Transcript

JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010, pp. 615–634

GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNERCOUPLES—THE RELATIONSHIP BETWEEN CHANGE OF JOBAND CHANGE OF RESIDENCE*

Mette DedingThe Danish National Centre for Social Research, Herluf Trolles Gade 11, DK-1052 Copenhagen K,Denmark. E-mail: [email protected]

Trine FilgesThe Danish National Centre for Social Research, Herluf Trolles Gade 11, DK-1052 Copenhagen K,Denmark. E-mail: [email protected]

ABSTRACT. We analyze the relationship between geographical residence and job mobility for Dan-ish dual-earner couples. We estimate the probability of moving residence and changing job, takingthe interdependence between the events into account. The results point to the importance of address-ing the interrelationship between residence and jobs. Furthermore, the change of residence mattermore than change of job and women respond relatively more to changes in their husbands’ job region.The findings imply that mobility promoting initiatives must focus on families rather than individu-als and recognize that for most families the choice of residence location dominates the choice of joblocation.

1. INTRODUCTION

Policy analysis typically focuses on labor market developments at the national level.Yet, in many countries, there are persistent regional disparities in employment per-formance (OECD, 2000) and Denmark is no exception.1 Denmark is a small countrywith only 5.4 million inhabitants, and dividing the country into 14 regions, the lowestunemployment rate in 1999 was 4.0 percent and the highest unemployment rate was9.8 percent. Hence, labor shortages in some regions coexist with continuously high unem-ployment in others (National Labor Market Authority, 2003). Part of a successful strategyfor reducing overall unemployment is to tackle obstacles to geographical labor mobility,defined as moving to another geographical region within the country, as well as promot-ing local job creation. Consequently, geographical mobility is an important component of aflexible labor force, and may play a self-equilibrating role in reducing regional disparitiesconcerning employment performance. The willingness to be geographically mobile mayshorten individual unemployment spells considerably and help providing the necessarylabor supply in all regions.

Internationally, geographical mobility is found to vary considerably between coun-tries. The Organisation for Economic Co-operation and Development (OECD) has pre-sented a cross-country comparison of residence migration rates, percent of residents whomigrate each year, within 17 countries (OECD, 2000). This cross-country comparison

*The authors thank Jos van Ommeren, Konstantinos Tatsiramos, Olle Westerlund, and the anony-mous referees for very useful comments. The usual disclaimer applies.

Received October 2006; revised: September 2008; accepted March 2009.1This is documented in Deding and Filges (2004a).

C© 2010, Wiley Periodicals, Inc. DOI: 10.1111/j.1467-9787.2010.00663.x

615

616 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

reveals that the level of internal migration is relatively high in the United States,Japan, Canada, Australia, New Zealand, and the United Kingdom2; whereas internalmigration rates in continental European countries are considerably lower.3 The gen-eral conclusion is that within continental European countries, regional unemploymentand wage differentials persist because of the relatively low geographical mobility. Whencomparing the level of residential mobility in Denmark to other OECD countries, onehas to be cautious because of the modest size of the country. However, according tothe Danish Economic Council (2002), it is safe to conclude that Danish mobility isnot higher than in other continental European countries: If Denmark is divided into5 regions; the number of inhabitants in each region (approximately 0.9 millions) aswell as the migration rate between regions (1.27 percent) is similar to the case ofBelgium.4

The vast majority of studies on geographic mobility focus on individual mobility (fora survey see Greenwood, 1985). Other studies include the family as the decision-makingunit. For example, Sandell (1977) analyzes the effect of migration on female labor forceparticipation. He uses a theoretical model in which families move if the gain to moving(the sum of discounted earnings in the new location) is greater than the gain to staying(the sum of discounted earnings in the current location). The same concept is used byMincer (1978), who analyzes the effect of family ties on migration. Mincer derives atheoretical model in which families move if the sum of their gains (positive or negative)is greater than the cost of moving. Noting that single individuals only consider their ownpersonal gains of moving location, Mincer points out that married couples’ decisions aredifferent from that of single persons. Some families will move even if there is a personalloss to one individual or some families may remain in the original location even if therewould be a gain to one individual from moving. Mincer refers to these people as eithertied-movers—those who would not have moved, had they been single—or tied-stayers—those who would have moved, had they not been married. Also Gardner, Pierre, andOswald (2001) and Nivalainen (2004) include the family as the decision-making unit. Aconclusion from these studies is that the mobility decision for families is very differentfrom that of individuals and, consequently, the entire family must be taken into accountwhen studying mobility especially for two-earner families. On the other hand, the existingstudies primarily focus on residential mobility, and the relationship between job mobilityand residence mobility is not explicitly addressed.

In this paper we analyze the job and residence mobility of households and explicitlyaddress the relationship between these two types of moves. We will not, as is usual donein these types of studies, define moves based on an administrative division of the country,e.g., municipalities or counties. Instead we use a functional division, i.e., between areaswhere the actual location and commuting behavior of the labor force characterizes it as acoherent local labor market. Such an area can be defined as a regional commuting area:a group of municipalities where people commute between municipalities within the area,but only seldom between areas. Several studies of the Danish labor market have indi-cated that although Denmark is a small country, it is characterized by these nonoverlap-ping regional commuting areas (see the Ministry of the Environment, 2001; or Andersen,2000). When dividing the country into nonoverlapping commuting areas, it is necessary todistinguish between intra- and interregional moves. A job move over a long distance, for

2Internal migration is defined as migration between larger regions within the countries and variesbetween 1.93 and 2.45 percent.

3The lowest level is 0.50 percent in Italy.4Alternatively, if we divide Denmark into two regions with app. 2.7 million inhabitants in each,

Denmark would be comparable to the case of Italy with a migration rate of 0.5 percent.

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 617

example between nonoverlapping commuting areas, has different consequences from ajob move within the same commuting area. For example, Zax (1991) and Zax and Kain(1991) show that job and residential moves may act as substitutes following an unexpectedchange in commuting distance related to intraregional moves. On the other hand, job andresidential moves more likely act as complements when interregional moves are consid-ered, as the distance between job and residence becomes far too long. In fact, Bartel (1979)shows that job mobility and migration (defined as a residence move to another state orcounty in the United States) are closely related. She shows that an analysis of migrationthat ignores the link between the decision to migrate and the probability of a job separa-tion may fail to understand the role played by socio-economic variables in the migrationprocess.

The purpose of this paper is to analyze geographical mobility from a labor marketperspective. We therefore only consider job and residence moves between commutingareas, i.e., only interregional moves. The reason for this is our interest in the relationshipbetween job and residence moves. Many intraregional moves are typical life-cycle moves—e.g., adjusting to changes in family size—while the interregional moves more likely arerelated to job considerations.5

We analyze the interdependence between interregional residence moves and inter-regional job moves for dual-earner households and the determinants of the mobility de-cisions. The case of dual-earner households deserves special attention because the twowage earners share a dwelling but have different workplaces, a situation which adds tothe complexity of their residence and job location choices. This issue is especially relevantto study in Denmark, because female labor force participation rates are very high—in1999, the participation rates were 88 percent for men and 81 percent for women, and theshare of dual-earners is about 70 percent.6 In addition, Denmark is a relatively egalitar-ian country in all respects, see Fuwa (2004). Thus, in the Danish case it is too simplisticto look at job or residential mobility for one spouse only. It is necessary to allow both thehusband’s and the wife’s job considerations to influence the family migration decision.More explicitly, we study the determinants of three mobility events: the couple movingresidence to another region, the husband changing job to another region, and the wifechanging job to a new region.

The theoretical foundation for the empirical analysis is search theory. We find thatsearch theory is relevant in this case because it is based on the idea of agents movingbetween different states and thus fits the dynamic structure and complexity of the prob-lem. Also, it is closely connected to the family migration hypothesis of moving when gainsexceed costs. Furthermore, uncertainty is explicitly treated in this framework. We assumethat job mobility is an individual decision taken by either the husband or the wife, but thatthe consequences for the whole household are part of the basis for decision. In line withthe studies on family migration mentioned above, we assume that moving residence is acouple-decision taken at the household level. We allow the three moving decisions to beinterdependent, and this endogeneity of the regressors is taken into account by applyingthe AGLS (Amemiya’s generalized least squares) estimator.

The paper is organized as follows. In the next section, the theoretical and empiricalmodel is presented. In Section 3, the data and the econometric framework are presented.Results are in Section 4 and, finally, concluding remarks are in Section 5.

5Although a couple can choose to move to another commuting area for family reasons, this movewill most likely initiate considerations about moving job region as well.

6Own calculations.

C© 2010, Wiley Periodicals, Inc.

618 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

2. THEORETICAL MODEL

The basic idea of the search theory framework is that workers are not fully compen-sated for their commuting costs by higher wages (or lower house prices) due to labor andhousing market search frictions (see e.g., Manning, 2003, for a theoretical and empiricalfoundation of this claim). Hence, workers are not in their optimal labor-housing marketsituation, but can improve their position through job and residential moves, taking thecost of moving into account. Workers are therefore assumed to continuously search forbetter jobs and dwellings, maximizing the discounted future flow of job utility and placeutility minus commuting costs, taking into account the costs of changing job or residence.The place utility is derived from specific dwelling and location characteristics net of costs,and the job utility is derived from the wage and other nonpecuniary benefits associatedwith the job. The cost of changing residence includes moving costs, as well as the (nonpe-cuniary) cost of getting settled in the new residence. Job moving costs include for instancethe effort to get acquainted with the new job. Job offers and residence offers arrive at anexogenously specified rate, and these offers are accepted or rejected instantly.

Whether a job or residence offer is accepted does not only depend on the gain, net ofmoving costs, in the job or residence utility. The implication of the model is that the choiceof job acceptance and choice of residence acceptance are related because a change in eitherjob or residence location changes commuting costs. The acceptance or rejection of a jobor residence offer thus takes changes in commuting costs into account. The theoreticalfoundation concerning individuals can be found in Deding and Filges (2004b) concerningnonoverlapping regional commuting areas and van Ommeren, Rietveld, and Nijkamp(1998) concerning overlapping regional commuting areas.

For single-earner households, the implication is that the commuting distance posi-tively affects the job and residential moving behavior. Empirically, this implication hasbeen well established (e.g., van Ophem, 1991; Zax, 1991; van den Berg, 1992; Henley, 1998;van Ommeren, Rietveld, and Nijkamp, 1999). However, for workers in dual-earner house-holds the implications are not that simple. Because workers in dual-earner householdsare constrained to live in the same residence as their spouse, job moves influence bothfuture residential moves as well as the spouse’s potential job move, whereas residentialmoves influence the future job moving behavior of both the spouses.

The theoretical link between the spouses’ job moving behavior is established in vanOmmeren et al. (1998) and van Ommeren (2000). When for example a change in one ofthe spouse’s job region implies a large increase in the commuting distance, then, accord-ing to theory, it is more likely that the household in the future will move residence inorder to decrease the commuting distance which in turn increases the probability thatthe other spouse changes job region—and all this is taken into account by the individualwhen deciding on the initial job move. Likewise, a change of residence is likely to changethe commuting distance of both spouses but depending on the spatial job locations of thespouses it has different consequences. If the spouse’s workplaces are close to each otherthe residence move will change the spouse’s commuting distance in the same directionand increase/decrease the probability of a change of job region if the commuting distancesincrease/decrease. If, on the other hand, the spouse’s workplaces are far from each other,the commuting distances of the spouses will most likely change in opposite directions andhave opposite impacts on the probability of a change of workplace region. The theoreticalhypotheses concerning the relationship between commuting distance and spatial mobil-ity have been validated empirically for Denmark by Deding, Filges, and van Ommeren(2009).

In the theoretical framework, the sole link between job and residence mobility is thecommuting costs. A job move which does not change the commuting costs has no impact on

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 619

the probability of moving residence and vice versa. However, the probability of moving—either job or residence—depends on all the parameters of the model, i.e., job and residenceutility, job and residence offer arrival rates, and moving costs. A number of demographicand other variables (e.g., age, education, homeownership, residence duration) are relevant,as they are suggested by economic theory to have an impact on either moving costs or theoccurrence of acceptable job and residence offers.

According to theory, age is negatively correlated with mobility because given a betterjob or a better residence has been accepted at one point in time, the next offer shouldbe better than the present in order to be accepted. Thus, as time goes by the probabilityof receiving job or residence offers above the reservation level decreases and so does theprobability of moving either job or residence.

We expect that having children increase the cost of moving residence due to forexample the difficulty of finding childcare in the new region and the cost of adjusting toa new neighborhood and a new school. These costs are likely to increase with the age ofthe child, because the older the child the more attached it is likely to be with the presentneighborhood (school and friends). Consequently, we expect families with older childrento be less likely to move residence than families with younger children.

Another important factor is education. Typically, higher educated individuals areperceived as more geographically mobile than lower educated individuals, both becausethe higher educated are more career-oriented and because they search for jobs in a largergeographical region. There can be several reasons for this pattern: First, Sabatier (1999)and Mauro and Spilimbergo (1999) show that low-skilled workers more often use local jobcenters (employment exchange services) whereas higher educated workers use advertise-ments and direct applications. The use of job centers implies receiving job offers in thelocal labor market, whereas ads and direct applications enable the contact to firms allover the country. Second, higher educated workers have been shown to use their socialnetwork more frequently when searching for a new job than the less educated workers,which also implies less geographical restrictions. Third, the career opportunities for thehigher educated within a limited geographical region can be more limited, so that careerdevelopment considerations require geographical mobility (Gurak and Kritz, 2000; Swainand Garasky, 2007).

The hypothesis in the previous paragraph relates to individual education and jobmobility. Another question is how education affects family mobility, i.e., residence mobility.To the extent that spouses have the same level of education, it is natural to expectthat higher educated couples are more mobile than lower educated couples. But whenspouses have different education levels, the question is whether the push factor (i.e., thehighest education) or the pull factor (i.e., the lowest education) is the strongest. Familymigration studies have found the husband’s career to be more important than the wifeand, thus, mobility should be higher when the husband is better educated than the wife(e.g., Shihadeh, 1991; Gurak and Kritz, 2000; Nivalainen, 2004; Swain and Garasky,2007). But this expectation may not hold in the Danish labor market that to a largeextent is characterized by dual-earner couples whom we expect to be equal. A specialcase is related to the locational choices of the so-called power couples (Costa and Kahn,2000) who are attracted to large urban areas with their abundance of managerial andprofessional jobs. At the other end of the scale, the results in Whisler et al. (2008), suggestthat smaller college towns (as for example Lafayette and Bryan-College Station) act asmarriage markets for power couples.7 They find very high rates of out-migration of powercouples from these small metro areas. In combination, these results suggest agglomeration

7For an analysis of marriage markets and education in Denmark, see Svarer and Nielsen (2006).

C© 2010, Wiley Periodicals, Inc.

620 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

of the college-educated couples in a selected group of urban centers in the United States.It is, however, beyond the scope of this paper to analyze these matters for Denmark.

The structure of the housing market matters for both the residence offer arrival rateand the transaction costs when moving. The Danish housing market is characterized by ahigh share of homeowners (about 60 percent are homeowners). The large transaction costsinvolved in selling and buying a new home increase moving costs and are clearly obstaclesfor labor mobility.8 However, not only owned homes may pose a barrier for mobility. InDenmark the market for rented housing is highly regulated and characterized by excessdemand.9 Although moving costs are lower for renters, it is reasonable to assume that themarket for renters is less flexible (you have to queue for a rented residence) implying thatthe probability of receiving a better residence offer is low. Thus, it is not clear whetherliving in owned housing is expected to have a positive or negative impact on residencemobility.

The higher the residence utility, the less likely it is to get a residence offer that is bet-ter than the present. Residence utility depends on many factors, among others the size ofthe residence or the presence of recreational opportunities in the neighborhood (e.g., greenspaces). Due to the modest size of the country there are no climate differences betweenregions. Also, it is not the case that some regions are more attractive than others becauseof other spatially tied amenities.10 Living in the metropolitan area (Copenhagen) may beof extra value though, because of the larger supply of cultural activities, so that mobilityout (in) of this region will be lower (higher). Furthermore, we expect that longer dura-tion in a given residence decreases residence mobility, because the reservation residenceutility is expected to increase over time.

The most obvious factor related to job utility is the wage. Unfortunately the wage isnot included in our data. But other factors also matter for the perception of the job. Forinstance, the working conditions differ between job categories, between firms, betweenlines of business, and between sectors. A job in the Copenhagen region can also be moreattractive because this is a very concentrated labor market, making it easier to changejobs without having to move residence at the same time. More experienced workers—interms of labor market experience or job tenure—are expected to be less job mobile, becausetheir reservation job utility increases over time.

Regional variables like unemployment, job vacancies or house prices are not includedin the analysis. According to theory they are expected to influence the regional job andresidence offer rate. However, as the workers are not restrained to receive offers from onespecific region and we do not analyze to which specific region the workers move, thesevariables are of no relevance for the present analysis. Besides, previous research hasshown that regional variables in fact do not influence the moving decision of employedworkers in Denmark, see Deding and Filges (2004a).

3. DATA AND ECONOMETRIC FRAMEWORK

The data used for the analysis are merged together from several administrativeregisters from Statistics Denmark and include information on age, education, family

8Oswald (1999) argues that the European housing market is the explanation for Europe’s unem-ployment problems: Because of the large number of homeowners, labor is simply not mobile enough.

9The market for rented housing is to a very large degree dominated by public housing, and therestrictions on rents are very strict both for public and private rented housing. Consequently, queues forrented housing exist in all regions of the country.

10Although some areas within each region can be more attractive than others. For example, thenorthern part of Zealand in the Copenhagen region or Skade in the Arhus region.

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 621

composition, residence and work location, and residence and job characteristics.11 The in-formation is from the end of 1999 and the end of 2000 (December 31). The data populationis all individuals in the labor force and for the present analysis, we restrict to dual-earnercouples12, where both spouses are 20–59 years and where both were employed both bythe end of 1999 and by the end of 2000.13 This restriction is made because we want tofocus on the relationship between family migration and geographic job mobility of bothspouses. This leaves us with about 450,000 couples.14 We analyze job changes and resi-dence changes in the year 2000, i.e., between the end of 1999 and the end of 2000.

Two aspects are important for the specification of the empirical model: First, thegeographical structure of the model and, second, the timing of events. Concerning thegeographical structure of the labor and housing markets and the geographical definitionof a change of job or a change of residence; as mentioned previously, we are only interestedin inter-regional moves, and not in intra-regional moves. Likewise, although commutingis an important variable of the model, the focus of the analysis is spatial migration ratherthan commuting per se. These considerations are discussed in more detail below.

Administratively, Denmark was, until 2007, divided into 14 counties and 275 munici-palities. However, for the present analysis we will as mentioned use a functional division;i.e., nonoverlapping commuting areas, as the geographic unit. A commuting area is definedas an area where people commute within the area, but only seldom between areas. Theregional structure is of course not constant over time. For example the work force movedwest away from the Copenhagen region in the 1980s, while this pattern was reversedin the 1990s where the work force moved east. Furthermore, over time the commutingareas are getting fewer and larger, i.e., people are commuting longer distances, on average13 kilometers (8 miles) in 1980 and 16 kilometers (10 miles) in 1995.

In 2000 (the time of the present analysis), Denmark could be divided into 45 nonover-lapping commuting areas consisting of between 1 and 55 municipalities.15 Copenhagen,consisting of 55 municipalities, is the largest commuting area both in terms of inhabitantsas well as geographical extent. Aarhus is the second largest consisting of 13 municipalities.A change of residence is defined as a move from one commuting area to another commut-ing area, and a change of job is defined as a change of workplace from one commuting areato another commuting area. A map of Denmark divided into the 45 commuting areas isshown in Figure 1. As mentioned Copenhagen is the largest region. Of the approximately2.5 million employed workers in Denmark, one-third work and/or live in the Copenhagenregion. In contrast, 8 percent of the employed workers work and/or live in the secondlargest region, the Aarhus region.

By choosing commuting areas as the geographical unit of the empirical analysis,we limit the number of typical life-cycle moves—like a couple moving to a larger homebecause they become parents. It should be mentioned that any definition of moves based ongeographical borders potentially have a problem with short moves just across the border.

11Register data from Statistics Denmark can only be accessed by registered users and are thusproprietary.

12The couples may be married or cohabiting. For convenience, we use the terms “husband” and “wife”whether the spouses are legally married or not.

13Being employed by the end of the year is defined as being employed in mid-November.14In the data, we have 548,465 couples where both are employed in 1999. Thus, we loose about

100,000 couples by restricting to employment for both spouses in both years. However, analysis of thedistribution of variables across the selected and unselected sample does not reveal any important differ-ences. The major reasons for being a dual-earner couple in the first year and not in the second year arechildbirths or entering education.

15This particular division is based on the Ministry of the Environment’s calculations; see the Ministryof the Environment (2001).

C© 2010, Wiley Periodicals, Inc.

622 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

FIGURE 1: Commuting Areas in Denmark, 2000.

However, by using the commuting areas this is not really a problem, because commutingareas are defined around a geographic center town that is important for the local identityof the inhabitants. Although a move across the border between two commuting areasmay not be very long in terms of distance, it may be a long move in terms of identity.Consequently, also these short moves are likely to be related to job considerations.

Another very important aspect of the model is time: although the theoretical model isformulated in continuous time, due to data restrictions, the empirical model is formulatedin discrete time. Theoretically, job or residence moves are independent conditionally oncommuting costs and takes place sequentially one at a time.16 Our data are not ableto capture this sequence, though, because we only have yearly data and, hence, yearlymobility. For instance if a residence move is followed by a job move, both moves will likelytake place within the same calendar year and, consequently, job and residence moves canoccur simultaneously in the empirical specification of the model.

16This condition is tested in van Ommeren et al. (1999): using information on the exact dates of joband residence moves for a sample of Dutch males, they find that residence and job moves are independentgiven commuting costs.

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 623

TABLE 1: Relationship between Change of Residence and Change of Job

Male Job Female Job Male and Female No JobChange Only Change Only Job Change Change Total

No residence change 9.7% 7.6% 1.6% 81.0% 100.0%Residence change 16.7% 16.9% 14.6% 51.8% 100.0%

Furthermore, as noted in the introduction, we argue that job and residence movesbetween nonoverlapping commuting areas to a large extent can be characterized as com-plements. If one of the household’s earners is offered a job in another area several hundredkilometers away, then according to the search model, if the job is accepted, the householdhas taken into consideration the need of moving residence in order to reduce the long com-muting distance as quickly as possible. Hence, it implies that the acceptance of a job “faraway” and a residence move are considered as parts of a whole package and takes placesimultaneously. In many cases therefore, the intervening housing-job location situationis of no importance. What matters for residence mobility is the fact that an interregionaljob move has occurred. Hence, although, in theory, the sole link between job and residencemoves is commuting distance, we allow the moves to be interdependent.17

In the econometric model, we estimate the probabilities of three events: the prob-ability of moving residence to another commuting area, the probability of the husbandchanging job to another commuting area, and the probability of the wife changing job toanother commuting area. According to the above arguments, these events can be viewedas interdependent, and this must be taken into account by the econometrical procedure.

We expect the three types of mobility to be positively correlated, i.e., we expect thatjob mobility has a positive impact on residence mobility and that residence mobility has apositive impact on job mobility. Furthermore, we expect the spill over from a residentialmove to the job moves to be larger than the other way around. This expectation is based onthe findings in Deding and Filges (2004c), where reasons for residence mobility betweenmunicipalities among Danish wage earners were analyzed. Of a representative sample ofwage earners that had moved during 1998, more than half of the moves were initiated byvarious family reasons, while a job move (by the respondent and/or his/hers spouse) wasinvolved in two-thirds of the residence moves. In other words, although the decision tomove residence often comes first, this move is very often followed by a job move.18

The interdependence between the moves is clear from inspection of the data usedin the present analysis. In Table 1, job changes during the year are shown dependingon whether a residential move has taken place or not. Among the couples who did notmove residence, 81 percent did not change job region; but among the couples who didmove residence, close to 50 percent of the couples also changed job region—split almostevenly in only the husband changing job region, only the wife changing job region, or bothspouses changing job regions.

Consider the estimation of the three geographic moves that can occur simultaneously.A change of job occurs if one spouse changes from a job in one commuting area to a jobin another commuting area between period t and period t + 1, and a change of residenceoccurs if a couple changes from a residence in one commuting area to a residence inanother commuting area between period t and period t + 1. In the model, each type ofmobility is allowed to affect the other mobility types, i.e., residence mobility affects job

17Besides, of course, allowing commuting distance to affect mobility as prescribed by theory.18Note that this result is based on a sample of both single and cohabitant wage earners (58 percent

of the respondents that had moved residence were cohabiting).

C© 2010, Wiley Periodicals, Inc.

624 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

mobility of both spouses, and his/her job mobility affects residence mobility and her/hisjob mobility.

The econometric specification of the model is given in (1) and (2). Let yr be an indicatorvariable for residence mobility, and ym

j and y fj be indicator variables for job mobility for

the male and the female, respectively. Defining y∗r , ym∗

j , and y f ∗j as the latent variables

corresponding to yr, ymj , and y f

j the model is as follows (where xr, xmj , and x f

j are vectors ofexogenous variables; � and � are coefficients; and εr, εm

j , and ε fj are error terms):

y∗r = �m

j ymj + �

fj y f

j + �′rxr + εr,

ym∗j = �r yr + �

fj y f

j + �m′j xm

j + εmj ,

y f ∗j = �r yr + �m

j ymj + �

f ′j x f

j + ε fj ,

(1)

and

yr = 1 if y∗r > 0,

= 0 if y∗r ≤ 0,

ymj = 1 if ym∗

j > 0,

= 0 if ym∗j ≤ 0,

y fj = 1 if y f ∗

j > 0,

= 0 if y f ∗j ≤ 0.

(2)

The probabilities of the couple moving residence, the husband changing job, or the wifechanging job are estimated by Amemiya’s generalized least squares (AGSL) for probitmodels with endogenous regressors (Amemiya, 1974, 1978). With the AGLS estimator,the endogenous regressors are treated as linear functions of the instruments and theexogenous variables, while correcting for the truncated distribution of the dependentvariable. In essence, the AGLS is a variant of the traditional generalized least squares(GLS) estimator adjusted to limited dependent variables. Furthermore, AGLS is superiorin terms of efficiency to traditional two-stage least squares (2SLS) or three-stage leastsquares (3SLS) estimators (Newey, 1987). In practice, the AGLS estimator is equivalentto a two-stage instrumental variables (2SIV) estimator: In each equation, the coefficientsof the endogenous regressors are estimated by instrumental variables using all variablesin the model, and then standard errors are adjusted according to the probit specification.

The exogenous factors of the model are divided into three subgroups: variables relatedto residence mobility (xr), variables related to male job mobility (xm

j ), and variables relatedto female job mobility (x f

j ). The variables for instance related to residence mobility directlyaffects residence mobility (the first line of equation 1), but are also used as instrumentsfor the residence mobility in the job mobility estimations (the second and third line ofequation 1); and likewise for the explanatory variables related to male and female jobmobility. An interpretation of this structure is that for instance variables in xr affectresidence mobility directly, but affect male and job mobility indirectly through the effectof residence mobility in these estimations. Identification of the coefficients in the modelis achieved by omitted variables, as the set of variables in xr, xm

j , and x fj are not identical.

As mentioned earlier a change of job or a change of residence is defined as a movefrom one commuting area to another commuting area during the year 2000. Of the 450,000couples we are considering, 2.6 percent moved residence to another commuting area inthis period, while 11.9 percent of the men and 9.8 percent of the women changed job toanother commuting area (Table 2). Geographical job mobility is thus relatively high and,furthermore, it is substantially larger than the geographical residence mobility.

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 625

TABLE 2: Means of Variables

Couples Men Women

Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

Change of residence (1999–2000) 2.6% (16.0%)Change of job (1999–2000) 11.9% (32.3%) 9.8% (29.8%)Distance between home and work

(in km)15.567 (28.557) 9.698 (19.002)

Max distance between home and work 20.073 (31.149)Distance difference 14.880 (29.266)20–29 years 14.6% (35.3%) 12.2% (32.7%) 17.0% (37.6%)30–39 years 33.8% (47.3%) 30.6% (46.1%) 33.4% (47.2%)40–49 years 34.2% (47.4%) 31.6% (46.5%) 31.5% (46.5%)50–59 years 22.0% (41.4%) 25.6% (43.7%) 18.0% (38.5%)Age difference 3.193 (2.969)Number of children 0.988 (1.020)Oldest child 0–2 years 7.0% (25.6%)Oldest child 3–6 years 10.9% (31.2%)Oldest child 7–17 years 38.5% (48.7%)Low education 24.7% (43.1%) 28.3% (45.0%)Medium education 66.6% (47.1%) 66.6% (47.2%)High education 8.7% (28.1%) 5.1% (22.1%)Finished education 0.9% (9.4%) 1.9% (13.8%)Male low—female low 10.2% (30.3%)Male low—female medium 13.9% (34.6%)Male low—female high 0.5% (7.4%)Male medium—female medium 47.8% (50.0%)Male medium—female high 1.8% (13.1%)Male high—female low 0.9% (9.6%)Male high—female medium 4.9% (21.5%)Male high—female high 2.8% (16.6%)Owned housing 80.0% (40.0%)Rented housing 19.4% (39.5%)Missing owner information 0.6% (7.9%)Size of residence per family member

(in m2)48.673 (25.042)

Residence duration (in years) 10.561 (9.488)Labor market experience (in years) 19.146 (9.042) 15.052 (8.210)Private sector 76.8% (42.2%) 52.4% (49.9%)Municipal (public) sector 12.4% (33.0%) 39.7% (48.9%)Government (public) sector 10.8% (31.1%) 7.8% (26.8%)Residence in Copenhagen area 35.1% (47.7%)Job in Copenhagen area 37.1% (48.3%) 36.1% (48.0%)Number of observations 451,964 451,964 451,964

In the following, we define the variables used in the econometric specification of themodel, see Table 2. All background characteristics are from 1999, i.e., prior to the possiblemove.

We do not have information on actual commuting costs, but use commuting distancein kilometers as a proxy. More specifically, commuting distance is defined from munici-pality center to municipality center. If a person lives and works in the same municipality,the commuting distance is zero by definition; while commuting distance by definition isgreater than zero if the person lives and works in different municipalities. Men commute

C© 2010, Wiley Periodicals, Inc.

626 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

more often and longer than women. In the selected sample 48 percent of the men do notcommute while 59 percent of the women do not commute. On average, men commute16 kilometers (10 miles) and women commute 10 kilometers (6 miles). In addition to theindividual commuting distances, we define the maximum commuting distance for the cou-ple (the largest commuting distance of either the husband or the wife) and the differencein commuting distances for the spouses. As explained previously, the hypothesis is thatthe couple seeks to reduce commuting costs by moving closer to the job with the highestcommuting cost; but that the probability of moving residence decreases with the distancebetween the spouses’ jobs.

Besides commuting costs a number of exogenous factors will matter for mobility inthe model, because they affect the structural parameters of the model, see Section 2.The individuals’ age distribution, used in the job mobility equations, is divided into fourcategories: 20–29 years, 30–39 years, 40–49 years, and 50–59 years. The average age of thecouple, divided into the same four categories, is used in the residence mobility equation.In addition, the age difference (husband’s age minus wife’s age) is used in the residencemobility equations. On average, the men in our sample are three years older than thewomen, which is a typical finding for couples in Denmark. The age difference is used tocorrect for cases where the age difference is so large that the spouses are at very differentstages of their life cycles.

Concerning children, the couples on average have one child aged 0–17 years (about44 percent of the couples do not have children in this age-group). As mentioned previously,we expect the age of the oldest child to matter most for the residence mobility decision. Weinclude three dummy variables for age of the oldest child in the family related to differentphases in a child’s life: 0–2 years (where the child is typically cared for at home or in a daynursery), 3–6 years (where the child is typically in kindergarten), and 7–17 years (wherethe child is at school).

The Danish educational system is basically characterized by two main tracks thatDanish students can choose after the completion of compulsory schooling (grade 9): thevocational track and the post secondary track. The apprentice or vocational track is similarto the German educational system where taking a vocational education would imply botha period of time with an employer (meister) and a period of time in a technical schoolwith exams. The other track begins with three years of high school and from there it ispossible to go on to university colleges19 or universities (with Bachelor, Master’s, and Ph.D.programmes). Depending on duration, these educations are labeled “short post secondary”(e.g., police training), “medium post secondary” (e.g., teacher or nurse training), and “longpost secondary’ (university degree such as Master or Ph.D.20).

Based on this educational system, we split education into three categories: low,medium, and high. Low consists of those having no qualifying education, medium of thosewho have vocational qualifications, short or medium post secondary education, and highof those who have a long post secondary education.21 Two-thirds of both men and womenhave medium educational attainment, while more men than women—8 percent com-pared to 5 percent—have higher educational attainment. As pointed out previously, the

19The university colleges are not identical to U.S. colleges. For a more thorough description of theDanish education system, see McIntosh and Munk (2007).

20It is not common for students to leave university with only a bachelor degree. Thus, less than1 percent of the employed workers in Denmark have a bachelor degree only. They are placed in thecategory “medium post secondary.”

21Although information on education is more detailed, we choose to work with three categoriesonly because we also define interactions between the spouses’ educations, and this becomes unfeasible ifeducation is split into more categories.

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 627

combination of education within the couple can be important and, hence, we define dif-ferent education combinations of the spouses. Not surprisingly given the individual ed-ucations, in almost half of the cases (48 percent) both spouses have medium education.On the other hand, only 2.8 percent of the couples are so-called power couples where bothspouses have high educations.

In addition to the educational level in 1999, a dummy for completing an educationduring the year 2000 (and thus moving to another educational category) is included inthe job mobility estimations. The share of individuals who finished an education in 2000is rather small, but twice as high for women compared to men (2 percent compared to1 percent).

The vast majority of the couples live in owned residences (80 percent), while19 percent lives in rented residences (the status of the rest is unknown). Hence, thedual-earner couples more often live in owned residences than the average Danish popu-lation (as mentioned earlier, on average 60 percent of Danes live in owned housing). Theaverage size of the residence is calculated per family member including children, and onaverage families have lived in the same residence (at the same address) for 10 years.

Actual labor market experience is measured in years and averages 19 years for menand 15 years for women. The Danish labor market is highly segregated: 77 percent ofthe men are employed in the private sector compared to 52 percent of the women, while12 percent of the men are employed in the local government public sector compared to40 percent of the women.

Finally, we define two dummy variables for living or working in the Copenhagencommuting area. Because this specific commuting area is very large both in terms ofinhabitants and jobs, mobility to other commuting areas from this specific area may berelatively low. Note, that there is a small degree of in-commuting into the Copenhagenarea—more people work there than live there; a pattern typical for capitals.

In order to achieve identification of the coefficients of the model, we distinguishbetween couple-specific variables and individual-specific variables. We let xr consist ofthe couple variables (column 1 in Table 2), xj

m consist of the male variables (column 3 inTable 2), and xj

f consist of the female variables (column 5 in Table 2). For instance, childrenare included in the residential mobility estimation, but not in the job mobility estimations.This implies that we assume that children can have a direct impact on residential mobility,and an indirect impact on job mobility through the effect on residential mobility. On theother hand, individual labor market experience are assumed to have a direct impact onindividual job mobility, but only an indirect impact on residence mobility or job mobilityof the spouse.

4. EMPIRICAL RESULTS

In Table 3, we present the estimated coefficients for the probabilities of changingresidence, the husband changing job, and the wife changing job to another commutingarea. The majority of the variables are highly significant, due to the large number ofobservations and, consequently, the significance of the estimated coefficients is less inter-esting than the size of the parameters. The discussion of the results is therefore based onthe calculated marginal effects presented in Table 4.22

In the first three rows of Table 4, we find the interaction coefficients, i.e., the esti-mated correlation between the moves. The largest interaction coefficient is found for achange in residence: A move of residence to another commuting area for a couple is asso-ciated with an increase in the probability of the wife changing job to another commuting

22Marginal effects are calculated at sample means.

C© 2010, Wiley Periodicals, Inc.

628 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

TABLE 3: The Probability of Changing Residence and Changing Job

Change of Residence Change of His Job Change of Her Job

Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.

Changed residence 1.738∗∗∗ (0.123) 0.649∗∗∗ (0.131)Changed his job 1.199∗∗∗ (0.181) 0.560∗∗∗ (0.048)Changed her job 1.016∗∗∗ (0.168) 0.294∗∗∗ (0.041)Distance between home and work

(in km)0.664∗∗∗ (0.008) 1.096∗∗∗ (0.012)

Maximum distance 0.145a (0.058)Difference in distance −0.206∗∗∗ (0.046)30–39 years 0.040∗ (0.015) −0.102∗∗∗ (0.009) −0.088∗∗∗ (0.009)40–49 years −0.083∗∗∗ (0.021) −0.179∗∗∗ (0.012) −0.182∗∗∗ (0.012)50–59 years −0.164∗∗∗ (0.026) −0.251∗∗∗ (0.014) −0.370∗∗∗ (0.014)Age difference 0.003 (0.002)Number of children −0.080∗∗∗ (0.011)Oldest child 0–2 years 0.062∗∗ (0.019)Oldest child 3–6 years −0.100∗∗∗ (0.024)Oldest child 7–17 years −0.274∗∗∗ (0.025)Medium education −0.028∗∗∗ (0.006) −0.026∗∗∗ (0.006)High education −0.081∗∗∗ (0.010) −0.003 (0.013)Finished education 0.159∗∗∗ (0.023) 0.180∗∗∗ (0.017)Male basic female medium

education0.096∗∗∗ (0.018)

Male basic female high education 0.201∗∗∗ (0.050)Male medium female basic

education0.081∗∗∗ (0.018)

Male medium female mediumeducation

0.142∗∗∗ (0.016)

Male medium female higheducation

0.267∗∗∗ (0.031)

Male high female basic education 0.216∗∗∗ (0.041)Male high female medium

education0.310∗∗∗ (0.023)

Male high female high education 0.359∗∗∗ (0.026)Rented housing 0.497∗∗∗ (0.011)Missing owner information 0.141∗ (0.045)Size of residence per family

member−0.271∗∗∗ (0.028)

Residence duration (in years) −0.012∗∗∗ (0.001)Labor market experience (in years) −0.015∗∗∗ (0.001) −0.039∗∗∗ (0.002)Labor market experience squared 0.002 (0.004) 0.071∗∗∗ (0.005)Municipal (public) sector −0.179∗∗∗ (0.009) −0.071∗∗∗ (0.006)State (public) sector −0.140∗∗∗ (0.009) −0.193∗∗∗ (0.011)Residence in Copenhagen area 0.036∗ (0.012)Job in Copenhagen area 0.183∗∗∗ (0.006) 0.230∗∗∗ (0.006)Constant −2.203∗∗∗ (0.041) −1.025∗∗∗ (0.016) −1.090∗∗∗ (0.014)Pseudo R2 0.1437 0.0732 0.0956Number of observations 451,964 451,964 451,964

∗∗∗Significant at 0.01% level, ∗∗significant at 0.1% level, ∗significant at 1% level, asignificant at 5% level.

area by 14 percent and an increase in the probability of the husband changing job by57 percent. The size of this marginal effect might seem dramatic, but keep in mind thatonly 2.5 percent of the couples move residence within the given year, while the job mobil-ity rate is much higher. As expected, this result shows that given a change of residence

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 629

TABLE 4: Marginal Effect on the Probability of Changing Residence and Changing Job

Change of residence Change of his job Change of her job

Mar. Eff Std. Err. Mar. Eff Std. Err. Mar. Eff Std. Err.

Changed residence 0.571∗∗∗ (0.044) 0.143∗∗∗ (0.038)Changed his job 0.115∗∗∗ (0.033) 0.111∗∗∗ (0.012)Changed her job 0.087∗∗∗ (0.027) 0.061∗∗∗ (0.010)Distance between home and work

(in km)0.119∗∗∗ (0.001) 0.164∗∗∗ (0.002)

Maximum distance 0.005a (0.002)Difference in distance −0.007∗∗∗ (0.002)30–39 years 0.001∗ (0.001) −0.018∗∗∗ (0.002) −0.013∗∗∗ (0.001)40–49 years −0.003∗∗∗ (0.001) −0.031∗∗∗ (0.002) −0.026∗∗∗ (0.002)50–59 years −0.005∗∗∗ (0.001) −0.042∗∗∗ (0.002) −0.047∗∗∗ (0.001)Age difference 0.000 (0.000)Number of children −0.003∗∗∗ (0.000)Oldest child 0–2 years 0.002∗∗ (0.001)Oldest child 3–6 years −0.003∗∗∗ (0.001)Oldest child 7–17 years −0.009∗∗∗ (0.001)Medium education −0.005∗∗∗ (0.001) −0.004∗∗∗ (0.001)High education −0.014∗∗∗ (0.002) 0.000 (0.002)Finished education 0.032∗∗∗ (0.005) 0.030∗∗∗ (0.003)Male basic female medium

education0.004∗∗∗ (0.001)

Male basic female high education 0.009∗∗∗ (0.003)Male medium female basic

education0.003∗∗∗ (0.001)

Male medium female mediumeducation

0.005∗∗∗ (0.001)

Male medium female higheducation

0.012∗∗∗ (0.002)

Male high female basic education 0.009∗∗∗ (0.002)Male high female medium

education0.015∗∗∗ (0.001)

Male high female high education 0.018∗∗∗ (0.002)Rented housing 0.024∗∗∗ (0.001)Missing owner information 0.006∗ (0.002)Size of residence per family

member−0.009∗∗∗ (0.001)

Residence duration (in years) −0.000∗∗∗ (0.000)Labor market experience (in years) −0.003∗∗∗ (0.000) −0.006∗∗∗ (0.000)Labor market experience squared 0.000 (0.001) 0.011∗∗∗ (0.001)Municipal (public) sector −0.030∗∗∗ (0.001) −0.010∗∗∗ (0.001)State (public) sector −0.023∗∗∗ (0.001) −0.026∗∗∗ (0.001)Residence in Copenhagen area 0.001∗ (0.000)Job in Copenhagen area 0.034∗∗∗ (0.001) 0.036∗∗∗ (0.001)Pseudo R2 0.1437 0.0732 0.0956Number of observations 451,964 451,964 451,964

∗∗∗Significant at 0.01% level, ∗∗significant at 0.1% level, ∗significant at 1% level, asignificant at 5% level.

to another commuting area, the probability of a simultaneous job change is very high,especially for men.

Looking at the opposite relationship, job changes are also positively correlated withthe probability of moving residence, although with a smaller magnitude. A change in the

C© 2010, Wiley Periodicals, Inc.

630 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

husband’s job region is associated with an increase in the probability of a residence moveby 11 percent, while the corresponding increase in the probability of the wife’s job changeis about 9 percent. The marginal effect of the husband’s job change is significantly largerthan the marginal effect of the wife’s job change (by a standard t-test). This implies thatthe husband’s job change is more important for residence mobility than the wife’s jobchange, although the difference between the coefficients is not very large.

The last interaction term that we consider is the relationship between the spouses’job moves. A change in the husband’s work region is associated with an increase in thewife’s probability of changing work region by 11 percent, while the opposite relationship is6 percent. This result—that the husband’s job move has a greater impact on the wife’s jobthan the other way around—again suggests that his job is more important for the family’smoving decisions than her job. Contrary to what we expected these results suggest thateven in an egalitarian country like Denmark his job is more important than hers, at leastin terms of geographical moving decisions. This result is in accordance with the results inBielby and Bielby (1992).

Taken as a whole, the estimated interaction terms are large and highly significant.Furthermore, compared to the other explanatory variables in the analysis, we see that thelargest marginal effects relate to the interaction parameters. This clearly confirms thatthe moves are interdependent. In particular we find that moving residence has a greatimpact on the probability of moving job for both spouses. This finding, in combination withthe results in Deding and Filges (2004c), suggests that for many couples the residencemove comes first and is then followed by a job move.

Looking at the other explanatory variables, in general we find the expected results.Commuting distance between home and work affects the probability of changing job pos-itively for both men and women, but more for women than for men. Together with thefact that women’s average commuting distance is about two-thirds of men’s average com-muting distance (see Table 2), an interpretation is that, even in a presumably egalitariancountry as Denmark women prefer to work closer to home than men. The possible ex-planations for the gender differences in commuting are analyzed for instance by Madden(1981) and Hanson and Johnston (1985) and include women’s shorter work hours, lowerwage and the fact that they have more household responsibilities. The shorter commutingfor women relative to men as well as the structural explanations are also present in theDanish context, see Deding, Lausten, and Andersen (2006).

As expected, and in line with the results in Deding et al. (2009), the probabilityof moving residence increases with the maximum commuting distance, but decreaseswith a larger difference in commuting distance between the spouses. However, it is notsurprising that the marginal effect of the commuting distance is much larger in thejob change equations than in the residence change equation. This is because adjustingthe commuting distance by moving residence probably is more costly than adjusting thecommuting distance by moving job.

Concerning age, we expect younger people to be more mobile than older people. Andthis is what we find with respect to job mobility: Both men and women are less likelyto change work region the older they are (the left out category is the age interval 20–29 years). For the probability of moving residence, the result is a little different. Comparedto couples on average aged 20–29 years, the probability of moving residence for the 30–39year olds is marginally higher. A reason for this may be that many Danish couples are intheir thirties before they have children and settle down. Being older, i.e., 40–49 years or50–59 years reduces the probability of moving residence as expected. The age differencebetween the spouses has no effect on the probability of moving residence. Although a largeage difference could imply that the spouses are at different points of their life-cycles, thisdoes not have an impact on the probability of moving residence, probably because only

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 631

few couples have very large age differences (the standard deviation of the age differenceis three years).

As expected, children matter for the probability of moving residence although themarginal effect is not very large. Both number of children and age of the oldest child(compared to having no children) matter, confirming the hypothesis of older childrenbeing more costly to move as the effect of having an older child is negative. In fact, theeffect of the oldest child being 0–2 years old is positive, possibly reflecting that having thefirst child often makes it necessary to adjust the size and location of the residence.

Concerning the impact of education on job mobility, other studies have found thathigher educated workers are more likely to be geographically mobile than lower educatedworkers (e.g., Greenaway, Upward, and Wright, 2002; Wayne, Webber, and Walton, 2002).Our findings do not confirm this. For men, we find that job mobility decreases witheducation (the left-out category is low education); while for women, we find that womenwith medium education has a lower probability of changing job region compared to womenwith low or high education. A reason for this finding can be that the labor market for higheducated workers is more concentrated to a few specific commuting areas, while thedemand for less educated workers is spread out across the country. Recall that Denmarkis a small country so that the commuting areas are very small for instance by U.S.standards. Furthermore, nearly all governmental administration—a very large employerof high educated individuals—takes place in Copenhagen limiting the real possibility of jobmobility. On the other hand, we do as expected find that having completed an educationduring 2000 has a positive impact on job mobility. Thus, both women and men have ahigher probability of changing job to another region when they finish their education.

On the other hand, turning to the effect of education on residence mobility we do findthe expected structure. Compared to couples where both spouses have low education, theprobability of moving residence increases with the couple’s educational level. The highestmarginal effect on moving is thus found for couples where both have a high education.Interestingly, when looking at different combinations of education men’s education doesnot appear to be superior to women’s or vice versa. For instance, the marginal effect ofa couple where the man has high education and the woman has low education is exactlythe same as the marginal effect for the couples where he has low education and she hashigh. Furthermore, the dominant educational level seems to be having a high educationwithin the couple, as the marginal effect are higher for any combination including thisthan for couples where both have medium educations.

Apparently, the results regarding education in the job mobility and residence mobilityequations are in conflict with each other. The explanation is probably that many low-skilled workers move job very often—apparently also across geographic region, but on ashort-term basis where they do not bring their families. Apparently higher-skilled workersmore often move with their families, when they move across commuting areas.

Looking at the housing variables, we find that relative to living in an owned residence,the probability of moving is higher if the residence is rented or the owner informationis missing. Thus, families living in an owned residence are less geographically mobile,as also found by e.g., Munch, Rosholm, and Svarer (2003) and Oswald (1999). The sizeof the residence per family member has a negative effect on the probability of moving,indicating that when the couple has found a residence that is large enough the probabilityof moving decreases as expected. The residence duration (defined as years living at thesame address) also has a negative effect on the probability of moving; however, the effectis negligible in size.

Similar to the change of residence equation, some labor market specific variablesare included in the job change equations. Years of labor market experience decreasethe probability of changing job region for both women and men, but with a decreasing

C© 2010, Wiley Periodicals, Inc.

632 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

magnitude for women (experience squared has a positive sign). This reflects that jobchanges (geographic or not) are more frequent in the beginning of the labor marketcareer. Compared to employees in the private sector, employment in the public sectordecreases the probability of changing jobs to another region for both men and women.This is expected, because job turnover is generally larger in the private sector than in thepublic sector, whether the state government or the local government sector.

Finally, we find that both residence mobility and job mobility is higher for couplesinitially living or working in the Copenhagen region, although only marginally for theresidence mobility. These results are contrary to expectations, but a reason can be thathousing prices have increased very much in the metropolitan area during the last decadeforcing people to move further away from the city in order to find affordable housing.Although Copenhagen is an attractive commuting area per se, the increasing housingprices may have outweighed this advantage.

5. CONCLUSION

In this paper, we analyze geographical residence and job mobility for Danish couplesand the interaction of the moves. Contrary to previous analyses on Danish data, thefocus is on the household as the decision-making unit and we only consider dual-earnercouples. The purpose of the paper is to analyze geographical mobility from a labor marketperspective. We therefore use a functional division of the country, i.e., commuting areas.Very often, a geographical residence change is followed by a geographical job change andvice versa; the theoretical link between the two being the commuting distance. Becausewe do not have the exact dates of the moves and the decisions of moving residence and jobto another commuting area are closely linked, we analyze the job changing probability ofboth spouses together with the probability of moving residence taking the interdependencebetween the three events into account. Hence, we focus on the relationship between familymigration and geographic job mobility of both spouses and the analysis is limited to dual-earner couples aged 20–59 years, where both are employed both before and after thepotential job and/or residence move. Consequently, we do not study life-cycle events suchas marriage, divorce, or retirement.

Looking at the empirical results, in general residence mobility is lower than jobmobility. The close link between the probabilities of the couple moving residence, thehusband changing job, and the wife changing job is evident. In particular, the link fromthe residence move to the job-changing probabilities is strong: moving residence increasesthe probability of both the wife and the husband changing to a new job region. Especiallythis is true for the men, where the association between residence mobility and job mobilityis very strong. Also the opposite relationship matters: a job change by either the husbandor wife increases the probability of moving residence. The marginal effect of the husband’sjob change is significantly larger than the effect of the wife’s job change. This implies thatthe husband’s job change is more important for residence mobility than the wife’s jobchange, although the difference between the coefficients is not very large.

Regarding the explanatory variables, by and large we find the expected results.Couples have the highest probability of moving residence when they are in their 30sand when they have their first child. Furthermore, the probability of moving residenceincreases with commuting distance but decreases with the difference in commuting dis-tance between the spouses. Higher educated couples have a higher probability of movingresidence, and this finding is gender neutral. Thus, higher education makes moving resi-dence more likely, no matter whether the wife or the husband have the highest educationalattainment. The important variable seems to be the highest level of education within thecouple, more than the educational level of either the husband or the wife.

C© 2010, Wiley Periodicals, Inc.

DEDING AND FILGES: GEOGRAPHICAL MOBILITY OF DANISH DUAL-EARNER COUPLES 633

The probability of changing job region decreases with age for both men and women.On the other hand, a relatively large gender-difference is found for the effect of thecommuting distance. An increase in commuting distance increases the wife’s probabilityof changing job to another commuting area significantly more than it affects the husband.At the same time, women’s average commuting distance is considerably shorter thanmen’s average commuting distance. An explanation for this difference is that althoughactive in the labor market, women are still the main responsible for the family and arethus inclined to work closer to home. Education, however, does not have the expected effecton job mobility: for both women and men the highest probability of moving job region isfound for individuals with the lowest level of educational attainment. For women, havinga high level of education implies an equally high probability of moving job region; but formen, the probability of moving job is significantly lower for individuals with a high levelof education compared to individuals with a low level of education. This suggests, thatlow-skilled labor in Denmark is much more geographically mobile than expected, but thatthey change job region without changing residence region as well.

Summing up, the analysis in this paper gives new insight about the mechanisms ofgeographical mobility of Danish dual-earner couples and about the link between theirresidential mobility and geographic job mobility. The findings stress the importance ofconsidering the family when studying geographical mobility. If policy-makers want to in-crease job mobility within the country and induce labor to shift between regions, the familyeffects must clearly be taken into account. Thus, mobility-promoting initiatives must fo-cus on families rather than individuals and, furthermore, it is necessary to recognize thatfor most families the choice of residence location dominates the choice of job location. Forinstance firms trying to attract high-educated workers outside the metropolitan regionshould consider family packages including attractive employment for both spouses as wellas an attractive residence.

REFERENCESAmemiya, T. 1974. “Multivariate Regression and Simultaneous Equation Models when the Dependent Variables

are Truncated Normal,” Econometrica, 42, 999–1021.———. 1978. “The Estimation of a Simultaneous Equation Generalized Probit Model,” Econometrica, 46,

1193–1205.Andersen, A.K. 2000. Commuting areas in Denmark, Copenhagen: AKF.Bartel, A.P. 1979. “The Migration decision: What Role Does Job Mobility Play?” The American Economic Review,

69(5), 775–786.Bielby, W.T. and D.D. Bielby. 1992. “I Will Follow Him: Gender-Roles Beliefs, and Reluctance to Relocate for a

Better Job,” The American Journal of Sociology, 97(5), 1241–1267.Costa, D.L. and M.E. Kahn, 2000. “Power Couples: Changes in the Locational Choice of the College Educated,

1940–1990,” The Quarterly Journal of Economics, 115(4), 1287–1315.Danish Economic Council. 2002. Dansk økonomi, efterar 2002 (Danish Economy, fall 2002). Danish Economic

Council, Copenhagen.Deding, M. and T. Filges. 2004a. “Arbejdsløs og geografisk mobil?” (Unemployed and geographically mobile?),”

in Niels Henning Bjørn (ed.), Bolig, Mobilitet og Marginalisering pa Arbejdsmarkedet (Dwellings, Mobilityand Marginalization in the Labor Market). The Danish National Institute of Social Research, Copenhagen,Denmark. Report 04:11.

———. 2004b. “Change of Job and Change of Residence—Geographical Mobility of the Labour Force,” WorkingPaper No. 9, The Danish National Institute of Social Research, Copenhagen, Denmark.

———. 2004c. “Derfor Flytter Vi—Geografisk Mobilitet i den Danske Arbejdsstyrke,” (That’s why We Move—Geographical Mobility in the Danish Labor Force). Report. No. 19. The Danish National Institute of SocialResearch, Copenhagen, Denmark.

Deding, M., M. Lausten, and A. Andersen. 2006. “Børnefamiliernes Balance Mellem Familie- og Arbejdsliv,”(Families with Children and the Balance between Family- and Working Life). Report. No. 32. The DanishNational Institute of Social Research, Copenhagen, Denmark.

Deding, M., T. Filges, and J. van Ommeren. 2009. “Spatial Mobility and Commuting: The Case of Two-EarnerHouseholds,” Journal of Regional Science.

C© 2010, Wiley Periodicals, Inc.

634 JOURNAL OF REGIONAL SCIENCE, VOL. 50, NO. 2, 2010

Fuwa, M. 2004. “Macro-Level Gender Inequality and the Division of Household Labor in 22 Countries,” AmericanSociological Review, 69(6), 751–767.

Gardner, J., G. Pierre, and A. Oswald. 2001. “Moving for Job Reasons,” Department of Economics, University ofWarwick.

Greenaway, D., R. Upward, and P. Wright. 2002. “Structural Adjustment and the Sectoral and GeographicalMobility of Labour,” Discussion Papers 3662, CEPR.

Greenwood, M.J. 1985. “Human Migration: Theory, Models, and Empirical Models,” Journal of Regional Science,25(4), 521–544.

Gurak, D.T. and M.M. Kritz. 2000. “The Interstate Migration of U.S. Immigrants: Individual and ContextualDeterminants,” Social Forces, 78(3) (March, 2000), 1017–1039.

Hanson, S. and I. Johnston. 1985. “Gender Differences in Work-Trip Length: Explanations and Implications,”Urban Geography, 6(3), 193–219.

Henley, A. 1998. “Residential Mobility, Housing Equity, and the Labour Market,” Economic Journal, 108, 414–427.Madden, J. 1981. “Why Women Work Closer to Home,” Urban Studies, 18, 181–194.Manning, A. 2003. “The Real Thin Theory: Monopsony in Modern Labour Markets,” Labour Economics, 10,

105–131.Mauro, P. and A. Spilimbergo. 1999. “How do the Skilled and the Unskilled Respond to Regional Shocks? The

Case of Spain,” IMF, Staff Papers 46(1).McIntosh, J. and M. Munk. 2007. “Scholastic Ability vs. Family Background in Educational Success: Evidence

from Danish Sample Survey Data,” Journal of Population Economics, 20(1), 101–120.Mincer, J. 1978. “Family Migration Decisions,” The Journal of Political Economy, 86(5), 749–773.Ministry of the Environment. 2001. “Pendlingen i Danmark ar 2000 og Udviklingen i 1990’erne,” (Commuting in

Denmark year 2000 and the Development in the 1990’es). Copenhagen.Munch, J., M. Rosholm, and M. Svarer. 2003. “Are Homeowners Really More Unemployed?” The Economic Journal,

116(514), 991–1013.National Labor Market Authority (Arbejdsmarkedsstyrelsen). 2003. Undersøgelse af Flaskehalse pa det Danske

Arbejdsmarked 2003 (Investigation of bottlenecks in the Danish labor market 2003), PLS RAMBØLL Manage-ment, Oktober 2003.

Newey, W.K. 1987. “Efficient Estimation of Limited Dependent Variable Models with Endogenous ExplanatoryVariables,” Journal of Econometrics, 36, 231–250.

Nivalainen, S. 2004. “Determinants of Family Migration: Short Moves vs. Long Moves,” Journal of PopulationEconomics, 17(1), 157–176.

OECD. 2000. OECD Employment Outlook. OECD, June, Paris.Oswald, A. 1999. “The Housing Market and Europe’s Unemployment: A Non-Technical Paper,” The University of

Warwick, United Kingdom.Sabatier, M. 1999. “Search Channels, Intensity and Youth Unemployment Duration,” Conference Paper.Sandell, S.H. 1977. “Women and the Economics of Family Migration,” The Review of Economics and Statistics,

59(4), 406–414.Shihadeh, E. 1991. “The Prevalence of Husband-Centered Migration: Employment Consequences for Married

Mothers,” Journal of Marriage and the Family, 53, 432–444.Svarer, M. and H.S. Nielsen. 2006. “Educational Homogamy: Preferences or Opportunities,” Working Paper, No.

12, CAM.Swain, L., and S. Garasky. 2007. “Migration Decisions of Dual-Earner Families: An Application of Multilevel

Modelling,” Journal of Family and Economic Issues, 28(1), 171–182.Van Den Berg, G.J. 1992. “A Structural Dynamic Analysis of Job Turnover and the Costs Associated with Moving

to Another Job,” Economic Journal, 102, 1116–1133.Van Ommeren, J. 2000. “Job and Residential Search Behaviour of Two-Earner Households,” Papers in Regional

Science, 79(4), 375–91.Van Ommeren, J., P. Rietveld, and P. Nijkamp. 1998. “Spatial Moving Behavior of Two-Earner Households,”

Journal of Regional Science, 38(1), 23–41.———. 1999. “Job Moving, Residential Moving, and Commuting: A Search Perspective,” Journal of Urban Eco-

nomics, 46, 230–253.Van Ophem, H. 1991. “Wages, Nonwage Job Characteristics and the Search Behavior of Employees,” Review of

Economics and Statistics, 71, 145–151.Wayne T., D.J. Webber, and F. Walton. 2002. “The School Leaving Intentions at the Age of Sixteen: Evidence from

a Multicultural City Environment,” Economic Issues, 7(1), 1–14.Whisler, R.L., B.S. Waldorf, G.F. Mulligan, and D.A. Plane. 2008. “Quality of Life and the Migration of the

College-Educated: A Life-Course Approach,” Growth and Change, 39(1), 58–94.Zax, J.S. 1991. “The Substitution between Moves and Quits,” Economic Journal, 101, 1510–1521.Zax, J.S. and J.F Kain. 1991. “Commutes, Quits and Moves,” Journal of Urban Economics, 29, 153–165.

C© 2010, Wiley Periodicals, Inc.