Social Science Research 44 (2014) 1–14

Contents lists available at ScienceDirect

Social Science Research journal homepage: www.elsevier.com/locate/ssresearch

Individuals’ openness to migrate and job mobility Johannes Huinink a,⇑, Sergi Vidal a, Stefanie Kley b a b

Institute for Empirical and Applied Sociology, University of Bremen, Germany Department of Social Sciences, University of Hamburg, Germany

a r t i c l e

i n f o

Article history: Received 16 August 2013 Revised 24 October 2013 Accepted 24 October 2013 Available online 6 November 2013 Keywords: Job mobility Job search Migration Openness to migrate Life course

a b s t r a c t In this article we extend the scope of the interdependence between migration and job mobility: We investigate whether an individual’s openness to migrate not only increases the probability of migration but also the likelihood to conduct a job search and exhibit job mobility. Using data from a three-wave panel study, which allows the analysis of temporal links between decision-making and subsequent events regarding migration and job mobility, a joint estimation of multiple equations is performed. We show that considering migration as an option for the future, which is our indicator of individuals’ openness to migrate, is positively associated with both migration and job mobility. It even increases job mobility independently of whether migration takes place or not. These findings contribute significantly to our body of knowledge about the interdependence of migration and job mobility. Additionally, they enhance our understanding of the mechanisms behind a common selectivity of migrants and job mobile individuals. Ó 2013 Elsevier Inc. All rights reserved.

1. Introduction The interdependence between migration and job mobility has become a topic of great importance in the scholarly literature. According to the neo-classical economic model, people are willing to migrate if they can obtain a higher income or find a job elsewhere (more suitable to their qualifications), given that migration costs do not exceed the expected gains (Sjaastad, 1962; Schultz, 1964; Todaro, 1969; Speare, 1971; da Vanzo, 1981; de Jong and Fawcett, 1981). This means that migration is seen as an instrument with which to gain a (better) job when job opportunities are unevenly distributed across geographies.1 More recent economic research, however, has criticized that the empirical evidence might be partly attributed to selective migration (Borjas, 1999). This means that individuals who migrate may have generally more aspirations, skills or a more suitable pool of opportunities than those who stay. Sophisticated statistical models can account for effects of self-selection to estimate unbiased effects of migration on employment outcomes (e.g. Nakosteen et al., 2008). Less advancement has been made in understanding the selective factors that commonly lead to migration and job mobility. In our research we contribute to this neglected area of research theoretically and empirically. We extend the empirical analysis of the interdependence between migration and job mobility by explicitly addressing the relevance of early stages of each decision process. In particular, we investigate whether already being open to migrate – which is a behavioral disposition (Ajzen 2005: 3) – impacts the likelihood of job search and of exhibiting job mobility. Being open to migrate in our empirical analysis is indicated by the information that an individual is considering migration as an option for the future. ⇑ Corresponding author. Address: Institute for Empirical and Applied Sociology, University of Bremen, Celsiusstraße, FVG-Mitte, R 1010, 28359 Bremen, Germany. Fax: +49 421 218 67341. E-mail address: [email protected] (J. Huinink). 1 Studies considering that migration may induce job mobility (Blau and Duncan, 1967; Linneman and Graves, 1983) are less numerous. It was shown that migration can have positive effects on individual occupational achievement in the long run (van Ham, 2002; Mulder and van Ham, 2005) and effects of family migration on job careers of men and woman have been studied (e.g. Boyle et al., 2009; Shauman and Noonan, 2007). 0049-089X/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ssresearch.2013.10.006

2

J. Huinink et al. / Social Science Research 44 (2014) 1–14

Individuals who are merely considering migration are explicitly distinguished from the group of those who are already planning it. Making concrete plans indicates that individuals have already decided in favor of migration – for example as a consequence of a job change – as previous, theory-guided research suggests (Kalter, 1997; Kley, 2011). Our model is based on the usual understanding that ‘‘migration is a change of residence that disrupts the basic ties with the local community and is a move that prevents commuting at least at any reasonable cost’’ (Clark, 1986: 20). Beyond the differences between movers and stayers in regard to employment outcomes, we already expect differences between those who are open to move and those who are not, because individuals’ openness to migrate likely coincides with higher aspirations and perceived opportunities elsewhere. Therefore, those open to migrate may be more likely to accomplish their (job-related) goals than those who are not open, independently of whether they actually migrate or not. These expectations are formulated in theory-based hypotheses that we derive on the basis of an enriched subjective expected utility (SEU) model (Ormel et al., 1999). To test this hypothesis we use data from a three-wave panel study gathered between 2006 and 2008 in two German towns with respondents aged 18–50 years at the time of the first survey (Huinink and Kley, 2011). We analyze the temporal dynamics of considering migration as an option for the future on both job mobility and migration. To achieve this we perform a joint estimation of multiple equations (i.e. multiprocess estimation) by combining hazard rate estimation with a categorical outcome estimation of panel data (Lillard and Panis, 1996). The consistency of our results is tested by the application of Markov Chain Monte Carlo methods (Browne et al., 2009). The findings show that considering migration as an option for the future is positively associated with job mobility. More specifically, openness to migrate has positive effects on job search behavior and increases job mobility rates independently of whether migration takes place or not, as expected. More generally, we find exemplary support for the assumption that an individual disposition regarding one life domain (migration) impacts behavior in another dimension of the life course (employment). 2. Previous research In their groundbreaking work, Blau and Duncan devoted a whole chapter to the relationship between migration and job mobility. They were probably the first to introduce behavioral dispositions towards migration into the discussion: ‘‘A man’s economic chances are improved by his motility, that is, his not being rooted to his place of birth but free to leave it for better opportunities elsewhere’’ (Blau and Duncan, 1967: 250). Motility refers to the ‘‘capacity to move’’ and, for these authors, ‘‘migration is simply an operational measure’’ of it. They deduce its significance from the empirical evidence that (interregional) migrants outperform to non-migrants in regard to occupational success. Blau and Duncan were well aware of the question of the correct direction of causality and selective migration. They found more evidence for selective migration of individuals with greater potential for occupational achievement but did not rule out that migration could promote the career simply because it improves the opportunity structure of job mobility for migrants (Blau and Duncan, 1967: 259, 274). Other studies also show that migration can trigger job mobility (e.g. Linneman and Graves, 1983; van Ham et al., 2001; van Ham, 2002; van Ham and Hooimeijer, 2009). Differences in local housing conditions or life-cycle-specific changes in the demands of local-specific amenities (e.g. because of the birth of a child) might be a reason to leave town leading to job mobility ‘‘to optimally coordinate job and residence sites’’ (Linneman and Graves, 1983: 265). Linneman and Graves (1983: 275) conclude ‘‘that the migration and job change decisions are interrelated and more complicated than previously recognized’’. Even though these authors do not look into the decision-making process explicitly, their analysis gives reason to expect that individuals’ openness to migrate plays an important part in improving the occupational career, because it enhances the likelihood of actually finding a (new) job. Referring to theoretical considerations offered by Simpson (1992), van Ham proposes conceiving of spatial mobility in general, i.e. short- and long-distance moves as well as circular mobility like commuting, as an instrument of occupational achievement (van Ham, 2002: 6). He and his colleagues address the relevance of behavioral dispositions to match job opportunities across space by what they call ‘‘spatial flexibility in job mobility’’ (van Ham et al., 2001; van Ham, 2002; van Ham and Hooimeijer, 2009). They define it as the ‘‘possibility of accepting a job at a greater distance’’ given more or less willingness and constraints to migrate or commute (van Ham et al., 2001: 924). A higher spatial flexibility, they argue, should positively affect the readiness to accept a new job at a greater distance, which either has migration as a consequence or implies long-distance commuting.2 In the rich body of previous literature, individual and household characteristics such as age, gender, education, and household composition have been identified as important predictors of spatial flexibility and the likelihood of accepting a new job at a greater distance. Van Ham et al. (2001) found that in the case of job mobility, age has a negative and educational achievement a positive effect on the likelihood that it is accompanied by increasing the distance to the workplace. Married women and mothers commute or perform a job-related migration less often than unmarried women without children (Hanson and 2 Indeed, as well as migration, commuting also has to be addressed in research on linking spatial mobility and job mobility (e.g. Simpson, 1980; Evers, 1989; Kalter, 1997; van Ommeren et al., 1999). The commuting distance positively impacts job mobility. Commuting distance is also positively correlated with the likelihood to migrate and migration can lead to a change in commuting distance (see Zax and Kain, 1991; Clark et al., 2003). Therefore, even though our study does not focus on commuting, we have theoretical reasons to account for it in our analyses. It is plausible to assume that it is positively correlated with an individual’s openness to migrate (van Ham et al., 2001).

J. Huinink et al. / Social Science Research 44 (2014) 1–14

3

Pratt, 1995; Clark and Withers, 1999; Clark et al., 2003). Married men and fathers are less likely to migrate for job reasons and commute more often than single men (van Ham et al., 2001: 935). To summarize: Previous research has made substantial conceptual contributions to the understanding of how migration can trigger job mobility, offering solid inputs to explain mechanisms of selective migration. However, due to lack of data, particular factors driving the potential selectivity behind higher rates of job mobility have not been explicitly analyzed so far. To take a step in this direction we ask whether a behavioral disposition to accept opportunities elsewhere, what we call individuals’ openness to migrate, partly explains higher rates of job mobility. In our analysis we assess its contribution to explaining job search behavior and job mobility in addition to other factors like age, gender, education, household composition, and commuting behavior.

3. Theoretical considerations and main hypotheses Migration and taking up a new job constitute important transitions in individual life courses. They can be understood as ‘‘instrumental goals’’ for generating subjective wellbeing (Ormel et al., 1999; Lindenberg, 2001). Major transitions in the life course with considerable consequences can be assumed to be the result of a stepwise decision process (Heckhausen and Gollwitzer, 1987). Building on previously developed migration models, e.g. by Brown and Moore (1970), Speare et al. (1975), De Jong and Fawcett (1981), and Kalter (1997), we distinguish between (1) merely considering migration as a promising welfare improvement tool, (2) planning a particular migration, and (3) realizing the migration.3 Considering migration as an option for the future is an indicator of individuals’ openness to migrate. It is conceived as a kind of behavioral disposition, i.e. a behavioral attitude individuals can have without having concrete plans to move. An attitude is generally defined as a disposition ‘‘to respond favorably or unfavorably to an object, person, institution, or event’’ (Ajzen, 2005: 3). Migration might be such an event. Attitudes are viewed to be quite stable but more ‘‘malleable’’ than other dispositions like personal traits (Ajzen, 2005: 6). Evaluative as they are, they are context-dependent. Therefore, the openness to migrate as a behavioral attitude might be influenced by the perception that regional living conditions in relevant life domains are better elsewhere than at the current place of residence (Kalter, 1997; Huinink and Kley, 2008; Kley, 2011). Openness to migrate can also be supported by feeling less committed to the place one lives and experiencing less constraints or perceiving lower cost to migrate. Entering the planning stage means taking the next step. It indicates that the decision in favor of migration has been made. This decision was shown to be primarily driven by the appearance of a concrete opportunity elsewhere, e.g. a promising job (Kalter, 1997; Kley, 2009, 2011; Kley and Mulder, 2010). Taking up a new job exerts a significantly greater influence on planning than on considering a migration. In our model we conceive those who are just considering migration – for whatever reason – but are not yet planning a move as a group of individuals who are open to migrate but have not made a firm decision about whether migration will be performed or not. Only a certain proportion of the individuals who are in the stage of considering migration will later enter the planning stage. Individuals may also stop considering migration – maybe because they lower their aspirations in regard to local living conditions or because they perceive high constraints with regard to a change of residence. One can model the decision process regarding job mobility in a similar way. Job search activities, as a precursor of job mobility, can be perceived as an indicator that someone plans to change their job. The decision to implement a biographical transition – like taking up a new job or migrating – is assumed to depend on expected welfare gains obtained by this transition. As a general model, we apply a version of the subjective expected utility model (SEU). Given alternative goals to pursue in a certain situation, individuals decide to work on the goal that promises the highest subjectively expected utility (welfare gain) for the future, e.g. by performing a certain biographical transition. To specify the concept of expected utility we adapt elements of expectancy theories of motivation (Vroom, 1964; De Jong and Fawcett, 1981; Schoemaker, 1982; Fawcett, 1985; Lindenberg, 2001; Huinink and Schröder, 2008). The expected utility or welfare gain is assumed to be positively correlated with (a) the salience of a welfare goal, and (b) the perceived gap between the current and the pursued level of achievement of that goal. Additionally, the expected utility is (c) negatively correlated with the situational constraints and costs of achieving the goal (efficiency) and (d) positively correlated with the expected likelihood of achieving it. Following this approach, the likelihood of an individual’s dissatisfaction with their current job and of looking for an alternative job should be higher the more important occupational success is valued, the higher aspirations in regard to working conditions and other outcomes are, and the higher the expected probability of finding a better job is. Here the individual’s openness to migrate comes into play. It opens up options for improving an individual’s employment situation and it supports an individual’s higher aspirations in regard to their occupational career. Individuals who are open to migrate feel less restricted to their local opportunities and are more motivated to pursue opportunities elsewhere or to make use of an opportunity once it arises. Openness to migrate should deter individuals from adapting their aspirations to what they think the local conditions allow them to achieve. Therefore, on average, openness to migrate should increase the likelihood of an individual searching for a better job (cf. Lippman and McCall, 1976). In addition, if someone is open to migrate, the job search scope is large and his or her chances of 3 Kley (2011) refers to the Rubicon model (Heckhausen and Gollwitzer, 1987) for a psychologically-based explanation of these stages. The model does not apply in cases the individual is not able to reflect on decision-making (forced moves).

4

J. Huinink et al. / Social Science Research 44 (2014) 1–14

actually finding a new job outside the city boundaries are higher than among those who are not open to migrate.4 However, a job search increases the probability of finding attractive job offers at any distance, given that an individual has not yet decided to move and, therefore, considers only distant jobs. Following the rule of efficiency, individuals should prefer the job closest to the current place of residence to avoid costs of migration or long-distance commuting (Simpson, 1992; Eliasson et al., 2003; van Ham, 2002). Therefore, individuals’ openness to migrate should increase the rate of job mobility in general and not only in cases where the new workplace is located at a greater distance. Additionally, the effect of openness to migrate should not be fully captured by increased job search activity as not all job changes are preceded by a job search. Our theory says that individuals who are open to migrate should be more likely to identify and accept suitable opportunities for a job change that may come up without explicitly having searched for a job. Our theoretical arguments raise some objectionable points. One could be that a desire to migrate forces individuals to become open for job mobility – even though they might be content with their current job (Morrison and Clark, 2011). This can be the case, for example, when people intend to migrate for reasons like partnership formation, partnership dissolution, or childbirth. However, previous research has shown that moves for family reasons are mainly short-distance moves (Mulder, 1993; Kley, 2009) and these seldom imply job mobility (van Ham et al., 2001; Huinink and Kley, 2008; Kley, 2011). In addition, we argue that planning a particular migration provides a stronger reason for considering a new job than merely being open to migrate. Therefore, in our model we account for whether individuals are in the planning stage or not. Another objection could be that discontent with the current job situation can be a reason to become open to migrate. If this is the case, accounting for job search as an indicator of job mobility intentions should rule out this possible source of bias in the effect of openness to migrate on job mobility. However, the association between job search and openness to migrate must be assumed to be circular. This endogeneity problem could only be tackled with explicit measures of job change intentions. Unobserved time-invariant subjective factors, like value orientations, other attitudes, personal traits and abilities, also mediate the associations between job mobility and migration processes (Courgeau, 1985; Oddland and Shumway, 1993; Mulder, 1993; van Ham, 2002; Mulder and van Ham, 2005). For instance, not controlling for individual characteristics, like work orientations, may inflate the positive association between openness to migrate, migration, and job mobility. To deal with unobserved influential time-invariant personal traits that commonly determine migration, job search, and job mobility (e.g. sociability, etc.) without invoking any direct influence of individuals’ openness to migrate, we apply joint multiple equation estimation. Other research has used this method in similar analyses (Detang-Dessendre, 1999; Detang-Dessendre and Mohlo, 1999). Since migration and job mobility probabilities are estimated simultaneously, no direction of causality between the processes is imposed. Concluding this discussion we state as our main hypothesis that openness to migrate works as a trigger for job search and job mobility. In particular we expect that: (1) Individuals’ openness to migrate is positively correlated with searching for a job at a greater distance. (2) Individuals’ openness to migrate has a positive effect on the likelihood of job mobility – even after accounting for job search. The effect is independent of whether job mobility is connected with migration or not. Realizing migration, then, is foremost a consequence of job mobility to avoid high commuting and other costs. Finally, we expect that the directions of all the effects commented above are true after controlling for personal traits (i.e. unobserved heterogeneity) that commonly affect migration, job search, and job mobility.

4. Data, variables, and methods 4.1. Data set We use data from a three-wave panel study gathered between 2006 and 2008 in two German towns, Magdeburg and Freiburg, with respondents aged 18 to 50 years at the initial interview (Huinink and Kley, 2011). The two cities, one in Eastern (Magdeburg) and the other in Western Germany (Freiburg), differ in economic conditions5 but are otherwise similar in important respects. They were selected on grounds of a classification of European city regions (Rozenblat and Cicille, 2003) to ensure their comparability with regard to the basic spatial distribution and concentration of opportunities (see Kley, 2009, for a thorough discussion). Both cities are of comparable size (about 200,000 inhabitants); both have universities and are not near another agglomeration within short commuting distance. Therefore, it seems justified to pool the data of both cities; the information whether a respondent lived in Magdeburg or in Freiburg in the first wave will be considered in the analysis. The fact that we have data from medium sized cities – and not from very big cities or from rural areas – should not make much difference regarding our hypotheses. On theoretical grounds, there is little reason to assume that the relationship proposed in our hypotheses should be substantially different in different geographical contexts. 4 We learned from the literature that not only openness to migrate but also commuting tolerance is part of ‘‘spatial flexibility’’ (van Ham et al., 2001). Men and women who are open to commuting longer distances may be motivated to look for a new job and to accept the possible need to commute over a longer distance. We cannot analyze this directly because of missing data on individuals’ willingness to commute. But consistent with van Ham and others we argue that age, education, and actual commuting behavior are good ‘‘instruments’’ controlling for willingness to commute. 5 For instance, in 2007 the unemployment rate was seventeen percent in Magdeburg but only nine percent in Freiburg; the average annual income per employee was €21,000 in Magdeburg and €26,000 in Freiburg. Previous research revealed that significantly more Magdeburg residents perceived their career and income prospects to be better elsewhere than in their city of residence, compared to respondents living in Freiburg. But the mechanisms explaining migration decision-making and actually moving were proven to be the same in the two cities (Kley, 2009).

J. Huinink et al. / Social Science Research 44 (2014) 1–14

5

All data was collected in Computer Assisted Telephone Interviews (CATI). Using the Random Digit Dialing method (Gabler and Häder, 2002), stratified samples of people aged 18 to 50 years living within the city boundaries of Magdeburg and Freiburg for at least twelve months were created at the beginning of 2006. For each of the cities, the sampling scheme envisaged strata consisting of 600 respondents who were not considering leaving town, and 600 respondents who were considering leaving town. A comparison of the main demographic characteristics with official statistics (Huinink and Kley, 2011) revealed an over-representation of women and persons aged 18–30 years, whereas men and persons aged 30–39 years are underrepresented. Respondents with lower educational attainment and non-German citizens are also underrepresented. This fact might lead to an over-estimation of the proportion of persons considering leaving town in the population. The first wave of the panel comprised a total sample of 2410 interviewees of whom 2288 agreed to take part in follow-up interviews.6 The second and third waves took place about one and two years after the first interview, yielding an observation window of about three years.7 The response rates in waves two and three were between 69% and 75%. In this article, data on 1172 respondents from the two cities who took part in all waves and who provided complete information on the main variables are used. A table with summary statistics can be found in the Appendix. As panel attrition is not randomly distributed across groups of respondents, we use longitudinal weights in the regression analysis (Huinink and Kley, 2011). However, as sensitivity tests show, results do not depend on weighting.8 We also successfully account for variables that have proved relevant for predicting attrition. Previous analyses of the migration process with these data using a Heckman selection model to control for panel attrition also showed no harmful effects of attrition (Kley and Mulder, 2010; Kley, 2011). 4.2. Variables In our analysis we consider three interdependent outcomes: migration, job mobility, and job search activities. Migration is measured as the first move beyond the city boundaries within the observation window (cf. Mulder and Hooimeijer, 1999). Ninety percent of the migrants from both towns moved a distance of more than 50 km. That is a threshold that is often used to distinguish migration from (local) residential moves in Europe (e.g. Long et al., 1988). The proxy ‘‘leaving town’’ therefore adequately captures migration in the sense of Clark (1986). Further migration events of the same person are not considered; they are very rare in this three-year panel study. Job mobility is operationalized as taking up a job, independent of being already employed or not. A job episode starts at the beginning of a job and ends at the beginning of the next one. This definition of an episode may include unemployment or other activities in case the individual does not make a direct transition from job to job. As periods of non-employment (as homemaker, pensioner or unemployed person) and periods of enrollment in education (in school, higher education or training) may affect the transition rate to a new job differently compared to periods of employment, ‘‘non-employment’’ and ‘‘in education’’ are included as time-varying covariates in the regression models.9 ‘‘Employed’’ includes both full-time and part-time employment, encompassing wage and salary earners as well as the self-employed. Marginal part-time jobs held by pensioners or persons enrolled in education are not part of this category. Moreover, transitions to a new job as secondary employment (e.g. part-time employment) while still holding a main job are not considered. Retrospective information allows us to track the month of job start and job end (if not censored) for all positions the interviewees held within the observation window. Several events of job mobility per individual are observed and modeled. Some of them occur before and some after a migration. Job search activities within the three-year observation window are tracked on a monthly basis from the retrospective record. Additionally, the respondents reported the area of their search activities on a 5-point scale ranging from local to over 100 km. The scale was dichotomized into search activities within a radius of up to 50 km and a radius beyond the 50 km threshold, in accordance with our measure of migration. In the models estimating job search activities, the exact monthly information is used. In the models estimating job mobility and migration probabilities, any job search activities that are currently taking place or have recently taken place within the last three months are considered as predictors. Our main independent variable, openness to migrate, is operationalized as considering migration out of town as an option for the future. In the first wave of the study, all respondents were asked whether they recently had considered leaving the city to live somewhere else.10 Those who answered this question affirmatively were asked whether they had plans to do so within the next twelve months.11 After the first wave, these questions were posed again in the two subsequent panel waves as well as in two extra follow-up questionnaires between the panel waves, as long as no move out of town was observed. 6

The response rate in the first wave was 52% in Magdeburg and 47% in Freiburg, with up to eight attempts to contact the target person. To avoid left-censoring of retrospectively recorded episodes of education, (un-)employment, and living arrangements the starting date of the status held at the beginning of the panel study was recorded. 8 As a sensitivity test, we also re-estimated the model (1) with various extreme weighting schemes by simulating random weights created from a uniform distribution and (2) with no weighting to test the stability of the coefficients. We found no considerable bias comparing the models. 9 We discarded the idea of simultaneously modeling transitions between employment, unemployment and other activity statuses because this would have led to a large transition matrix and made the analysis inefficient. 10 The exact wording in the questionnaire was, translated from German: ‘Have you recently thought about moving away from Magdeburg/Freiburg to live somewhere else?’ 11 The exact wording in the questionnaire was, translated from German: ‘Do you have plans to move away from Magdeburg/Freiburg within the next twelve months?’ 7

6

J. Huinink et al. / Social Science Research 44 (2014) 1–14

The data on consideration of migration or making plans show enough within-individual variation to be used as a time-varying covariate that switches every time a new status is reported. We apply duration analysis for migration as well as for job mobility. The duration of residence in town is measured in months since birth if the respondent has always lived there, or since arrival in town if the respondent moved there. The duration until the start of a new job is measured in months since the last job change in case of previous employment, or since age 18 in case of no previous employment. Other measures are used as covariate controls in the regression analysis. Among demographic characteristics, we account for gender as well as a mean-centered measure of age. A dummy indicator of higher educational attainment (i.e. university or technical college degree vs. lower educational attainment) works as a proxy for human capital. Another type of relevant human capital investment is prior migration experience, which also indicates a disposition like self-efficacy that may lead to reconsidering and repeating moves (Detang-Dessendre, 1999). The variable indicates whether someone had previously moved across the city boundaries or not. The retrospective record allows us to track the current job status (employed vs. non-employed), the job episode order since January 2006, and job-related characteristics for each episode. For the first and second (or higher order) job episodes observed, different time-varying indicators were constructed. Each episode starts at the time the job begins – in most cases before January 2006 – and ends when the next one begins or with censoring, as explained above. Based on the latest research findings, we also included commuting behavior in our model to account for the effect of current commuting costs on both job mobility and migration (van Hamm, 2002; Clark et al., 2003). The respondents reported the time traveled one way in minutes and the frequency of traveling to work. We constructed a time-varying indicator that reflects relatively high costs of commuting; the frequency must be more than once a week and the travel time over one hour each way, following common practice in the literature (Schneider and Meil, 2008). A travel time above one hour approximates the distance that separates the local labor market from other labor markets in alternative regions. In the models estimating job mobility and migration, a variable indicates the time passed since the respondent terminated his or her last employment until he or she found a new job, in order to account for time dependencies. Migration decisions are frequently taken at the household level, where the occupational careers of the partners, children’s schooling and other private matters are considered. The retrospective record also permits the construction of family-related indicators for partnership (ref: single) and whether the respondent has children (ref: no children) in every month of observation. As we cannot model the influence of other household members on the respondent’s decision-making process directly,12 we have to rely on these indirect indicators. In general, the expectation is that couples and family households experience higher constraints on mobility. Other common predictors used in migration research, like home-ownership or existing social networks, proved to only add redundant information to the model according to efficiency tests (AIC). Therefore, we discarded them from the analysis. Obviously, other variables in our model, in particular previous migration experience, are confounded with both home-ownership and social networks. Poor satisfaction with employment opportunities in town can also be correlated with the desire to change jobs, as can poor satisfaction with some other local amenities, which may prompt the desire to leave town. To control for this possibility in our models, a scale measuring perceived opportunities in town was constructed on the basis of the individual importance of life goals and the perception of opportunities to achieve these goals at the current place of residence. In the first wave, respondents reported the importance of their individual life goals relating to partnership, income, hobbies, children, occupation, health, social contacts, and standard of living on seven-point Likert scales. Then they were asked to what extent they could accomplish each of these goals at their current place of residence, measured accordingly. For each life domain, the individual importance score was subtracted from the individual accomplishment score and summed for all life goals. Low values indicate low perceived opportunities in town compared to personal ambitions whereas high values mean good perceived opportunities for the respective person. Additionally, searching for a job was included as a time-dependent indicator in the models for job mobility and migration. It captures search behavior during the three months prior to event times (t-3). 4.3. Methods Data analysis is based on regression methods for longitudinal data. We jointly estimate (i.e. multiprocess estimation) a mixture of hazard regression for discrete durations and panel models for competing outcomes similar to other research (cf. Lillard and Panis, 1996). We consider three outcomes for which we model the measures introduced above: (1) migration (i.e. moving beyond the city boundaries), (2) job mobility (i.e. getting a new job), and (3) job search (i.e. searching within a radius of up to 50 km from the place of residence, searching beyond a radius of 50 km). Modeling three outcomes allows to represent three interrelated pathways in which an individual’s openness to migrate affects job mobility, namely: directly, through job search, or as a consequence of migration. RM

log

12

hti 1

RM hti

!

T

T

¼ aRM ðtÞ þ bRM xti þ cRM wti þ dRM i

Migration intentions of the partner were only asked in the first wave.

ð1Þ

J. Huinink et al. / Social Science Research 44 (2014) 1–14

!

JM

log

log

htji

JM

1  htji

T

T

¼ aJM ðtÞ þ bJM xtji þ cJM wtji þ dJM i

7

ð2Þ

 r T T yti ¼ ar0i þ br xti þ br wti þ dri ; r ¼ 1; 2 y0ti

ð3Þ

where r = 1 if the search radius 650 km, r = 2 if the search radius >50 km. The first two outcomes are estimated as discrete time hazard models – Eqs. (1) and (2) – (Allison, 1982), where the hazard hti is defined as the conditional probability of an event during interval t for individual i given that no event occurred in the RM JM previous interval. hti denotes the hazard of residential change during interval t for individual i and htj represents the risk of starting a new job episode j during interval t for individual i.13 The discrete hazard is a response probability for a binary variable; thus, we use a logit model for the estimation after restructuring the data to define a dichotomous censoring variable, which takes the value 1 if the event is observed in interval t and the value 0 otherwise.14 Time dependence of the hazard function is captured in a(t) that is a linear function of duration of residence in the migration equation. For job mobility, it is a linear spline estimation of slope effects of episode duration for the first 12 months and after 12 months. Observable interdependence between processes is captured in the model by allowing outcomes of each equation to be predictors for the other outcomes. The vector wti includes the outcomes of the other equations as time-varying statuses. Other time-varying and time-constant covariates are included in the vector xtji. Finally, di is a normally distributed individual-specific random effect that will account for time-fixed unobserved heterogeneity. The inclusion of the individual level stochastic terms for all equations is justified by the fact that we aim at simultaneously estimating their variances and covariances to make all processes (i.e. equations) involved interdependent. Job search activity is modeled as a multinomial logit model in a longitudinal framework (Eq. (3)). We estimate the individual odds of engaging in job search activities within a local (r = 1) or a wider search radius (r = 2) in time t to the odds of not engaging in any search activity (r = 0) in the same period. In the equation, ar0i captures the intercept for each competing risk at the beginning of the observation window. The rest of the notation of the equation can be read as commented for the other equations. The problem of the independence of irrelevant alternatives for multinomial logit applications is ruled out by allowing each outcome to have its own but correlated random term (Steele et al., 1996). Multiprocess estimation, outlined first by Lillard (1993), is a widespread strategy in a hazard regression setting (e.g. Lillard and Panis, 1996; Baizan et al., 2003; Billari and Philipov, 2004; Kulu, 2008). The outcomes of the processes to be estimated are allowed to influence each other by relevant characteristics that we cannot include in the model. The strategy is based on a maximum likelihood estimation of the correlation between individual-specific random effects for each estimated outcome. Random effects are extracted from a joint multivariate normal distribution with as many draws as outcomes in the model and are meant to capture time-invariant unobserved heterogeneity. Since the multivariate normal distribution does not allow for a closed form of the likelihood, we overcome this by integrating out the random terms from the likelihood by numerical integration. We use the software aML 2.0 (Lillard and Panis, 2002), which contains a numerical integration algorithm based on Gauss-Hermite Quadrature.15 Eq. (4) shows the error structure matrix of our equations, where there are four r2d terms, which denote the variances of the person-specific residuals to be estimated, as well as the six covariances between all residuals, which are denoted by the qdd terms.

0

dRM i

B JM B di B B r1 @ di dr2 i

1

00 1 0 C BB 0 C C B B C  N BB C C; C @@ 0 A A 0

0

r2dRM B q RM JM B d d B @ qdRM dr1 qdRM dr2

qdJM dRM rdJM qdJM dr1 qdJM dr2

qdr1 dRM qdr1 dJM r2dr1 qdr1 dr2

11

qdr2 dRM C qdr2 dJM C CC CC qdr2 dr1 AA rdr2

ð4Þ

The identification of unobserved heterogeneity is possible due to the observation of multiple episodes per individual in each hazard model and status variation in the job search model. However, in our model we observe migration only once and then the variance of the residual term is weakly identified. Here, we follow the suggestion made by Baizán et al. (2003) or Billari and Philipov (2004) to fix the residual term to a given level, i.e. not estimating it, but to allow for correlation with the estimated individual-specific residuals of the other outcomes. As a sensitivity test, we ran the models using Markov Chain Monte Carlo (MCMC) methods for generalized linear mixed models (Hadfield, 2010). MCMC is a Bayesian-like method that allows for approximating the integral for the random effects for dependent variables that are non-Gaussian, like indicator variables, not basing them in arbitrary numerical integration approaches. However, as MCMC requires very lengthy computing time it was only used to evaluate the point estimates 13

The hazard for the job mobility equation includes the suffix j because we identify several job episodes per individual. We minimize common biases due to right-censoring and left-truncation as we are able to use complete information on both the entry and exit for each outcome and for all observations in the analysis (Guo, 1993). 15 As the likelihood function has no closed-form solution, we have to integrate the residual out using a numerical integration algorithm. aML offers an algorithm based on Gauss-Hermite Quadrature where the default is to approximate a multivariate normal distribution (i.e. three or more dimensions) by 6 integration points in each dimension. The final models were also tested by increasing the number of integration points to 20, with an equivalent exponential increase of computing time, but the substantive interpretation of the results remained unchanged. 14

8

J. Huinink et al. / Social Science Research 44 (2014) 1–14

and confidence intervals of our final choice model results. We used the free software R and the library MCMCglmm (Hadfield, 2010) in the R environment to run the sensitivity test. The estimates from MCMC are more robust than those performed by quasi-likelihood estimates typical for random effects logistic regressions/discrete hazard models (Browne et al., 2009). In particular, averaging results of chained estimations by MCMC will help to confirm whether the coefficients resulting from numerical estimation with aML are plausible within confidence intervals that were constructed with MCMC. 5. Findings An initial descriptive analysis of the data reveals a positive correlation between considering and planning migration in the first wave of the survey with conducting a job search, job mobility, and migration (Huinink et al., 2011). Moreover and most importantly for our analysis here, it was found that considering migration is associated with job mobility independently of whether migration takes place later or not. Entering the stage of planning migration for whatever reason greatly increases the likelihood of realizing it. These descriptive results initially support our hypotheses about the relevance of individuals’ openness to migrate for job mobility. Motivated by these findings, we make use of multiprocess estimations of the three outcomes (job search, job mobility, and migration), which allow us to link different trajectories and control for other important variables as well as for latent heterogeneity correlated with our dependent variables. 5.1. Unobserved heterogeneity In Table 1 we present correlations among the residual terms for the different outcomes in order to show the degree of interdependence between the processes after controlling for the set of predictors commented above. Since the results of simultaneous estimation show that all residual correlation terms are statistically significant, in the rest of the section we will only present the results of such models because uncorrelated residual models are biased. Regarding the signs of the residual correlation coefficients, Table 1 shows the expected effects of unobserved heterogeneity at the individual level on the association of the outcomes. The correlation between migration and job mobility is positive, confirming that unobserved individual propensities exist that select job mobile individuals towards migration. A job search, no matter what the search radius, is also positively associated with migration. This confirms that individuals prone to migrate are more prone to engage in search activities, too, whereas the scope of the search is mostly irrelevant. In accordance with what one would expect, the positive correlations between heterogeneity components of job search and job mobility are very strong. And finally, we observe a negative association in the heterogeneity of searching for jobs locally and in a broader area. This suggests that persons seldom commit equal search efforts to a new job locally (650 km) compared with far away (>50). 5.2. Job search We discuss now results of the job search equations’ estimation (see Table 2). When we allow job search activity to depend on unobserved individual features that also affect migration and job mobility, we find no significant association between considering migration and a local job search and a small but significant coefficient in the case of a long-distance job search. Individuals with a higher propensity for migration are more likely to engage in search activities. Planning migration is positively associated with job search; weakly in regard to a small radius and strongly in regard to a large radius. Because it is possible that the decision to migrate is not conditional on a job offer or that migration might have non-occupational reasons, the search for a job at the destination might be part of the planning stage of migration. The effects of possible confounders with considering migration are as expected. After a migration took place, we observe a decline in the probabilities of engaging in any type of search activity. We may conclude that migration is not speculative, since the job search activity is concentrated in the pre-migration stages. As one would expect, being employed or being enrolled in education are – compared to being non-employed – negatively associated with job search. Long-distance commuting obviously motivates a job search locally, which seems plausible. Having a high level of human capital, i.e. higher

Table 1 Unobserved heterogeneity: Variances and correlations. Variances

Migration Job mobility Job search (650 km) Job search (>50 km) ⁄ ** ***

Significance level: 0.1. Significance level: 0.05. Significance level: 0.01.

1 0.47*** 8.98*** 6.68***

Correlations Migration

Job mobility

Job search (650 km)

Job search (>50 km)

1 0.34** 0.17*** 0.24***

– 1 0.56*** 0.56***

– – 1 0.11***

– – – 1

9

J. Huinink et al. / Social Science Research 44 (2014) 1–14 Table 2 Multinomial logit coefficients of job search activities at short and long distance (reference category: no search). Short distance: 650 km radius Coef.

Long distance: >50 km radius S.E.

Coef.

S.E.

Before migration Not considering/planning migration Considering migration Planning migration After migration

Ref. 0.00 0.31* 0.35***

(0.10) (0.17) (0.13)

Ref. 0.38*** 1.56*** 1.99***

(0.14) (0.14) (0.11)

Non-employed Employed (1st episode) Employed (higher episode) Commutes (over 1 h)

Ref. 1.06*** 1.99*** 0.51**

(0.05) (0.14) (0.22)

Ref. 1.40*** 2.79*** 0.41

(0.07) (0.12) (0.29)

High education In education Partnership Children

0.63*** 0.01 0.72*** 0.97***

(0.14) (0.10) (0.12) (0.20)

1.11*** 1.37*** 0.06 1.39***

(0.12) (0.09) (0.11) (0.30)

0.27 0.05*** 0.73*** 0.41** 1.58*** 34,148

(0.17) (0.01) (0.20) (0.19) (0.18)

3.34*** 0.07*** 1.22*** 1.65*** 1.46*** 34,148

(0.32) (0.01) (0.21) (0.20) (0.20)

Previous migration Perceived opportunities in town City (Freiburg) Age (over 35) Sex (female) N

N = individual-month observations. Coef. = coefficients, S.E. = standard errors. Interdependent models include individual-specific random terms for each equation, which are allowed to be correlated among themselves and with the random terms of the migration and job mobility equations. * Significance level: 0.1. ** Significance level: 0.05. *** Significance level: 0.01.

Table 3 Discrete time log-hazards of job mobility and migration. Job mobility Coef.

Migration S.E.

Before migration Not considering/planning migration Considering migration Planning migration After migration

Ref. 0.24* 0.47*** 0.21

(0.15) (0.17) (0.19)

Non-employed Employed (1st episode) Employed (higher episode) Commutes (over 1 h) Search job (btw. t and t-3)

Ref. 2.49*** 1.75*** 0.78* 1.99***

High education In education Partnership Children Previous migration Perceived opportunities in town City (Freiburg) Age Sex (female) Baseline duration (until new job) Up to 12 months 12 months or more Residence episode duration N

Coef.

S.E.

Ref. 1.62*** 4.03***

(0.50) (0.42)

(0.18) (0.35) (0.46) (0.21)

Ref. 0.27 0.03 0.39 1.66***

(0.26) (0.46) (0.80) (0.31)

0.35** 0.45*** 0.08 0.04 0.08 0.01 0.06

(0.16) (0.15) (0.14) (0.18) (0.15) (0.01) (0.15)

0.45 0.17 0.01 0.16 0.03 0.02 0.02

(0.29) (0.26) (0.28) (0.37) (0.46) (0.02) (0.30)

0.004*** 0.02

(0.00) (0.13)

0.01*** 0.19

(0.00) (0.26)

0.06** 0.001*

(0.03) (0.00)

34,148

0.00 30,932

(0.00)

N = individual-month observations. Coef. = coefficients, S.E. = standard errors. Job mobility refers to getting a full-time or part-time job conditional on ending previous employment (if applicable). Migration refers to the first long-distance move of more than 50 km. t refers to monthly units of time. * Significance level: 0.1. ** Significance level: 0.05. *** Significance level: 0.01.

10

J. Huinink et al. / Social Science Research 44 (2014) 1–14

education, is generally positively associated with job searches. Living in a union has a positive effect on local job searches only, whereas children decrease the probability of searching for jobs at any distance. Interestingly, previous migration experience strongly supports long-distance job searches. Similar to individuals’ openness to migrate, this might be due to the fact that migration experience encourages individuals to look for jobs that may require migration. The more opportunities a respondent perceives in town, the less likely they are to search for a job elsewhere. The estimated effect for the short-distance job search points in the same direction, because respondents living in Freiburg, the more prosperous of the two cities, seldom engage in job search activities at a greater distance, even if they perceive the local opportunities as scarce. Older respondents prefer local job searches. Finally, women are more likely to search locally than men. This effect can be explained by a lower spatial flexibility among women due to household responsibilities (Hanson and Pratt, 1991, 1995; Cooke et al., 2009). 5.3. Job mobility Now we turn to the equation for job mobility (Table 3, first column of coefficients). The model shows that the association between migration and job mobility is mediated by the migration decision-making process. Planning migration is positively related with job mobility, while having migrated does not impact job change. This finding confirms that migration is rarely speculative. Individuals want to be certain about a job opportunity before deciding to migrate. Planning migration, the stage where the decision in favor of it has already been taken, has a positive effect on job mobility, because a job offer elsewhere has been accepted and preparations for the move are taking place (Huinink and Kley, 2008; Kley, 2011). After controlling for the planning stage and for actual migration, considering migration as an option for the future – without having concrete plans – still increases the probability of job mobility significantly. This is true regardless of whether the job opportunities are outside of or within the town of residence and this finding confirms our core hypothesis. Those who are open to migrate experience higher job mobility than those who are not. The effect of considering migration as an option for the future is independent from a possible indirect effect of job search since we control for the latter. Job search activity here refers to any spatial area and a time span up to three months before the new job started. As expected, a job search largely explains the variation in job mobility but not completely. Not all job changes are the result of previous job searches and opportunities for a job change may just come up. The positive effect of considering migration corroborates our hypothesis that those individuals who are open to migrate and have higher aspirations are more prone to accept such opportunities and exhibit job mobility. Allowing job mobility outcomes to be commonly affected by unobserved factors that impact migration and job search does not diminish the significance of considering migration on job mobility. Compared to the model without correlated errors (not shown here) it is slightly reduced. This indicates that some selection mechanisms are present in this association. To complete the picture, some comments on the other predictors are necessary. Respondents who are non-employed are more likely to start a new job compared to those who are employed or enrolled in education. Commuters with long journey times are significantly more likely to change jobs than all other status groups. This finding supports the idea that commuting motivates job mobility due to high monetary, social, and other costs. Among sociodemographic characteristics, only education and age have significant effects. Higher education is positively associated with job mobility. The value of the age coefficient is very small because much of its effect is already captured by the covariate of duration dependence (i.e. time spent at risk before starting a new job). Other characteristics, like the place of residence or perceived opportunities, do not increase the probability of job mobility, presumably because the main part of their influence is mediated by the job search. 5.4. Migration The migration equation (Table 3, second column of coefficients) sheds light on the direct effects of considering or planning migration as well as of job mobility and job search.16 As expected, planning migration is the most important predictor for actually realizing migration, since individuals at this stage have already made a firm decision. Many of the control variables that are known to be predictors of migration, therefore show no significant influence. Acceptance of a job offer elsewhere may impact a decision in favor of migration. In fact, previous job status, i.e. whether a person was non-employed, employed, or enrolled in education, and the situation after beginning a new job show no effect on actual migration after controlling for the status of planning migration. This finding corroborates previous findings that jobrelated migration normally takes place after the job has been found (Kley, 2009; Kley and Mulder, 2010); in other words, migration is an instrumental behavior to reduce the costs derived from finding a new job at a greater distance from the former residence. Searching for a job has an independent effect on migration. There might be a subgroup of individuals who need a job at the destination once the decision to migrate has been taken because the migration may have been initiated for other reasons. Other aspects, like job mobility, human capital, or perceived local opportunities, are likely to be predictors of considering and 16

Only minor changes in the coefficients are observed between the models that account for time-constant unobserved heterogeneity and those that do not.

11

J. Huinink et al. / Social Science Research 44 (2014) 1–14 Table 4 Sensitivity of the coefficient of considering migration, numerical integration vs. MCMC. In equation for

Migration Job mobility Job search (650 km) Job search (>50 km)

Coefficient of ‘considering migration’ estimated by Numerical integration

MCMC average

95% Confidence interval

1.62*** 0.24* 0.00 0.38***

0.75** 0.32** 0.31** 0.92***

0.02 0.05 0.11 0.72

1.71 0.59 0.51 1.21

Numerical integration results are obtained from the interdependent models shown above. MCMC average coefficient is averaged from the results obtained from 24,000 iterations. * Significance level: 0.1. ** Significance level: 0.05. *** Significance level: 0.01.

planning migration and are therefore only weakly associated with migration behavior once its pre-action stages have been accounted for. 5.5. Results of Markov chain Monte Carlo estimation In order to test the robustness of our results based on numerical integration, we performed an MCMC estimation of the interdependent model specified above, averaging the results of 24,000 iterations of the estimation. The level of correlation between iterations of the MCMC estimation was relatively low, ensuring a good selection of the prior distribution for the calculation of the posterior distribution. In Table 4 we show the coefficients of considering migration for the four equations estimated simultaneously, comparing the results of the numerical integration with those of the MCMC. The results of the MCMC indicate that there is a significant positive effect of considering migration on all outcomes. The results of the MCMC reproduce the direction of the sign of the coefficients estimated numerically. The exception is a positive effect of considering migration on job search within a short distance, which was found to be not significantly different from 0 by numerical integration in the interdependent model. This finding points to the fact that individuals who consider migration as an option may not necessarily be interested in improving their job situation by migrating. Moreover, the coefficient of job search over long distances by numerical integration falls below the 95% confidence interval of the MCMC coefficient estimate, but both coefficients are positive and significantly different from 0. A larger sample would have been helpful to obtain similar results across numerical integration and MCMC estimation. However, other coefficients not shown here are very similar when the numerical integration is compared with the MCMC estimates. The results in general confirm a positive impact of considering migration on job mobility. 6. Conclusion In our study, we examined the interdependence between migration and job mobility from a decision-making perspective and investigated the relevance of considering migration – our indicator of an individual’s openness to migrate – for job mobility. Our research contributes to the theoretical framework on the interdependence between migration and job mobility by combining a modified version of the subjective expected utility model with a three-stage model of migration that encompasses migration decision-making and actual migration, and by including the job search as a preceding step for job mobility. It was argued that being open to migrate supports aspirations in regard to employment and enhances perceived employment opportunities, particularly at larger geographical scales. For that reason, we expected that the openness to migrate (1) is positively associated with job search – particularly for jobs at a greater distance – and so indirectly enhances job mobility – and (2) has direct positive effects on job mobility in general. Testing our hypotheses with data from a tailor-made panel study from two German cities, we confirmed that individuals’ openness to migrate – measured by considering migration as an option for the future – is positively related to job search activities in a greater radius, which enhances the number of opportunities and chances of actually finding a new job. In addition, we find a positive effect of openness to migrate on job mobility – independently of the distance to the new workplace. This effect is smaller than the effect of openness to migrate on job search, which is reasonable because what is accounted for in the job mobility equation is partly ‘channeled’ by the wider job search. However, a direct effect of openness to migrate on job mobility reflects that job opportunities are generally identified and accepted more often when an individual is open to migrate. We introduced a measure of planning migration in the models that is positively associated with both migration and job mobility. Accounting for planning migration rules out that considering migration reflects a firm migration decision, as previous research has proved planning migration to be an indicator of the decision in favor of migration. Therefore, our measure of considering migration as an option for the future is a consistent indicator that adequately approximates the attitude openness to migrate.

12

J. Huinink et al. / Social Science Research 44 (2014) 1–14

The findings are robust against commonly-used individual predictors of migration and job mobility like age, sex, education, living arrangement and more. It was not within the scope of this article to study their effects in greater detail. In particular, gender issues and family migration and other constrains to job-related migration could not be focused on. This study contributes to the literature that addresses the self-selection of migrants by theorizing on a factor that potentially mediates the interdependence between migration and job mobility. It enhances our understanding of the mechanisms behind a common selectivity of migrants and job mobile individuals. Although individuals’ openness to migrate captures part of the effect of unmeasured personal traits, latent heterogeneity commonly affecting migration and job mobility still exists. Therefore personal traits, like work orientations, sociability and other individual dispositions are still important after including openness to migrate, but their potential bias on our estimates is controlled by the statistical model. We are also aware of the fact that even though we account for a number of relevant time-varying factors to minimize this potential biases we could not do so for all potential unmeasured sources of time-varying heterogeneity. For further studies on the association between job mobility and migration it might be fruitful to investigate the influence of occupational skills and the spatial distribution of occupational opportunities (cf. van Ham, 2002). Because it is very likely that some occupations enhance mobility more according to task-specific requirements or due to an occupation-related uneven distribution of job opportunities across space, we may want to analyze whether occupational characteristics not accounted for here are a key to distinguishing who ends up considering migration and who does not. Furthermore, we did not distinguish between upward and downward mobility because of data restrictions. To conclude, the interdependence between migration and job mobility is more complicated than assumed in the past. According to our findings, considering migration for non-economic reasons may lead to job changes that were not intended. Other studies show that job mobility may well be a consequence of migration and not the reason for it (cf. Morrison and Clark, 2011). Therefore, research on the interdependence between migration, job mobility, and other life domains, most prominently family, should be intensified in the future. This would also contribute to a better understanding of the dynamics of individual life courses and self-selection in general. Appendix A Descriptive statistics of the analysis variables (weighted). Proportion Considering migration Planning migration Second or higher residential episode

0.24 0.10 0.09

In employment Second or higher employment episode Commutes (over 1 h)

0.68 0.04 0.04

Job search (650 km radius) Job search (>50 km radius)

0.07 0.05

Higher education Enrolled in school, vocational training or studies

0.36 0.28

Partnership Children

0.71 0.41

Ever migrated (before observation) City: Freiburg (ref.: Magdeburg)

0.72 0.49

Age: 36–50 years (ref.: 18–35) Sex: female

0.50 0.57 Mean value (Std. dev.) *

*

Perceived opportunities in town

3.82 (7.28)

N N N N

149 401 34,148 1172

of of of of

migration events job mobility events spells persons

Index with minimum = 42, maximum = 16.

J. Huinink et al. / Social Science Research 44 (2014) 1–14

13

References Ajzen, I., 2005. Attitudes, Personality and Behavior. Open University Press McGraw-Hill, Berkshire and New York. Allison, P.D., 1982. Discrete-time methods for the analysis of event histories. Sociological Methodology 13, 61–98. Baizan, P., Aassve, A., Billari, F.C., 2003. Cohabitation, marriage and first birth: the interrelationship of family formation events in Spain. European Journal of Population 19, 147–169. Billari, F.C., Philipov, D., 2004. Women’s education and entry into a first union. A simultaneous hazard comparative analysis of Central and Eastern Europe. Vienna Yearbook of Population Research 2004. Vienna: Austrian Academy of Sciences. Blau, P.M., Duncan, O.D., 1967. The American Occupational Structure. Wiley, New York. Borjas, G.J., 1999. The economic analysis of immigration. Handbook of Labor Economics 3, 1697–1760. Boyle, P., Gayle, V., Feng, Z., 2009. A new look at family migration and women’s employment status. Journal of Marriage and Family 71, 417–431. Brown, L., Moore, G., 1970. The intra-urban migration process: a perspective. Geografiska Annaler 52B, 1–13. Browne, W., Steele, F., Golalizadeh, M., Green, M., 2009. The use of simple reparametrizations to improve the efficiency of Markov chain Monte Carlo estimation for multilevel models with applications to discrete time survival models. Journal of the Royal Statistical Society A 172, 579–598. Clark, W.A.V., 1986. Human Migration. Sage Publications, Beverly Hills, CA. Clark, W.A.V., Withers, S., 1999. Changing jobs and changing houses: mobility outcomes of employment transitions. Journal of Regional Science 39, 653– 673. Clark, W.A.V., Huang, Y., Withers, S., 2003. Does commuting distance matter? Commuting tolerance and residential change. Regional Science and Urban Economics 33, 199–221. Cooke, T.J., Boyle, P.J., Couch, K., Feijten, P., 2009. A longitudinal analysis of family migration and the gender gap in earnings in the United States and Great Britain. Demography 46, 147–167. Courgeau, D., 1985. Interaction between spatial mobility, family and career life cycle: a French survey. European Sociological Review 1, 139–162. Da Vanzo, J., 1981. Microeconomic approaches to studying migration decisions. In: De Jong, G.F., Gardner, R.W. (Eds.), Migration Decision Making. Multidisciplinary Approaches to Microlevel Studies in Developed and Developing Countries. Pergamon Press, New York, pp. 90–129. Da Jong, G., Fawcett, J., 1981. Motivations for migration: an assessment and a value-expectancy research model. In: De Jong, G.F., Gardner, R.W. (Eds.), Migration Decision Making. Multidisciplinary Approaches to Microlevel Studies in Developed and Developing Countries. Pergamon Press, New York, pp. 13–58. Detang-Dessendre, C., 1999. Reciprocal link between exit from unemployment and geographical mobility. Environment and Planning A 31, 1417–1431. Detang-Dessendre, C., Mohlo, I., 1999. Migration and changing employment status: a hazard function analysis. Journal of Regional Science 39, 103–123. Eliasson, K., Lindgren, U., Westerlund, O., 2003. Geographical labour mobility: migration or commuting? Regional Studies 37, 827–837. Evers, G.H.M., 1989. Simultaneous models for migration and commuting: macro and micro economic approaches. In: Van Dijk, J., Folmer, H., Herzog, H.W., Schlottman, A. (Eds.), Migration and Labour Market Adjustment. Kluwer, Dordrecht, pp. 177–197. Fawcett, J.T., 1985. Migration psychology. New behavioral models. Population and Environment 8, 5–14. Gabler, S., Häder, S., 2002. Idiosyncrasies in telephone sampling - the case of Germany. International Journal of Public Opinion Research 14, 339–345. Guo, G., 1993. Event-history analysis for left-truncated data. Sociological Methodology 23, 217–247. Hadfield, J., 2010. MCMC Methods for multi-response generalized linear mixed models: the MCMCglmm R Package. Journal of Statistical Software 33, 1–22. Hanson, S., Pratt, G., 1991. Job search and the occupational segregation of women. Annals of the Association of American Geographers 81, 229–253. Hanson, S., Pratt, G., 1995. Gender, Work, and Space. Routledge, London. Heckhausen, H., Gollwitzer, P.M., 1987. Thought contents and cognitive functioning in motivational versus volitional states of mind. Motivation and Emotion 11, 101–120. Huinink, J., Kley, S., 2008. Regionaler Kontext und Migrationsentscheidungen im Lebensverlauf. In: Kalter, Frank (Eds.), Migration und Integration. Kölner Zeitschrift für Soziologie und Sozialpsychologie. Sonderheft, pp. 162–184. Huinink, J., Kley, S., 2011. Migration Decisions in the Course of Life. GESIS Data Archive, Cologne. ZA5228 Data file Version 1.0.0, doi:10.4232/1.11063. Huinink, J., Vidal, S., Kley, S., 2011. Effects of Residential Mobility on Job Mobility over the Life Course. CIQLE WP 2011–2. Yale University, New Haven. Huinink, J., Schröder, T., 2008. Skizzen zu einer Theorie des Lebensverlaufs. In: Diekmann, A., Eichner, K., Schmid, P., Voss, T. (Eds.), Rational Choice: Theoretische Analysen und empirische Resultate. VS, Wiesbaden, pp. 291–308. Kalter, F., 1997. Wohnortwechsel in Deutschland. Ein Beitrag zur Migrationstheorie und zur empirischen Anwendung von Rational-Choice-Modellen. Leske & Budrich, Opladen. Kley, S., 2009. Migration im Lebensverlauf. Der Einfluss von Lebensbedingungen und Lebenslaufereignissen auf den Wohnortwechsel. VS, Wiesbaden. Kley, S., 2011. Explaining the stages of migration within a life-course framework. European Sociological Review 27, 469–486. Kley, S., Mulder, C.H., 2010. Considering, planning, and realizing migration in early adulthood. The influence of life-course events and perceived opportunities on leaving the city in Germany. Journal of Housing and the Built Environment 25, 73–94. Kulu, H., 2008. Fertility and spatial mobility in the life course: evidence from Austria. Environment and Planning A 40, 632–652. Lillard, L.A., 1993. Simultaneous equations for hazards: marriage duration and fertility timing. Journal of Econometrics 56, 189–217. Lillard, L.A., Panis, C.W.A., 2002. aML Multilevel Multiprocess Statistical Software: Release 1.0. Los Angeles, California: EconWare. Lillard, L.A., Panis, C.W.A., 1996. Marital status and mortality: the role of health. Demography 33, 313–327. Lindenberg, S., 2001. Intrinsic motivation in a new light. Kyklos 54, 317–342. Linneman, P., Graves, P.E., 1983. Migration and job change. A multinomial logit approach. Journal of Urban Economics 14, 263–279. Lippman, S.A., McCall, J.J., 1976. The economics of job search: a survey. Economic Inquiry 14, 155–189. Long, L., Tucker, J., Urton, W.L., 1988. Migration distances: an international comparison. Demography 25, 633–640. Morrison, P.S., Clark, W.A.V., 2011. Internal migration and employment: macro flows and micro motives. Environment and Planning A 43, 1948–1964. Mulder, C.H., 1993. Migration Dynamics: A Life Course Approach. Thesis Publishers, Amsterdam. Mulder, C.H., Hooimeijer, P., 1999. Residential relocation and the life course. In: Van Wissen, L.J., Dykstra, P.A. (Eds.), Population Issues. An Interdisciplinary Focus. MacMillan, Kluwer Academic/Plenum Publishers, pp. 159–186. Mulder, C.H., van Ham, M., 2005. Migration histories and occupational achievement. Population Space and Place 11, 173–186. Nakosteen, R.A., Westerlund, O., Zimmer, M., 2008. Migration and self-selection: measured earnings and latent characteristics. Journal of Regional Science 48, 769–788. Oddland, J., Shumway, J.M., 1993. Interdependence in the timing of migration and mobility events. Papers in Regional Sciences 72, 221–237. Ormel, J., Lindenberg, S., Steverink, N., Verbrugge, L.M., 1999. Subjective well being and social production functions. Social Indicators Research 46, 61–90. Rozenblat, C., Cicille, P., 2003. Les villes européennes. Analyse comparative. Montpellier, DATAR. Schneider, N.F., Meil, G. (Eds.), 2008. Mobile Living Across Europe I. Relevance and Diversity of Job-Related Spatial Mobility in Six European Countries. Barbara Budrich Publishers, Opladen. Schoemaker, P.J.H., 1982. The expected utility model: its variants, purposes, evidence and limitations. Journal of Economic Literature 20, 529–563. Schultz, T., 1964. Transforming Traditional Agriculture. Yale University Press, New Haven, CT. Shauman, K., Noonan, M., 2007. Family migration and labor force outcomes: sex differences in occupational context. Social Forces 85, 1735–1764. Simpson, W., 1980. A simultaneous model of workplace and residential location incorporating job search. Journal of Urban Economics 8, 330–349. Simpson, W., 1992. Urban Structure and the Labour Market: Worker Mobility, Commuting and Underemployment in Cities. Clarendon Press, Oxford. Sjaastad, L.A., 1962. The costs and returns of human migration. The Journal of Political Economy 70, 80–93. Speare, A., 1971. A cost-benefit model of rural to urban migration in Taiwan. Population Studies 25, 117–130.

14

J. Huinink et al. / Social Science Research 44 (2014) 1–14

Speare, A., Goldstein, S., Frey, W.H., 1975. Residential Mobility, Migration, and Metropolitan Change. Ballinger, Cambridge. Steele, F., Diamond, I., Wang, D., 1996. The determinants of the duration of contraceptive use in China: a multilevel multinomial discrete hazards modelling approach. Demography 33, 12–33. Todaro, M.P., 1969. A model of migration and urban unemployment in less-developed countries. American Economic Review 59, 138–148. Van Ham, M., 2002. Job Access, Workplace Mobility, and Occupational Achievement. Eburon, Delft. Van Ham, M., Hooimeijer, P., 2009. Regional differences in spatial flexibility: long commutes and job related migration intentions in the Netherlands. Applied Spatial Analysis 2, 129–146. Van Ham, M., Mulder, C.H., Hooimeijer, P., 2001. Spatial flexibility in job mobility: macrolevel opportunities and microlevel restrictions. Environment and Planning A 33, 921–940. Van Ommeren, J., Rietveld, P., Nijkamp, P., 1999. Impacts of employed spouses on job-moving behaviour. International Regional Science Review 22, 54–68. Vroom, V.H., 1964. Work and Motivation. Wiley, New York. Zax, J., Kain, J., 1991. Commutes, quits and moves. Journal of Urban Economics 29, 153–165.