probit model. For the second model, in which size of the destination is determined jointly with the decision to migrate, we rely on the above test results and estimate the model under the assumption that sample selection bias is not significant. Here we employ a multinomial logistic specification and estimate the following model using a maximum likelihood procedure.8 Extending the binary model from (4), the probability that observation i chooses outcome ; is pii = Pr(Y1=j) = 1 1 +SL2exp(X,/3J ' exp(X,/3;) .1 +SL2exp(X,/3J where; = 1, 2, 3, 4 denotes the four categorical outcomes (not migrating, migrating to a small MSA, migrating to a midsize MSA, migrating to a large MSA); the base outcome (j = 1) is the choice of not migrating. In the results that follow, we present the relative risk ratios, which are equal to the exponentiated value of a coefficient. This allows for an easy interpretation: the relative risk ratio indicates the change in the probability of observing outcome k, relative to the probability of not migrating, for a one unit change in the corresponding variable. 2. Endogeneity Migration and education are endogenous if individuals have unobserv-able characteristics that are reflected in a higher propensity to invest in human capital through both education and migration. To correct for unobserved heterogeneity in the propensity to invest in human capital, we control for whether the current state is the home state of either spouse and whether or not the household head (read: husband) has previously moved in order to take a job. The inclusion of these variables weakens the endogeneity concerns as they capture some of the otherwise unob-servable propensity to invest in human capital. Indeed with the inclusion of these variables, Hausman tests allow us to reject the null hypothesis of endogeneity at the 5% confidence level. Column F in table 1 highlights 8 Although we assumed a normal distribution in the first case, we assume a logistic error distribution here. As we have no a priori beliefs concerning the distribution of the errors, the choice of error form was determined solely by ease of computation. While it would have been more consistent to estimate multinomial probit regressions, convergence was difficult. The binary choice regressions were run under the assumption of logistic errors, and the results were very similar. = 1 = 2,3,4 This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 486 C ompton/Pollak the importance of these variables. While the results are robust to inclusions and omissions of other control variables, the omission of these variables significantly alters the impact of the education variables. 3. Control Variables In the following discussion, we focus on couples' joint education profile as a determinant of migration, controlling for other factors such as family, location amenities, employment, and occupation. Appendix B shows means and standard deviations for the full sample, which includes all couple-year observations, and separately for the migrant observations. The means show the strong importance of the couples' home state and previous migration behavior. Close to 80% of the couple observations are residing in the home state of at least one spouse. Compare this with only 45% of migrant observations. In 26% of all observations, the head had moved previously for a job; the corresponding figure for migrant observations is 38%. In our sample, migrants are younger, slightly less likely to have preschool children, less likely to be homeowners, and less likely to be nonwhite. Migrants' origins tend to be characterized by slower growth, higher housing value, and greater distance to any large MSAs. In terms of labor force attachment, migrants have lower job durations and are more likely to be unemployed or out of the labor force. IV. Regression Results The regression analysis provides no evidence that colocation pressure is a determinant of migration. Tables 2-5 provide key regression results. For the binary model, the marginal effects of the key variables are presented—the change in the probability of migrating due to a change in the dummy variable from zero to one. For the multinomial model, the results are presented as relative risk ratios. For a dummy variable, this is the change in the probability of migrating to the indicated destination size over the probability of not migrating that occurs due to a discrete change in the variable from 0 to 1. A. The Effect of Power Couple Status on Migration Table 2 presents the results for key variables in the binary model. Column A corresponds to column E in table 1 and controls for sample selection using maximum likelihood estimation. The regression results show that it is not the joint education profile of a couple but only the husband's education that affects the decision to migrate. The analysis hinges on the behavior of couples in which one spouse has a college degree and the other does not (i.e., the part-power couples). In column A of table 2, we pool the part-power couples in which the husband is a college graduate with the part-power couples in which the This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Table 2 Binary Model: Probability of Migration A B c D Power: both spouses have college degree .022* .022* .026* (.000) (.000) (.010) Half power: only one spouse has a college degree .013* (.000) Husband power: only husband has college degree .020* (.000) .021* (.027) Wife power: only wife has college degree .001 (.786) -.016 (.126) Husband power2: husband has a college degree .019* (.000) Wife power2: wife has a college degree .001 (.614) Origin—midsize MSA -.001 -.001 -.001 -.001 (.666) (.685) (.686) (.697) Origin—large MSA -.007* -.007* -.007* -.013* (.011) (.012) (.012) (.015) Power x Large MSA -.007 (.491) Husband power x Large MSA -.002 (.857) Wife power x Large MSA .057 (.052) Currently residing in home state of at least one spouse -.022 -.021* -.022* -.044* (.000) (.000) (.000) (.000) Head has previously moved for a job .004* .004* .004* .008 (.045) (.047) (.049) (.058) P .478 .533 .511 -.807 (.511) (.376) (.490) (.181) Wald coefficient tests (Prob > x2)' Power = Half power .054 Power = Husband power .707 .606 Power = Wife power .000 .004 Husband power = Wife power .001 .005 Husband power2 = Wife power2 .000 Wald test of independent equations (p = 0) .432 .258 .395 .032 Note.—For ease of interpretation, the marginal effects of the variables are presented. For the dummy variables, this is dyldx for a discrete change of dummy variable from 0 to 1. p > \ z | in parentheses. For p, the coefficient is given, with its standard error in parentheses. Standard errors are adjusted for clustering on household identifier. ;:" Coefficients are significant at the 95% confidence level or higher. This content downloaded from 147.251.185.127 on Mon. 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/tenns 488 C ompton/Pollak wife is a college graduate. Point estimates indicate that power couples are more likely to migrate than both part-power couples and low-power couples, although the Wald test shows that the difference between the coefficients on power couples and part-power couples is not statistically significant at a 95% confidence level. Holding all else constant, the probability of migration for a power couple is 2.2% higher than the probability of migration for a low-power couple. In column B, we split the part-power couples into those in which the husband is a college graduate and those in which the wife is a college graduate. We find that the migration propensity of power couples is indistinguishable from that of part-power couples in which the husband has a college degree and that the migration propensity of low-power couples is indistinguishable from that of part-power couples in which the wife has a college degree. When we ignore joint education specifications and control for the husband's education, we find no statistically significant effect of the wife's education on the probability of migration (col. C). Wald tests on the coefficients confirm that the effects of husband's and wife's education on the propensity to migrate are statistically different. These results suggest that it is not the joint education profile of a couple that affects migration, but only the husband's education. Finally, in column D we find that while the probability of migrating is lower for couples living in large MSAs compared to couples living in small MSAs, the interaction variables show that there is no differential impact of origin size by education. The coefficients on other variables are as expected. The variables included to capture past behavior—whether or not the couple is residing in the home state of at least one spouse and whether or not the head has previously moved for a job—remain constant in size and significance across model specifications. Living in the home state of at least one spouse reduces the probability of migrating by 2.2%, while prior migrants are 0.4% more likely to migrate compared to couples in which the head has never previously moved for a job. As noted above, these binary regressions are run simultaneously with a sample selection equation. The statistical significance of the variable p provides a test of sample selection bias. In all but column D, we can accept the null hypothesis that p = 0, that is, that the two equations are independent and controlling for sample selection is not providing any further information. Table 3 presents the results from the multinomial logit regressions, which confirm the crucial role of husband's education in migration, especially migration to large MSAs. These regressions do not control for sample selection because sample selection bias was significant in only one of the binary regressions. For columns A-C, then, this omission is unimportant. The first regression, presented in column A, again does not distinguish This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 489 between the two groups of part-power couples. The propensity to migrate to large MSAs is higher for both power and part-power couples, compared to low-power couples. The results for these regressions are presented as relative risk ratios, indicating that power couples are 2.8 times more likely to migrate to a large MSA compared to low-power couples, while part-power couples are 3.4 times more likely. Column B distinguishes between part-power couples in which the husband holds a college degree and part-power couples in which the wife holds a college degree. As in the binary regressions, this distinction is important. Compared to low-power couples, both power couples and part-power couples in which the husband has a college degree are more likely to migrate to medium MSAs and to large MSAs. Part-power couples in which the wife holds a college degree are indistinguishable from low-power couples in their propensity to migrate to cities of all sizes. Column C considers only the education of each spouse and not their joint profile. Again we find that couples in which the husband holds a college degree are more likely to migrate to a large MSA. The estimated effect of a wife's college education on the probability of migrating to a large MSA is negative, although the precision of this estimate is low. In all three regressions presented in columns A-C, couples living in large MSAs are less likely to migrate to small MSAs. Couples living in midsize MSAs are more likely to migrate to large MSAs (with slightly lower levels of significance). These trends indicate a general reluctance to move to a smaller metropolitan area and a general preference for moving to a larger MSA. The regression results presented in column D indicate that the education category of the couple has some impact on this preference for larger MSAs. Couples in which only the wife has a college degree are less likely to migrate from one large MSA to another large MSA and more likely to migrate from a large MSA to a midsize MSA, compared to low-power couples. However, the results for column D are questionable. Not only is sample selection bias an issue in this specification, but we also find a violation of the assumption of independence of irrelevant alternatives. Hausman test results suggest that the omission of the small or medium alternatives alters the results for the large alternative. We see again the robust significance of the past behavior variables. Residing in the home state of at least one spouse lowers the probability of migration to any sized destination. Interestingly, having previously moved for a job has no discernible effect on the probability of migrating to a medium or large MSA but increases the probability of migrating to a small MSA. This suggests that we may be capturing return migrants. Both the binary probit regressions and the multinomial logit regressions show that the probability of migration, and the migration to destinations of different sizes, depends not on the number of spouses holding a college This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms o m m ^ o go o k m m o ^t-o o o r\ o -^t- ^- <3 ^- <3 r\ o CO K O O i/l CO M O M O m h-v 0 2 3 Cl, S ' U CL, J> p *T3 * I B o -C CD "t» feD CD 0 I g i o u oj Cl, A 0 Cm > | £i 1, u a. u ^ ' ; O C O II ' <_m CL) CL) CL) CL) O S S S 2 w o o o £ "rt Ph Ph Ph >-Ih 11 <: *^ S3 £f o „ S 6 j CM to ^ _ U bC bJ} bD "§ " .S .9 .9 rt 9 "O "O g _2 _3 _3 2 S "u T> "u l3 S x x x 490 This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms O c~> ^- rx. o go C> \D \D M O O Q-< O LO O O h-v o m m ^ un iTl rH rH co ^ ri rH (S in go lo lo O O h-v o o o ^- ^-M fS N O \0 O O GO rH co fn o »0 \C rH CL) t-*-."-! rJ3 rt Cj o CD Dh CL) Dh CL) g o a o a o I Cl, rt Cm rt Dh G CO £ ^ ^ CL) S u ^ J IS CL) M 1 u ^ - Ixti ?j I a o h d bn ^ fS K K X 6 6 K d CL) Cm 1 A > -t3 C rt ^ „ ^ ii ii ^ n ^ ^ _ o CL) CL) " * * 3 3) 0 0 S3 "3 £h o a. * I, II < 0 II ^ ^ ^ ^ R „ .a » 0 0 „ S B ,, _,_, oj b/: bn bn II "o "o " .c .c .c sh rt B "3 "3 "3 u _a _Q S _5 _5 _5 ^ B o u2 u2 £ * * * p>_ X X 3www 491 This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 492 C ompton/Pollak degree (low-power, part-power, and power) but only on whether the husband holds a college degree. In a recent paper, Jurges (2006) performs similar analyses on German panel data and finds no evidence of a colo-cation effect. He distinguishes between "traditional" and "egalitarian" couples, defined by the husband's relative share of the housework on weekends. His results are consistent with ours for "traditional" couples— only the husband's education is a significant determinant of migration. For more "egalitarian" families, migration is as likely for those in which either spouse has a high level of education as for those in which both spouses have a high level of education. B. The Effect of Postgraduate Degrees and Urban Occupations on Migration To link the colocation hypothesis to the education profile of couples we must assume that, for couples with college degrees, occupational specialization makes colocation pressure more acute. But many college-educated individuals, women especially, are in occupations that are relatively portable and are not concentrated in large urban areas—the occupational specialization story is more plausible for an economist married to a lawyer than for a high school teacher married to a nurse. Fifty percent of power couples and 55% of power singles lived in large urban areas in 1990. Economists and lawyers tend to have advanced degrees and are relatively concentrated in large areas: 59% of economists and 55% of lawyers lived in large MSAs in 1990. In contrast, teachers and nurses are less likely to have advanced degrees and are not concentrated in large areas: 33% of teachers and 38% of registered nurses live in large MSAs in 1990.9 The migration behavior of couples in which both spouses have postgraduate degrees—call these "super-power couples"—or both spouses have occupations that are concentrated in large MSAs might be very different from the behavior of power couples without advanced degrees or urban occupations. We apply the same regression analyses as above to investigate whether colocation effects can be found for couples in which both spouses have occupations that are concentrated in large MSAs or both spouses have advanced degrees (tables 4 and 5). There are 1,778 observations in which at least one spouse has an advanced degree. These include 257 observations in which both spouses have advanced degrees, 986 observations in which only the husband has an advanced degree, and 535 observations in which only the wife has an advanced degree. Likewise, there are 3,493 observations in which at least one spouse has an urban occupation, including 1,740 observations in which both spouses have urban occupations, 4,716 in which only the husband has an urban occupation, and 1,282 obser- 9 Full results are available upon request. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Table 4 Binary Model: Probability of Migration Part II E F G H* Power: both spouses have college degree .022* .020* .019* .022* (.000) (.000) (.000) (.000) Husband power: only husband has college degree .020* .019* .019* .020* (.000) (.000) (.000) (.000) Wife power: only wife has college degree -.0002 .001 .001 .001 (.956) (.840) (.877) (.909) Percentage of husband occupation in large MSA .004 (.661) Percentage of wife occupation in large MSA -.009 (.514) Both spouses have urban occupations .004 (.281) Only husband has urban occupation .005* (.037) Only wife has urban occupation .005 (.225) At least one spouse has advanced degree .004 (.225) Both spouses have advanced degrees -.003 (.653) Only husband has advanced degree .005 (.266) Only wife has advanced degree .004 (.537) Origin—midsize MSA -.001 -.001 -.001 -.001 (.645) (.629) (.718) (.724) Origin—large MSA -.007* -.006* -.006* -.007* (.009) (.015) (.016) (.008) Currently residing in home state of at least one spouse -.022* -.021* -.022* -.035* (.000) (.000) (.000) (.000) Head has previously moved for a job .004* .004* .004* .005* (.041) (.044) (.042) (.047) P .585 .631 .569 (.240) (.172) (.650) Wald coefficient tests (Prob > x2)'- Power = Half power .785 .732 .972 .758 Power = Husband power .0001 .001 .002 .001 Power = Wife power .0004 .002 .002 .005 Wald test of independent equations (p = 0) .067 .010 .381 Note.—For ease of interpretation, the marginal effects of the variables are presented. For the dummy variables, this is dyldx for a discrete change of dummy variable from 0 to l._p> |z | in parentheses. For p, the coefficient is given, with its standard error in parentheses. Standard errors are adjusted for clustering on household identifier. Due to convergence problems, the regression in column H is run as a single probit, without correcting for sample selection. ;:" Coefficients are significant at the 95% confidence level or higher. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Table 5 Multinomial Logit Model: Probability of Migrating to (Small, Medium, Large) Destination Size Small Medium Large Small Medium Large Power: both spouses have college degree 1.798 (.012)* 2.574 (.000)* 2.960 (.000)* 1.753 (.016)* 2.441 (.000)* 2.914 (.000)* Husband power: only husband has college degree 1.414 (.128) 1.758 (.007)* 5.226 (.000)* 1.361 (.181) 1.629 (.018)* 5.178 (.000)* Wife power: only wife has college degree .532 (.165) 1.304 (.335) 1.034 (.944) .749 (.461) 1.286 (.359) 1.016 (.972) Percentage of husband occupation in large MSA .629 (.544) 1.469 (.562) .773 (.818) Percentage of wife occupation in large MSA 6.914 (.103) .232 (.113) .152 (.298) Both spouses have urban occupations 1.671 (.064) .924 (.751) .980 (.955) Only husband has urban occupation 1.144 (.527) 1.516 (.005)* .881 (.614) Only wife has urban occupation 1.409 (.324) 1.290 (.292) .952 (.893) Origin—midsize MSA .655 (.020)* 1.049 (.743) 1.637 (.046)* .674 (.030)* 1.025 (.867) 1.613 (.052)* Origin—large MSA .434 (.001)* .779 (.246) 1.135 (.705) .483 (.004)* .787 (.261) 1.122 (.730) Currently residing in home state of at least one spouse .276 (.000)* .299 (.000)* .329 (.000)* .270 (.000)* .298 (.000)* .329 (.000)* Head has previously moved for job 1.574 (.005)* .983 (.900) 1.171 (.440) 1.563 (.005)* .970 (.825) 1.170 (.445) Wald x2 924.39 939.20 Pseudo R1 .1083 .1093 Wald coefficient tests (Prob > x2)'- Power = Husband power .365 .010* .041* .3411 .074 .039* Power = Wife power .010* .023* .038* .040* .033* .037* Husband power = Wife power .043* .353 .001* .162 .462 .001* Hausman test for IIA: Excluding small .510 .924 .474 .974 Excluding midsize .222 .746 .184 .633 Excluding large .756 .292 .780 .215 G H Small Medium Large Small Medium Large Power: both spouses have college degree 1.505 (.111) 2.759 (.000)* 2.176 (.015)* 1.520 (.103) 2.758 (.000)* 2.304 (.008)* Husband power: only husband has college degree 1.275 (.308) 1.800 (.007)* 4.511 (.000)* 1.253 (.355) 1.811 (.006)* 4.400 (.000)* Wife power: only wife has college degree .728 (.411) 1.301 (.336) .953 (.920) .745 (.441) 1.280 (.364) .958 (.929) At least one spouse has advanced degree 1.559 (.087) .873 (.545) 1.715 (.079) Both spouses have advanced degrees 1.459 (.506) .784 (.638) .525 (.523) Only husband has advanced degree 1.662 (.076)* .831 (.466) 1.866 (.050)* Only wife has advanced degree 1.310 (.563) 1.047 (.899) 1.625 (.335) Origin—midsize MSA .682 (.034)* 1.037 (.804) 1.636 (.045)* .682 (.034)* 1.035 (.813) 1.626 (.048)* Origin—large MSA .487 (.005)* .765 (.210) 1.181 (.621) .486 (.005)* .762 (.203) 1.156 (.667) Currently residing in home state of at least one spouse .273 (.000)* .297 (.000)* .334 (.000)* .274 (.000)* .296 (.000)* .335 (.000)* Head has previously moved for job 1.590 (.004)* .961 (.773) 1.202 (.371) 1.598 (.004)* .957 (.749) 1.200 (.375) Wald x2 946.12 963.37 Pseudo R1 .1084 .1088 Wald coefficient tests (Prob > x2): Power = Husband power .535 .0585 .014* .481 .072 .030* Power = Wife power .088 .017* .111 .093 .013* .087 Husband power = Wife power .187 .324 .001* .221 .284 .002* Hausman test for IIA: Excluding small .673 .970 .706 .978 Excluding midsize .197 .725 .148 .750 Excluding large .792 .322 .855 .419 Note.—The coefficients are presented as relative risk ratios. For dummy variables, this is the change in the ratio of the probability of migrating to the indicated destination size and the probability of not migrating due to a discrete change in the variable from 0 to l.p> \z\ in parentheses. Standard errors are adjusted for clustering on household identifier. ;:" Coefficients are significant at the 95% confidence level or higher. 496 C ompton/Pollak vations in which only the wife has an urban occupation. We first include a measure of the urbanization of occupations—the percent of all individuals in the occupation that reside in MSAs greater than 2 million.10 This variable is added as a continuous variable (col. E) and as a dichotomous variable (col. F) that defines urban occupations as those in which at least 40% of individuals in the occupation live in large MSAs. This cutoff was chosen to correspond roughly to the proportion of all individuals living in large MSAs. In 1990, approximately 36% of all couples and 43% of singles lived in large urban areas. Urban occupations then are those that are more than proportionately located in large MSAs.11 We create categories to describe "urban" and "part-urban" couples by occupation in the same way we describe power and part-power couples by education— controlling for couples in which both spouses have urban occupations, where only the husband has an urban occupation and where only the wife has an urban occupation. In columns G and H we add indicators for postgraduate college degrees, again creating categories to describe the joint profile of the couple. We find few significant effects of urban occupation or advanced education in either the binary or the multinomial equations. Those variables that are significant confirm the importance of the husband's career in migration. Couples in which only the husband is employed in an urban occupation are more likely to migrate than couples in which neither has an urban occupation (table 4, col. F). The multinomial logit regression results (table 5) are difficult to interpret. Couples in which only the husband is employed in an urban occupation are more likely to migrate to a midsize MSA compared to couples in which neither spouse is employed in an urban occupation, while couples in which only the husband has an advanced degree are more likely to migrate to small and large MSAs. Thus even under a more stringent definition of who is most affected by occupation specialization we find no evidence that colocation pressure affects migration behavior. Our results here are consistent with a recent study by McKinnish (2006) in which the mobility rates of both spouses' occupations (measured by the percentage of workers in education/oc-cupation groups who have moved across state lines in the past 5 years) affect the probability of household migration, but the result is considerably stronger for the husband's occupation mobility. The coefficients for college education maintain statistical significance and have similar magnitude when we control for the concentration of occupations in large urban areas. The stability of the college education 10 Rates of occupation urbanization are calculated from the 1990 public use files of the Census (Ruggles et al. 2004). " The regression results are robust to alternative percentages used to define urban occupations. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 497 variables after controlling for occupation suggests that in addition to employment opportunities, amenities and/or returns to education are pulling college graduates to large MSAs.12 V. The Four Processes Underlying the Location Data The regression results presented above suggest that the colocation problem facing power couples does not manifest itself through differences in their migration behavior. If colocation pressure lies behind the concentration of power couples in large MSAs, it does not operate through couple migration. In this section, we argue that not only is colocation pressure not affecting the migration of power couples, but the migration of power couples is not likely to be important in affecting location patterns. The observed location trends of power couples may arise through differences by city size in power couple migration, power couple formation—either through marriage or increased educational attainment—and/ or power couple dissolution (divorce rates).13 Using the PSID and Census data we examine these dynamic processes. A. Migration The overall migration patterns of couples and unmarried individuals reported in table 6 are broadly consistent with the predictions of human capital theory.14 Higher skilled individuals are more mobile than lower skilled individuals, power couples are more likely to migrate than are low-and part-power couples, and power singles are more likely to migrate than are low-power singles. The observed patterns are not consistent with Mincer's hypothesis that dual-career couples are less mobile than single-career couples. The PSID transition data suggest that the migration of power couples is not the primary explanation for the increasing concentration of power couples in large MSAs—the migration patterns of power couples are not substantially different from those of power singles or part-power couples in which the husband has a college degree. There is, however, some evi- 12 This suggestion depends on the assumption that individuals do not change occupations: the insignificance of the urban occupation indicators may be caused by individuals changing occupations following migration so that their occupation premigration is not indicative of their occupation postmigration. 13 We ignore the dynamics of employment, although these are likely to differ by education and occupation. Basker (2003) develops a model of job search and migration that highlights the distinction between highly educated individuals, who tend to search first and then migrate, and less educated individuals, who tend to migrate without a job and then search. 14 Recent immigrants are underrepresented in the PSID sample. However the location patterns observed in the Census—those we are trying to explain—remain when immigrants are excluded from the Census sample. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Table 6 Migration Patterns, by Household Type Probability of Destination, Condi- Migrating tional on Mil grating 95% Confi- 95% Confi- dence Interval dence Interval Mean Lower Upper Mean Lower Upper Power couple: Small metropolitan area 7.34 5.06 9.62 23.3 17.0 29.6 Midsize metropolitan area 8.01 6.12 9.90 36.4 29.2 43.5 Large metropolitan area 6.20 4.86 7.54 40.3 33.0 47.7 Part-power couple—husband has BA: Small metropolitan area 5.67 3.79 7.55 20.0 13.2 26.8 Midsize metropolitan area 9.61 7.23 11.99 39.3 30.9 47.6 Large metropolitan area 4.83 3.45 6.21 40.7 32.3 49.1 Part-power couple—wife has BA: Small metropolitan area 4.34 2.31 6.36 25.5 12.6 38.5 Midsize metropolitan area 4.24 2.20 6.29 42.6 27.9 57.2 Large metropolitan area 3.20 1.70 4.69 31.9 18.1 45.7 Low-power couple: Small metropolitan area 3.22 2.74 3.71 36.9 32.7 41.2 Midsize metropolitan area 4.29 3.70 4.89 37.7 33.5 42.0 Large metropolitan area 2.69 2.26 3.11 25.3 21.5 29.2 Single power men: Small metropolitan area 17.43 12.36 22.51 23.7 16.3 31.0 Midsize metropolitan area 12.35 9.13 15.56 33.6 25.4 41.8 Large metropolitan area 5.66 4.09 7.24 42.7 34.2 51.3 Single power women: Small metropolitan area 16.85 12.43 21.26 22.2 15.9 28.5 Midsize metropolitan area 10.55 8.08 13.02 38.0 30.7 45.4 Large metropolitan area 6.24 4.80 7.68 39.8 32.4 47.2 Single low-power men: Small metropolitan area 8.14 6.88 9.41 33.2 28.4 37.9 Midsize metropolitan area 5.27 4.34 6.20 31.1 26.4 35.7 Large metropolitan area 4.65 4.01 5.30 35.8 30.9 40.6 Single low-power women: Small metropolitan area 6.32 5.37 7.28 28.0 23.9 32.1 Midsize metropolitan area 3.96 3.37 4.54 35.7 31.6 40.1 Large metropolitan area 2.59 2.25 2.93 36.3 31.9 40.7 Note.—Annual migration rates from 1970 to 1996 estimated from PSID sample. The estimates assume that the migration rates for each category remain constant for 1970-96, but when we split the PSID sample into the 3 decades, we find no significant differences in the migration rates of power couples between the 1970s and 1980s nor between the 1970s and 1990s. Couples include those who are observed married or cohabiting in year 1 and year 2. Singles are also limited to those who did not change marital status during the year. Large metropolitan areas are those with at least a 2 million population, midsize metropolitan areas have populations between 250,000 and 2 million, and smaller metropolitan areas and counties that are not contained in a metropolitan area are included in the final category. Power couples are those in which both spouses have completed at least 4 years of college or hold college degrees, part-power couples include only one spouse with a college education, low-power couples are those in which neither spouse has a college degree. Couples are limited to those in which the wife was 23-37 years of age and the husband was 25-39 years of age. Singles fall into the same age categories. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 499 dence of a revealed preference for large MSAs among power couples. The average rate of migration of power couples living in large MSAs is lower than the rates of those living in medium and small MSAs, although the differences are not statistically significant. Among migrants, power couples are more likely to migrate to medium and large MSAs than to small MSAs. These trends, however, are not limited to power couples and may simply reflect a general preference for large urban areas. For every group considered, migration rates of those in the large areas are lower than the rates of those in small areas, and this difference is much larger, and statistically significant, for single power men and women. An unmarried man or woman with a college degree is much more likely to migrate from a small MSA than a power couple. Conditional on migration, single power men and single power women are also more likely to migrate to large MSAs than small MSAs, as are part-power couples in which the husband has a college degree. Low-power couples are more likely to migrate to smaller areas than larger, and there is no statistically significant difference among the destination choices of part-power couples in which the wife has a college degree. The data suggest a revealed preference for large urban areas among college-educated singles and among all couples in which the husband is college educated. Again we see the need to distinguish between the two types of part-power couples, those in which the husband is college educated and those in which the wife is college educated. B. Education by City Size The proportion of power couples in large MSAs may have increased simply because individuals living in large MSAs have a greater incentive to invest in human capital than those living in small MSAs.15 The wage premium that is earned by workers in large urban areas may be a compensating differential to offset higher living costs and urban disamenities but must also be due in part to higher productivity of workers in MSAs. Without higher productivity, it is difficult to explain why firms are willing to locate in these high-wage areas. The causes of this higher productivity may lie in the ability of MSAs to attract more able workers or because MSAs create more productive workers by inducing human capital accumulation and labor-market matching. Glaeser and Mare (2001) investigate this question using a variety of data sources and argue that the urban wage premium is due to a combination of wage-level effects and wage-growth effects. They find that recent migrants to MSAs do experience real wage gains but that the wage premium is highest for long-term residents of MSAs. Thus many of the urban wage gains accrue over 15 Glaeser (1999) suggests that urban density may increase interactions with highly skilled role models and facilitate coordination, thus increasing the demand for human capital. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 500 C ompton/Pollak Table 7 Percent of Population Who Have a High School Diploma but Not a College Degree Who Are Enrolled in Credit Education % Change % Change 1980 1990 2000 1980-90 1990-2000 Married men: Non-MSA or MSA fewer than 1 million 5.6 6.3 6.6 11.8 5.4 MSA population between 1 and 2 million 8.5 8.3 8.9 -2.1 6.6 MSA population over 2 million 8.2 9.0 8.2 9.8 -8.6 Married women: Non-MSA or MSA fewer than 1 million 5.0 8.0 7.8 58.8 -2.6 MSA population between 1 and 2 million 6.8 9.1 8.5 34.1 -6.8 MSA population over 2 million 7.1 10.0 9.5 40.5 -4.9 Unmarried men: Non-MSA or MSA fewer than 1 million 10.0 9.7 12.2 -2.9 25.4 MSA population between 1 and 2 million 12.2 11.4 14.3 -6.3 25.1 MSA population over 2 million 12.9 13.3 16.1 3.3 20.5 Unmarried women: Non-MSA or MSA fewer than 1 million 11.5 13.9 15.3 21.5 9.7 MSA population between 1 and 2 million 14.6 14.9 16.7 1.8 12.6 MSA population over 2 million 14.7 16.8 19.1 13.9 13.6 Note.—Calculated by authors using the Census mtegra ted public use census samples (Ruggles et al. 2004). time as urban workers accumulate more human capital and benefit from the better coordination of urban labor markets. Table 7, which shows the proportion of high school graduates without college degrees currently enrolled in higher education, is consistent with this hypothesis. These data are from the integrated public use samples (IPUMS) of the U.S. Census for 1980, 1990, and 2000. The IPUMS samples are l-in-100 national random samples of the U.S. population created by the Census Bureau as part of each decennial enumeration (Ruggles et al. 2004). The individual level records of the IPUMS samples include information on age, location, marital status, education, and current education enrollment. The samples also allow the linkage of spousal records enabling us to determine the joint profile of married couples. With a few minor exceptions, the probability of being enrolled in education increases with population size for all groups in each year.16 For example, in 1990, among married men with high school diplomas, 6.3% of those living in small MSAs were enrolled in credit courses compared to 9.0% of those living in the largest MSAs, those with populations over 5 million. The differences across city size indicate that the formation of power couples and power singles through educational attainment is more likely 16 The categories of small, midsize, and large MSAs are different for this table than for the rest of the analyses because the public use files for 2000 do not identify places with populations less than 400,000. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 501 in larger MSAs than in smaller MSAs. Holding marriage and migration constant, an increasing concentration of power couples in large MSAs would occur if the proportion of married individuals in large MSAs who are enrolled in credit education is higher than the proportion of married individuals in smaller areas enrolled in credit education. For both married men and married women, the gap in enrollment rates between the large and small MSAs grew or remained constant between 1980 and 1990 and then fell between 1990 and 2000. C. Marriage and Divorce by City Size The relationship between city size and marriage is not straightforward. Edlund (2005) suggests that since large cities attract highly skilled individuals they also attract individuals (i.e., women) who seek to marry highly skilled individuals. Drewianka (2003) argues that marriage rates are affected not by the size of the city itself but by the percentage of the adult population that is single. The effect is twofold. With a relatively large single population, the probability of meeting a potential spouse whose quality exceeds a specified level in a given period of time is higher because people meet potential matches at a faster rate. But the large pool of potential matches also causes marriage market participants to increase their reservation quality level. Drewianka's empirical results suggest that the reservation quality effect dominates such that a 10% increase in the single population lowers the hazard rate for entering marriage between 7% and 10%. Gould and Paserman (2003) find a strong negative relationship between women's marriage rates and the inequality of male wages within an urban area. They argue that higher levels of male wage inequality can explain 25% of the decline in marriage rates between 1970 and 1990. Taken together, these papers suggest a finding similar to that of Robst and McGoldrick (1996) on the relationship between city size and labor market success: it is not the size of the market that is important, but its composition. The human capital model of migration suggests a link between migration, city size, and marital instability. Increased labor force attachment of wives makes it more likely that the location preferences of spouses will differ, which increases marital instability. This suggests two hypotheses. First, since power couples are more likely to experience colocation problems, they may be more likely to experience marital instability than part-power and low-power couples. Second, since colocation problems are less likely to occur in large labor markets, power couples living in small cities may have more marital breakups than those living in larger cities. Colocation problems may affect the location patterns observable in census data by increasing the relative stability of marriages in larger employment markets rather than by inducing migration. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 502 C ompton/Pollak Table 8 Marriage Patterns, by Size of Area Probability of Marrying 95% Confidence Interval Conditional on Marrying, % with College-Educated Spouse 95% Confidence Interval Mean Lower Upper Mean Lower Upper Unmarried men with less than BA: Small metropolitan area 14.0 12.5 15.5 9.9 6.3 13.6 Midsize metropolitan area 12.8 11.5 14.1 7.9 4.8 10.9 Large metropolitan area 11.5 10.6 12.4 10.9 8.1 13.8 Unmarried men with BA or more: Small metropolitan area 14.2 9.9 18.5 48.8 32.8 64.8 Midsize metropolitan area 14.0 10.9 17.2 51.4 39.4 63.4 Large metropolitan area 12.6 10.5 14.7 61.1 52.7 69.3 Unmarried women with less than BA: Small metropolitan area 11.8 10.6 13.0 7.6 4.6 10.7 Midsize metropolitan area 8.6 7.8 9.4 11.8 8.4 15.2 Large metropolitan area 7.6 7.1 8.2 11.7 9.1 14.4 Unmarried women with BA or more: Small metropolitan area 11.4 7.9 15.0 61.0 45.4 76.6 Midsize metropolitan area 10.9 8.5 13.3 59.3 48.7 69.9 Large metropolitan area 10.8 9.1 12.5 64.1 56.5 71.7 Note.—Annual rates from 1970-96 PSID data. Unmarried indiv: [duals are defined as becoming married if they are observed to be legally married or living common-law m year 2. Mincer (1978) notes that expectations of marital instability reduce the incentives for either spouse to become a tied-mover or tied-stayer, further amplifying marital instability. If we allow for expectations of marital instability, modeling migration as a response to maximizing net family benefits may be less satisfactory than modeling migration as a potentially inefficient bargaining solution to the colocation problem; these issues are more fully explored in Lundberg and Pollak (2003). The results reported in table 8 suggest that the formation of power and part-power couples through marriage is more likely to occur in large urban areas compared to smaller urban areas. Table 8 shows the probability that an unmarried individual in year 1 will be married in year 2, conditional on the size of the year 2 location.17 The probability of marriage decreases by city size for low-power singles, but there is no statistically significant difference in marriage rates among power singles of either sex. That is, power singles are just as likely to marry if they live in a large 17 These are higher than might be expected for the total population because individuals in the sample must have been the head or spouse of a household at some point during the survey years, but this should not bias the comparison across city size. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 503 Table 9 Probability of Marriage Breakup, by Size of Area 95% Confidence Interval Means Lower Upper Power couple: Small metropolitan area Midsize metropolitan area Large metropolitan area Part-power couple: Small metropolitan area Midsize metropolitan area Large metropolitan area Low-power couple: Small metropolitan area Midsize metropolitan area Large metropolitan area 1.95 .75 3.14 1.60 .74 2.47 1.27 .65 1.89 2.50 1.53 3.47 2.61 1.62 3.60 1.81 1.13 2.49 3.14 2.67 3.62 3.77 3.22 4.32 4.01 3.50 4.52 Note.—Annual rates calculated from 1970-96 PSID. Couples are defined as having a marriage breakup if they are no longer married or cohabiting m the second year. MSA or a small MSA, but low-power singles are more likely to marry in a small MSA. The patterns of assortative mating also suggest that power couples are more likely to be created in large MSAs than in small MSAs. While there is no clear difference in assortative mating patterns for women by city size, there is a more distinct relationship between assortative mating patterns and city size for men: college-educated men living in small and midsize MSAs are as likely to marry a college-educated spouse as a non-college-educated spouse, that is, those that marry are equally likely to create a part-power couple as a power couple. College-educated men living in large MSAs, however, are more likely to marry college-educated women than non-college-educated women, that is, conditional on marrying, the probability of creating a power couple from that marriage is greater than 50%. Finally, we consider the probability that a couple will divorce by city size. While the confidence intervals on these data are too large to make any definitive statements, the differences in the means shown in table 9 suggest that power couples are less likely to divorce than low-power couples. Furthermore, low-power couples are more likely to divorce if they live in large metropolitan areas than in small MSAs, whereas power couples are more likely to divorce if they live in a small city. These estimates, while not precise, suggest that the stability of power couple marriages may be higher in large MSAs. VI. The Decline in Power Couple Concentration in 2000 Finally, we extend the time span of the census-based location analysis of Costa and Khan (2000) by using data from the 2000 census, data This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 504 C ompton/Pollak Table 10 Proportion of Household Groups Living in MSAs Greater than 2 Million 1970 1980 1990 2000 Power couple 40.1 43.3 49.7 46.9 Part-power couple 37.1 37.3 40.1 38.2 Part-power, husband has college degree 38.6 40.1 42.6 41.1 Part-power, wife has college degree 35.3 34.9 37.9 37.1 Low power couple 31.2 32.3 31.3 33.5 Single power men 51.3 53.4 55.1 55.5 Single power women 48.5 53.4 54.2 54.6 Single low-power men 38.7 42.3 39.8 40.2 Single low-power women 39.9 44.2 40.8 41.1 All couples in age range 32.9 35.0 35.7 37.3 All single men in age range 40.8 45.4 43.2 44.0 All single women in age range 41.1 46.3 43.9 44.9 Note.—Power couples are those m which both spouses have completed at least 4 years of college or hold college degrees, part-power couples include those couples m which only one spouse has completed at least 4 years of college or holds a college degree, and low-power couples are those m which neither spouse has 4 years of college or a college degree. Couples are limited to legally married couples residing m the same household. For married couples, the sample is limited to those m which the wife was 23-37 years of age and the husband was 25-39 years of age. Singles fall into the same age categories. Calculations by authors using the census integrated public use census samples (Ruggles et al. 2004). unavailable to Costa and Kahn. Table 10 shows the proportions of power, part-power, and low-power couples living in large MSAs in each of the 4 census years (1970, 1980, 1990, 2000), using the 1990 MSA definitions for all years. Since our regression results clearly show that the gender of the college-educated spouse matters for migration behavior, we separate part-power couples by which spouse holds a college degree. The data show that the proportions of couples living in large MSAs are lower when the wife is more highly educated than when the husband is more highly educated. The data also show that between 1970 and 1990, the proportion of power couples living in large MSAs increased dramatically, as noted by Costa and Kahn. However, the rising trend in concentration did not continue to 2000. Between 1990 and 2000 the proportion of power couples and part-power couples living in large MSAs declined, while the proportion of low-power couples living in large MSAs increased. Can either of the two hypotheses put forth to explain the rising concentration of power couples in large MSAs also explain the fall? Trends in women's labor supply did not reverse during this decade: labor force participation of married women remained steady between 1990 and 2000, and the proportion of women in specialized professional occupations increased during the decade.18 This suggests that the colocation pressure 18 The labor force participation rate of married women aged 23-55 increased from 43.5% in 1970 to 58.0% in 1980, 70.8% in 1990, and 70.9% in 2000. The labor force participation rates for most subgroups—married women with and This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 505 facing power couples did not weaken over this decade. But neither do we see evidence of a weakening of the relative attractiveness of large MSAs for the college educated. The fall in the proportion of power couples living in large MSAs was not mimicked in the category of power singles, and the proportions of both college-educated single men and college-educated single women living in large MSAs increased modestly between 1990 and 2000 (from 55.1 to 55.5 and from 54.2 to 54.6, respectively). VII. Conclusion Using census data, Costa and Kahn (2000) show that between 1970 and 1990 couples in which both spouses were college graduates (power couples) were increasingly likely to be located in the largest metropolitan areas. Costa and Kahn argued that these changes in location patterns were driven by the migration of power couples trying to solve their colocation problem—a plausible-sounding hypothesis to anyone familiar with academic job markets. In this article we investigate the dynamic processes underlying the changes in location patterns found by Costa and Kahn. Using data from the Panel Study of Income Dynamics, we find that power couples are not more likely to migrate to the largest metropolitan areas and are no less likely than other couples to migrate from such areas once they are there. The dynamic processes that appear to contribute most to the observed trends in location patterns include (i) the migration of power singles to large metropolitan areas, (ii) differences in rates of marriage and divorce of men and women with and without college degrees in cities of various sizes, and (iii) differences in the likelihood that individuals acquire additional education in cities of various sizes. The observed trends in location patterns are primarily due to differences in the rates at which power couples form and dissolve in cities of various sizes rather than to the migration of power couples to the largest metropolitan areas. Our analysis of the PSID migration data hinges on our analysis of the behavior of part-power couples. Unlike Costa and Kahn, who group couples by the number of spouses with a college degree—power, part-power, and low-power couples—we decompose part-power couples into two subcategories: those in which only the husband has a college degree and those in which only the wife has a college degree. We find that couple migration patterns are determined by the husband's education profile, not by the couple's joint education profile. The migration patterns of couples in which both spouses have college degrees look like those of part-power without young children, married women with and without college education— follow the same pattern, with a slight decline in the LFP rates between 1990 and 2000 for most groups (less than 1 percentage point fall). Calculations by authors using the census integrated public use census samples (Ruggles et al. 2004). This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 506 C ompton/Pollak couples in which only the husband has a college degree. The migration patterns of part-power couples in which only the wife has a college degree look like those of low-power couples. Our conclusion that trends in location patterns are not generated by power couples migrating to solve their colocation problem is strengthened by data from the 2000 census. The 2000 census shows a decline in power couple concentration between 1990 and 2000, a decline that is difficult to reconcile with the colocation hypothesis. The plausible-sounding hypothesis based on academic job markets is misleading because the typical power couple is not one PhD married to another PhD but a high school teacher married to a nurse. These more typical power couples do not need to locate in New York or Los Angeles to solve their colocation problem. But even when we look at super-power couples we find no support for the colocation hypothesis. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Appendix A Table Al Large MSAs (Population over 2 Million) 1970 1980 1990 2000 New York-Northern New Jersey-Long Island, NY-NJ-CT-PA 18,071,522 17,412,203 17,953,372 19,451,757 Los Angeles-Anaheim-Riverside, CA 9,980,859 11,497,549 14,531,529 16,036,587 Chicago-Gary-Lake County, IL-IN-WI San Francisco-Oakland-San Jose, CA Philadelphia-Wilmington-Trenton, PA-NJ-DE-MD Detroit-Ann Arbor, MI Washington, DC-MD-VA-WV Dallas-Fort Worth, TX Boston-Lawrence-Salem-Lowell-Broc- ton, MA-NH-ME-CT Houston-Galveston-Brazoria, TX Miami-Fort Lauderdale, FL Atlanta, GA Cleveland-Akron-Lorain, OH Seattle-Tacoma, WA San Diego, CA Minneapolis-St. Paul, MN-WI St. Louis, MO-IL Baltimore, MD Pittsburgh-Beaver Valley, PA Phoenix, AZ Tampa-St. Petersburg-Clearwater, FL Denver-Boulder, CO MSAs are defined as "large" if their population is greater than 2 million m 1990. The analysis here uses the MSA definitions, i.e., county components, from the 1990 definitions. 7,778,948 7,937,290 8,065,633 8,783,199 4,754,366 5,367,900 6,253,311 6,873,645 5,749,093 5,680,509 5,899,345 5,661,399 4,788,369 4,752,764 4,665,236 5,031,963 3,040,307 3,250,921 3,923,574 4,739,999 2,351,568 2,930,568 3,885,415 4,909,523 3,709,642 3,662,888 3,783,817 4,440,881 2,169,128 3,099,942 3,711,043 4,493,741 1,887,892 2,643,766 3,192,582 3,711,102 1,684,200 2,138,136 2,833,511 3,857,097 2,999,811 2,834,062 2,759,823 2,910,616 1,836,949 2,093,285 2,559,164 3,023,741 1,357,854 1,861,846 2,498,016 2,820,844 1,981,951 2,137,133 2,464,124 2,872,109 2,429,376 2,376,968 2,444,099 2,569,029 2,089,438 2,199,497 2,382,172 2,491,254 2,556,029 2,423,311 2,242,798 2,331,336 971,228 1,509,175 2,122,101 3,013,696 1,105,553 1,613,600 2,067,959 2,278,169 1,238,273 1,618,461 1,848,319 2,252,103 This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Appendix B o s- 13 h n Table Bl Variable Means Nonmigrants Migrants Full Sample Mean SD Mean SD Mean SD Sample size 21,317 638 21,955 Education variables: Both spouses have college degree .110 .312 .223 .416 .113 .316 Only one spouse has a college degree .151 .358 .215 .411 .153 .360 Only husband has a college degree .089 .285 .169 .375 .091 .288 Only wife has a college degree .062 .241 .045 .208 .062 .240 At least one spouse has advanced degree .079 .269 .161 .368 .081 .273 Both spouses have advanced degrees .012 .107 .017 .130 .012 .108 Only husband has an advanced degree .043 .203 .108 .311 .045 .207 Only wife has an advanced degree .024 .153 .036 .186 .009 .096 Occupation variables: Percent of husband's occupation in large MSA .348 .101 .349 .114 .348 .101 Percent of wife's occupation in large MSA .316 .103 .303 .109 .316 .103 Both have urban occupation .079 .270 .083 .277 .079 .270 Only husband in urban occupation .213 .409 .276 .448 .215 .410 Only wife in urban occupation .058 .234 .069 .254 .584 .234 Size of origin variables: Origin—midsize MSA .422 .494 .475 .500 .423 .494 Origin—large MSA .256 .436 .208 .407 .254 .435 Propensity to migrate variables: Either spouse living in home state Head moved previously for a job Labor force attachment: Number of months duration, husband's job Number of months duration, wife's job Husband unemployed Wife unemployed Husband not in the labor force Wife not in the labor force Total family income ($) Basic demographics: Husband aged 25-29 Husband aged 30-34 Wife aged 23-27 Wife aged 28-32 Child aged 0-5 present in household School aged child present in household Homeowner Either spouse nonwhite Origin MSA attributes: Average growth rate of origin MSA Average housing value in origin MSA Distance to nearest large MSA (pop. > 2 million) .798 .401 .448 .262 .440 .378 60.31 59.47 49.03 31.05 44.89 15.68 .044 .205 .050 .029 .169 .042 .021 .145 .044 .332 .471 .408 44,200 29,206 44,706 .307 .461 .422 .398 .489 .364 .302 .459 .386 .407 .491 .384 .563 .496 .539 .310 .462 .343 .656 .475 .455 .295 .456 .218 -.076 .124 -.109 108.74 74.57 120.37 148.43 170.40 207.28 .498 .788 .409 .485 .265 .441 54.56 59.98 59.36 27.73 30.60 44.56 .218 .044 .205 .201 .030 .170 .205 .022 .147 .492 .338 .472 27,866 44,214 29,167 .494 .310 .463 .481 .397 .489 .487 .304 .460 .487 .406 .491 .499 .562 .496 .475 .311 .463 .498 .650 .477 .413 .292 .455 .126 -.077 .124 82.27 109.09 74.83 343.28 150.14 178.07 510 C ompton/Pollak References Basker, Emek. 2003. Education, job search and migration. Working Paper no. 02-16, Department of Economics, University of Missouri-Columbia. Bielby, William T., and Denise D. Bielby. 1992. I will follow him: Family ties, gender-role beliefs, and reluctance to relocate for a better job. American Journal of Sociology 97, no. 5:1241-67. Bird, Gerald A., and Gloria W. Bird. 1985. Determinants of mobility in two-earner families: Does the wife's income count? Journal of Marriage and the Family 47, no. 3:753-58. Cooke, Thomas, and Adrain Bailey. 1996. Family migration and the employment of married women and men. Economic Geography 72, no. 1: 38-48. Costa, Dora L., and Matthew E. Kahn. 2000. Power couples: Changes in the locational choice of the college educated, 1940-1990. Quarterly Journal of Economics 115, no. 4:1287-1315. Drewianka, Scott. 2003. Estimating social effects in matching markets: Externalities in spousal search. Review of Economics and Statistics 85, no. 2:409-23. Duncan, R. Paul, and Carolyn Cummings Perrucci. 1976. Dual occupation families and migration. American Sociological Review 41, no. 2:252-61. Edlund, Lena. 2005. Sex and the city. Scandinavian Journal of Economics 107, no. 1:25-44. Eklof, Jan A., and Sune Karlsson. 1999. Testing and correcting for sample selection bias in discrete choice contingent valuation studies. SSE/EFI Working Paper Series in Economics and Finance no. 171, Stockholm School of Economics. Frank, Robert H. 1978. Why women earn less: The theory and estimation of differential overqualification. American Economic Review 68, no. 3: 360-73. Gardner, Jonathan, Gaelle Pierre, and Andrew J. Oswald. 2001. Moving for job reasons. Unpublished manuscript, Department of Economics, Warwick University. Glaeser, Edward L. 1999. Learning in MSAs. Journal of Urban Economics 46, no. 2:254-77. Glaeser, Edward L., and David C. Mare. 2001. Cities and skills. Journal of Labor Economics 19, no. 2:316-42. Gould, Eric D., and M. Daniele Paserman. 2003. Waiting for Mr. Right: Rising inequality and declining marriage rates. Journal of Urban Economics 53, no. 2:257-81. Green, Anne E. 1997. A question of compromise? Case study evidence on the location and mobility strategies of dual career households. Regional Studies 31, no. 7:641-57. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms Power Couples 511 Greenwood, Michael J. 1997. Internal migration in developed countries. In Handbook of population and family economics, ed. M. R. Rosenzweig and O. Stark. Amsterdam: Elsevier. Jacobsen, Joyce, and Laurence M. Levin. 1997. Marriage and migration: Comparing gains and losses from migration for couples and singles. Social Science Quarterly 78, no. 3:688-709. Jurges, Hendrik. 2006. Gender ideology, division of housework, and the geographic mobility of families. Review of Economics of the Household 4, no. 4:299-323. Lundberg, Shelly, and Robert A. Pollak. 2003. Efficiency in marriage. Review of Economics of the Household 1, no. 3:153-68. McKinnish, Terra. 2006. Spousal mobility and earnings. Unpublished manuscript, Department of Economics, University of Colorado, Boulder. Mincer, Jacob. 1978. Family migration decisions. Journal of Political Economy 86, no. 5:749-74. Morrison, Donna R., and Daniel T. Lichter. 1988. Family migration and female employment: The problem of underemployment among migrant married women. Journal of Marriage and the Family 50, no. 1:161-72. Nivalainen, Satu. 2004. Determinants of family migration: Short moves vs. long moves. Journal of Population Economics 17, no. 1:157-75. Polachek, Solomon W., and F. W. Horvath. 1977. A life cycle approach to migration: Analysis of the perspicacious peregrinator. In Research in labor economics, vol. 1, ed. R. Ehrenberg. Greenwich CT: JAI Press. Robst, John, and KimMarie McGoldrick. 1996. Gender differences in overeducation: A test of the theory of differential overqualification. American Economic Review 86, no. 2:280-84. Ruggles, Steven, Matthew Sobek, Trent Alexander, Catherine A. Fitch, Ronald Goeken, Patricia Kelly Hall, Miriam King, and Chad Ronnan-der. 2004. Integrated public use microdata series: Version 3.0. Minneapolis: Minnesota Population Center. Sandell, Steven H. 1977. Women and the economics of family migration. Review of Economics and Statistics 59, no. 4:406-14. Shihadeh, Edward S. 1991. The prevalence of husband-centered migration: Employment consequences for married mothers. Journal of Marriage and the Family 53, no. 2:432-44. Sjaastad, Larry A. 1962. The costs and returns of human migration. Journal of Political Economy 70, no. 5, pt. 2:80-93. Spitze, Glenna D. 1984. The effect of family migration of wives' employment: How long does it last? Social Science Quarterly 65, no. 1: 21-36. University of Michigan Institute for Social Research. 2002. 1969-1993 Panel Study of Income Dynamics, http://www.isr.umich.edu/src/psid. U.S. Census Bureau. 1988. County and city data books. Washington, DC. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms 512 C ompton/Pollak U.S. Census Bureau. 1993. County and city data books. Washington, DC. van de Ven, Wynand P. M. M., and Bernard M. S. Van Praag. 1981. The demand for deductibles in private health insurance. Journal of Econometrics 17, no. 2:229-52. Vella, Francis. 1992. Simple tests for sample selection bias in censored and discrete choice models. Journal of Applied Econometrics 7, no. 4:413-21. -. 1993. A simple estimator for simultaneous models with censored endogenous regressors. International Economic Review 34, no. 2: 441-57. -. 1998. Estimating models with sample selection bias: A survey. Journal of Human Resources 33, no. 1:127-69. This content downloaded from 147.251.185.127 onMon, 19 Mar 2018 14:41:38 UTC All use subject to http://about.jstor.org/terms