Skip to navigation

Skip to main content


Why do People Stay? The Insider Advantages Approach -- Thomas Straubhaar and Peter Fischer

Why do People Stay?
The Insider Advantages Approach
Empirical Evidence from Swedish Labour Markets

by Peter A. Fischer*, Einar Holm+,
Gunnar Malmberg+ and Thomas Straubhaar**_

Summary 1

Abstract 4

Introduction 5

Explaining (im)mobility 6

Migration and Immobility in Swedish Labour Markets 9

3.1 The Data 9

3.2 Patterns of (Im)Mobility 10

3.3 Stayers and Movers in Sweden 13

Immobility, Insider Advantages and Duration Dependence: A Bivariate Probit Model of the Determinants of Immobility 20

4.1 The model 20

4.2 Econometric issues 21

4.3 Estimation Results 22

4.3.1 Duration of stay and insider advantages 22

4.3.2 Traditional determinants of migration 29

4.3.3 Life-course events 31

4.3.4 Place of birth and occupation 32

4.3.5 Group-specific heterogeneity 33

4.3.6 Business cycle effects on (im)mobility 36

Summary and conclusions 38

References 40

This paper is produced as part of a CEPR research programme on European Migration from Economic Analysis to Policy Response, supported by a grant from the Commission of the European Communities under its Human Capital and Mobility Programme (no. ERBCHRXCT940515).

The question ìwhy do people stay?î stands for more than just the opposite of asking why people move. Classical migration theories have proved quite successful in identifying reasons for migration. They explain spatial mobility in terms of differences between macro-factors in regions of origin and potential destination, group dynamics and networking on the meso level and socio-economic characteristics and behavioural strategies on the micro level. But traditional migration theories offer less satisfactory answers as to why the large majority of people almost always stays geographically immobile; at least in the classical sense of migration.

World-wide, only about 2 percent of the total population live outside their home countries. In many European countries, internal mobility has had a tendency to decrease substantially since the late Sixties and early Seventies. As a crude rule of thumb, international emigration propensities have in recent years in Europe been around or below 0.5 per cent of average population, average internal emigration rates around or below 2 per cent. In Sweden, half of all internal migration was within a distance of five kilometres. Immobility, not labour market migration is in fact the dominant mass phenomenon of European populations at the end of this century.

Economics usually treats mobility of people in a way similar to the movement of capital or goods. It sees migration as an essential tool to guarantee the efficient allocation of production factors. Immobility is generally evaluated negatively, as a rigidity that causes welfare costs. Immobility of people is often cited as one of the fundamental reasons for the high and stable structural unemployment rates in Europe. But is it really inefficient for people to be immobile? In this paper we extend standard explanations of mobility by putting forward what we call the insider advantage theory of immobility. We argue that it usually makes sense for most people not to consider moving because much of their knowledge, information and abilities that grant them high productivity in work and an optimal use of leisure time are location-specific. Staying immobile allows them to accumulate such insider advantages. Even if substantial macro-economic differences persist on the aggregate level, moving away will generally not be worth it from an individual point of view, due to the costs of accumulated location-specific insider advantages being lost in case of emigration.

With respect to mobility, nobel laureate Gary S. Becker was probably the first to emphasise that part of the knowledge an individual acquires is often firm-specific and cannot be transferred to an other employment. Migration may therefore result in a ñat least temporary- decrease of potentially achievable relative wages, because firm-specific abilities are ësunkí in case of a change of workplace. Although no doubt important and illustrative, labour economistís exclusive concentration on firm-specific insider advantages is in our opinion far too rigid to satisfactorily explain immobility. There are also place- and society-specific insider advantages that increase oneís location-specific productivity. Furthermore much points towards leisure-oriented location-specific advantages being more important in explaining immobility than work-oriented insider advantages; especially in highly developed societies. Place-specific insider-advantages enable the individual to make best use of his limited financial resources. Knowing about local opportunities and those offers that best suit individual preferences can save a lot of money and help maximise the utility achievable during leisure time. Society-specific insider advantages broadly emanate from the social relations and political activities an immobile individual and his family build up within the society in which one is residing. Having friends who can look after or play with the children or having neighbours who give their helping hand when something has to be repaired can make everyday-life a lot easier and pleasant.

Empirically, the insider-advantage theory implies that the longer people stay at a certain location, the less likely they will be to move further on. Immobility should be highly duration dependent. Individuals whose skill are less location-specific and who have less place-specific ties are more likely to move.

In this paper, we use a new unique micro data set on mobility between Swedish labour markets to test empirically for the determinants of (im)mobility. Having information on socio-economic characteristics, family ties and the migratory history of all people resident in Sweden during 1985-95 we draw a sample that contains all movers between labour markets, 20 per cent of foreign-born and 2 per cent of native-born stayers. Analysing it statistically provides significant empirical support for our insider-advantage theory. Simply looking at the relation between probability and duration of stay shows that 20 per cent of those who moved 12 months ago moved again in 1994, while out of those who stayed in the same labour market for 119 months the corresponding share was only 0.3 per cent. Simultaneously accounting for socio-economic micro variables, macro-attributes, family ties and life-course events demonstrates how the partial effect of duration on mobility depends on the value of all the other explanatory variables. At mean characteristics the increase in probability of staying is ñalike all other effects- rather small (less than half a per cent for one additional year of staying), because average people are very, very unlikely to move anyway. But for more migration prone, young single individuals the partial effects of having stayed ten years instead of one year may increase the estimated probability of staying by more than 15 per cent.

Our results confirm the importance of individual micro-characteristics, above all age and education, in determining why people move. Somewhat surprisingly, traditional macroeconomic regional attributes like differences in average wages, employment or vacancy rates perform very poorly in explaining immobility. There is one notable exception, however. People living in the three metropolitan areas of Stockholm, Gothenburg and Malmà are significantly less mobile than similar individuals resident in other regions of the country. Metropolitan regions seem to offer particular scope for the accumulation of location-specific insider advantages, thereby keeping people immobile.

Ties to other people and places prove more important than macro variables in determining the probability to stay. Being married or having children as well as owning an own house boost the probability to stay. Apart from the importance of family variables our results underline the life-course event-specific character of much of actual mobility. More than two thirds of all moves in our sample are undertaken by people who are between 20 and 35 years old. 24 per cent of all moves are related to engaging in education. The formation or dissolution of households constitute other events that increase the likelihood of moving over labour market borders.While (being) unemployed turns out to be a mobility encouraging factor in Sweden, its magnitude is strongly dependent on age and family variables. We estimate that a 24 years old individual who becomes unemployed but is married to a working wife is not more likely to move than an employed single of the same age.

In brief, our empirical investigation of (im)mobility in Swedish regional labour markets identifies immobility as a strong and persistent behavioural strategy for the large majority of people. The estimation results support our argument that insider advantages and duration effects of staying are crucial in gaining a better understanding of the immobility phenomenon. Once one has accumulated sufficient insider-advantages, staying immobile is efficient. Most people will move only if a bad economic policy forces them to do so. Seen from this point of view, a good economic policy should rather encourage local opportunities for people to make efficient use of their labour than to try to encourage mobility beyond removing market imperfections.

Migration research has been quite successful in explaining changes in migration flows. Less satisfactory are its answers as to why the overwhelming majority of people remains immobile, despite persistent regional wealth differences and economic integration proceeding. We suggest to complement traditional theories with an insider advantages approach towards immobility. Most people do not move because staying immobile they have accumulated work- and leisure-oriented insider advantages that are location-specific and would be lost in case of emigration. Therefore, the longer people have stayed -and the more insider advantages they have accumulated-, the less likely they are to move. Using a new micro dataset covering all people resident in Sweden in 1994 and their mobility experience since 1985, we find a strong positive duration dependence of the probability to stay even after controlling for a large set of alternative factors. Traditional micro-economic characteristics prove helpful in explaining immobility, while regional macro-economic differences have very little impact on individual mobility decisions. A large part of moves between Swedish labour markets seems related to specific life-course events, of which getting unemployed is only one. Factors that are not dependent on ones own work but ought to increase location-specific insider-advantages, like having a working partner, children or owning a house, increase the probability to stay further. JEL-codes: F22, J60, R23Keywords: Immobility of people, insider advantages , migration, Swedish labour markets. Fischer, Holm, Malmberg and Straubhaar Why do People Stay? The Insider Advantages Approach and Empirical Evidence

Migration and mobility have for long caught a lot of attention in theory as well as in public debate. But in actual fact the large majority of people does not move. Worldwide, only about 2 percent of the total population live outside their home countries (ILO/IOM/UNHCR 1994). For Europe, records on internal mobility are not substantially different (Salt, Singleton and Hogarth 1994). In many European countries internal mobility has had a tendency to decrease substantially since the late Sixties and early Seventies. Not migration, immobility is in fact the dominant mass phenomenon of European populations at the end of this century, and it is one of the fundamental reasons for the high and stable structural unemployment rate.

Traditional migration theory has been quite successful in explaining causes and consequences of international and internal spatial mobility. At present we have several satisfactory answers as to why migration streams may fluctuate over time and place. Less clear it is, however, why migrants remain a tiny minority all over the world. In this paper we extend standard explanations of mobility by adopting what we call the insider advantage approach towards immobility. We argue that usually most people do not even consider moving because they have accumulated too many location-specific insider advantages over time. Even if substantial macro-economic differences persist on the aggregate level, moving away will generally not be worth it from an individual point of view, due to the costs of accumulated location-specific insider advantages that would be sunk in case of emigration. The longer people stay, the less likely they will be to move. Therefore (im)mobility should be highly duration dependent.

In this paper we use a newly elaborated very comprehensive micro dataset on people resident in Swedish labour markets in 1994 to test our insider advantages approach. Due to the availability of extensive information about each individual it has been possible to test the partial effect of various factors on the probabilities to stay within the same regional labour market. Results will be provided of estimated bivariate probit models identifying the impact on staying propensities of (a) duration variables, (b) classical determinants of mobility, especially individual socio-economic characteristics and region-specific macroeconomic conditions, (c) family variables proxying ties to people and activities, (d) life course events (e) occupation and place of birth and (f) business cycle effects.

In what follows, we first introduce the insider-advantage approach towards explaining immobility. Section three introduces the data and provides some statistical information on immobility in Swedish labour markets. Section four introduces a bivariate probability model of the decision to stay and discusses the results of its estimation on Swedish data. Part five concludes.

Explaining (im)mobility
Classical migration theories explain mobility in terms of differences between attributes of places on the macro level, group dynamics and networking on the meso level and socio-economic characteristics and behavioural strategies on the micro-level (for a new and comprehensive comparative multidisciplinary survey on theories of migration see Hammar et al. 1997, and for a shorter review that more explicitly treats issues of economic methodology Bauer and Zimmermann 1997). Studies using aggregate data have shown that changes in place-specific living standards and labour market conditions may explain fluctuations in migration flows (Ghatak, Levine and Wheatley Price 1996). This success, however, is less pronounced as far as absolute levels are concerned. Despite considerable macro-economic differences persisting both between many countries as well as within,_ and despite technological and political decreases in obstacles to migration, the vast majority of people has not and does not consider moving._ To complicate things even further, migration does not necessarily diminish between places that seem very similar on an aggregate level.

Although contributing insights as to why some people move, classical explanations are somewhat unsatisfactory in explaining why most people never consider moving. While Stark (1991) and Stark, Helmenstein and Yegorov (1997) concentrate on optimal duration of migration and on return we suggest a new approach to explain immobility which we call the insider-advantage approach. The insider advantage theory in itself accommodates several elements of traditional explanations but derives some new conclusions with respect to the underlying dynamics of mover-stayer decisions.

Under the conventional static view a micro-level decision maker compares her or his present and future level of utility in different macro-level units on the basis of her or his present stock of assets and abilities. In most cases this is not a realistic judgement because a certain part of the abilities and assets of every human being are location-specific, in other words they can only be used (or are only existent) in a specific place. These are what we call insider-advantages. They are not transferable to other places of work and residence. An important part of these abilities has to be obtained within a location specific learning process which requires time, information and temporary immobility. Mobility turns such investments into lost sunk costs, i.e. costs which are tied to a specific project or - in this case - a specific location and which are lost in the case of emigration. Henceforth, immobility makes sense to a majority of people because the loss of location specific assets and abilities induced by migration would be too severe and because it is immobility which allows to accumulate insider-advantages.

With respect to mobility, Becker (1962) and his scholars emphasised that part of the knowledge an individual acquires is often firm-specific and can not be transferred to another employment. Migration may therefore result in a decrease of potentially achievable relative wages because firm-specific abilities are 'sunk' in case of a change of workplace. It seems to us, however, that the exclusive concentration on firm-specific insider advantages is far too rigid to satisfactorily explain immobility._

Figure 1 gives a graphical representation of the structure of our insider-advantage idea. It differentiates insider-advantages further according to their origin (work- or leisure-related) and specificity (firm-, place- or society-specific).

Place-specific advantages make the individual particularly attractive for all or at least some firms in his region of work. Examples of such insider advantages are expertise in the location-specific preferences, desires and habits of clients or insider knowledge of the peculiarities of the political situation in a region. Society-specific advantages broadly emanate from the social relations and political activities an immobile individual builds up within the society in which he/she is residing. Firm, place- and society-specific work-oriented insider advantages lead to higher revenues for the individual, in the form of wages or other income.

Leisure-oriented location-specific insider advantages allow a resident to reach a higher utility level with a given set of monetary or other resources and time. Examples of place-specific leisure-oriented insider advantages can range from information about the ëgood-value-for-moneyí Italian restaurant round the corner to knowledge about cultural events and the local housing

Figure 1: The Insider-Advantage Approach Towards Immobility

_source: own illustration.

market._ Society-specific leisure-oriented insider advantages capture the utility increase a decision maker and his family get from having friends, being socially integrated, accepted and active at a certain place of residence. These insider advantages result from a locational investment in ësocial capital' which encompasses a wide range of human contacts, from family relations and friendships to membership of clubs and political parties. Mobility generally induces the loss of most of these abilities and assets and requires new investments in obtaining a ëticket to entryí at a new place of residence. Place of origin related networks of ëcompatriotsí at the place of destination may lower this loss.

The more developed economies are and the relatively more important leisure time becomes in comparison to gaining income, the more leisure-oriented insider advantages ought to matter for the decision to stay or to go. Admittedly leisure-oriented insider advantages are for the most part difficult to quantify in monetary terms. Nevertheless is should be possible to control their importance indirectly. If duration of stay increases the probability of staying irrespective of employment or if e.g. having children decreases individual propensities to move, then this is likely to be due to leisure-oriented insider advantages rather than work-oriented ones.

Intertemporally, insider-advantages may be partly recovered and ëupdatedí if one returns, but they nevertheless strongly increase the (opportunity) costs of staying away. There may be some cases where the benefits of being an outsider create the very incentive to move. A researcher specialised in the political economy of policy making in African states may be more scarce and thus in higher demand outside the African continent than in the places his knowledge is about. But such situations tend to be exceptional. Generally it is e.g. highly unlikely that the specialist on farming in Scandinavia where summers are short will be in great demand in sun abundant Southern Italy. While outsider advantages may be an incentive to move for a few, our argument is that insider advantages are important to explain the immobility because they are reason enough to stay for most.

There is some similarity between our insider-advantage approach and the human capital approach. The human capital approach emphasises the point that people are very different in their characteristics and their abilities and that migration may be a form of investment on which the return will occur within a given future time span. The insider-advantage approach stresses that during periods of immobility at a particular location individuals invest in the accumulation of location specific skills, abilities and assets. By this they can increase the realisable individual utility at this location significantly. Consequently, even if on an aggregate level considerable locational differences in average incomes, unemployment risks or endowment with natural amenities may be observed, an individual may actually rightly expect that a move might decrease her personal utility due to the incurred loss on non-transferable knowledge and the costly need to acquire new insider-advantages in order to get into a similar relative position at the new location._

Important empirical implications of the insider-advantage approach to immobility are that mobility patterns should be duration dependent and that the degree of transferability of skills, abilities and personal relations ought to be of central importance for observed (im-)mobility behaviour. The more people have accumulated location-specific insider-advantages and the less transferable their abilities and current ëlife-projectsí are, the more likely it is that they stay immobile. Somebody who has moved recently and thus has lost his accumulated insider-advantages should be more likely to move again. The longer she or he keeps staying at the new place of residence, the further the probability of an additional move will decrease again.

Migration and Immobility in Swedish Labour Markets
3.1 The Data
The empirical investigation in this paper relies on data from the TOPSWING database. This new database constructed at the Institute of Economic and Social Geography of the University of Ume and the Spatial Modelling Centre in Kiruna links information from various official statistical registers and censuses as provided by Statistics Sweden. It covers anonymised micro data for the total population resident in Sweden between 1985 and 1995, i.e. more than 9 million people. It includes information about individualís place of residence and work, age, gender, education, income, employment, profession as well as household conditions and links to family members (see the appendix at the end of this paper for further information on the variables used)._

In this paper we are interested in inter-regional migration that is not purely residential but involves both, a change of residence and a change of workplace. Our data does not constrain us to use politically defined county borders. Instead, we will use economically defined labour market regions as spatial resolution of our analyses in what follows. Labour market regions are defined by Statistics Sweden and the Swedish Department of Finance as to minimise inter-regional commuting (Finansdepartementet 1994). Using these labour market regions we separate the territory of Sweden into 108 different regional labour markets. For this study we have drawn a sample of all movers over labour market regions during the years of the study (1994 and 1989) and a control group of 2% of Swedish-born stayers and 20% of non-moving immigrants. For calculations individual observations have been weighted as to represent the true proportion in total population.

3.2 Patterns of (Im)Mobility
As in many other European countries, mobility in Sweden has not only been low but has also exhibited a tendency to decrease. Figure 2 shows how emigration propensities decreased relatively steadily in the 1970s and 1980s and have remained rather stable since._ Only for the last three years of the period of observation does the graph show a slow upswing in mobility patterns.

Figure 2: Internal Migration in Sweden 1967-95

_EINBETTEN Excel.Chart.8 \s___

source: Statistics Sweden.

Figure 3: Mobility by Distance of Moves in Sweden 1994

_EINBETTEN Excel.Chart.8 \s___

Source: Calculations based on data from Statistics Sweden as provided in the TOPSWING database, Department of Social and Economic Geography, Ume University.

In the 1960s the urbanisation process was still going strong and Sweden experienced an intra-regional redistribution of population from the countryside to the cities and an interregional redistribution from the northern parts of the country to the major industrial and metropolitan areas in the South (BorgegÂrd et al 1995). As in many other European countries this trend was followed in the 1970s by a suburbanisation tendency and a weak urbanisation trend in the 1980s with small interregional net-redistribution. While industrial areas have had a population decrease in the last decades the winners have been university regions, the main metropolitan areas and other regions with important public sectors.

The recent moderate upswing of mobility rates in the mid-1990s does not represent a resurgence of individual mobility. As Holm et al (1995) have demonstrated, it is due to an increase in the number of foreign born immigrants_ who were more mobile than natives in the first years after immigration and who had a stronger effect on population redistribution than changes in the rate of internal migration or (the small) regional variations in fertility and mortality. The most recent upswing in internal migration is also due to a simple demographic effect: the baby-boom generation is at present in their most mobile age.

Bengtson and Johanson (1993) point out some possible explanations to the trend of decreasing mobility in Sweden. They emphasise the low regional income variations, the increase in public job opportunities and the growing number of two income families. Trends that are also found in other developed economies. Indeed, (im)mobility patterns have been similar to those observed in other European countries (Fischer 1998). Studies of immobility in the Swedish context thus might reveal some Sweden-specific characteristics, but generally speaking the determinants affecting the patterns of immobility should be very similar to the situation in other European countries.

Figure 2 clearly illustrates the importance of distance in determining migration. Mobility between municipalities is distinguished from mobility between counties only by a (sizeable) scale factor. The low spatial resolution of the TOPSWING database allows us to show how the number of moves occurring in Sweden has been a strongly decreasing function of the distance involved (figure 3). In 1994, roughly one third of all recorded moves covered less than 1.5 kilometres of distance and around half of all migrations remained within a distance of 5 kilometres, i.e. constituted mainly residential mobility. The mobility rates of around 2 per cent of population that we find in our analysis of mobility between Swedish labour markets corresponds to the number of moves over a distance larger than 20 kilometres in figure 3. As can be seen, such ëlabour market movesí represent just one fourth of total mobility in Sweden. In brief, figure 3 illustrates that: the shorter the distance involved, the more numerous are the migration events. Therefore, the smaller the geographical entities, the more short moves will cross their border and the higher will therefore be the recorded mobility rates.

During 1994 about 15 per cent of the population in Sweden changed address, but only 2,3 per cent of the population had moved over the border of a labour market region. If the data period is extended to ten years (1985-1995) still 87% of the population lived in the same labour market region.

Apart from showing the demarcation of the labour market regions, map 1 gives net migration rates for 1985-95 in per cent of average population by regional labour market. It illustrates the regional centralisation patterns that have been common in Sweden. People have mainly moved from the inner parts of the country to the South and to the coast. In Northern Sweden only the labour market regions on the coast around the local centres of Ume and Lule report positive net migration rates while all the regions in the very Northwest recorded negative net migration rates. In the middle and South of the country internal migration was mainly directed towards the metropolitan areas of Stockholm, Gothenburg and Malmà and to local centres that were generally located along the coast. In one respect, map 1 is deceiving. Despite the very Northern regions showing the most negative net migration
Map 1: Net migration in Swedish Regional Labour Markets 1985-95

_ EINBETTEN Word.Picture.8 _____Source: own calculations based on data from Statistics Sweden as provided in the TOPSWING database, Department of Social and Economic Geography, Ume University.

rates in proportion to their population, the majority of internal moves has not been from inner Lapland to the South but within the middle and South of Sweden. This is due to the fact that Northern Sweden is so much more sparely populated than the South of the country, that the Northern migrants which are few in absolute numbers have a relatively large impact on relative net migration figures.

Disparities in average labour income per worker between the different labour market regions of Sweden have been limited in size, but not really trivial. In our data, the lowest regional average income from work amounted to 60 per cent of the highest one. But net migration rates and average income level per worker lack any linear correlation in our data (the correlation coefficient amounts to just 0.004)._

3.3 Stayers and Movers in Sweden
Figures 4 to 12 provide some of the most illustrative descriptive analyses of immobility in the Swedish Labour Markets in 1994 as derived from our dataset._ Note the importance of distinguishing between these overall relations and causal partial effects. For example, figure 4 showing the distinct age-specificity of staying and figure 5 illustrating the strong total relation between duration of stay and propensity to move both provide results that concern overall relations. But ageing and staying obviously go along with each other. Only our statistical analysis below will be able to provide causal partial effects and prove that duration of stay is important as such, not just in the sense of a mere effect of ageing.

First of all, figure 4 shows how small a minority movers are in Sweden. Second, it reveals the strong age-specificity of propensities to stay. More than two thirds of all movers have been between 20 and 35 years old. The figure also demonstrates the strong cohort-effect on the general level of interregional mobility.

Figure 4: Stayers and Movers between Labour Markets by Age, 1994.


Figure 5: Propensity to Stay in 1994 by Duration of Stay 1985-93


Figure 6: Propensity to Stay in 1994 by Previous Moves 1985-93


Figure 7: Propensity to Stay in 1994 by Years since Immigration


Source: Own calculations based on data from Statistics Sweden as provided in the TOPSWING database, Department of Social and Economic Geography, Ume University.

With respect to our insider advantages hypothesis, figure 5 lends considerable support to the implication that propensities to move are dependent on the duration of stay. While about every fifth of those individuals who have migrated 12 months ago moved again in 1994, less than half a per cent of those who have stayed for 119 months moved again.

Figure 6 documents that the propensity to move in 1994 was higher among those who had already moved previously, one or several times during 1985-1993, than among those who had stayed immobile during these ten years. A (very) small fraction of the population are obviously frequent movers. Probably the few notorious movers are people who never accumulate place-specific insider advantages._

The insider-advantage hypothesis also implies that individuals who have immigrated from abroad should at the outset be more mobile than natives, as they have already lost their location-specific insider advantages and have not yet accumulated many new ones. In a large number of studies it has been demonstrated that immigrants in Sweden show on average a much higher propensity to move than the population (c.f. BorgegÂrd et al 1995). Figure 7 demonstrates that in our data differences are also considerable with respect to those who have recently immigrated to Sweden. While propensities to move internally are very much higher for those who have immigrated during the last two years than for natives, the propensity to stay increases rapidly afterwards and becomes similar to Swedish-born._

Figure 8 provides movers and stayers by their highest level of education. The Swedish classification distinguishes between seven different levels which are firstly classified according to the length of the educational

Figure 8: Stayers and Movers in 1994 by Highest Level of Education Completed

a) absolute numbers


b) group-specific propensities to stay


Figure 9: Stayers and Movers by Marital Status


Figure 10: Propensity to Stay in 1994 by Number of Children


Figure 11: Propensity to Stay by Living in an Own House


Figure 12: Propensity to Stay by Labour Market Status


Source: Own calculations based on data from Statistics Sweden as provided in the TOPSWING database, Department of Social and Economic Geography, Ume University.

period usually needed to achieve it and secondly according to academic demands._ Graph a) shows absolute numbers, b) group-specific propensities to stay for the total population. If we define as medium skilled all those individuals belonging to classes 3-5, then they made up for 70.1 per cent of all movers. In the total population, just 57 per cent were medium skilled._ Only 12.4 per cent of all movers and 9.5 per cent of the total population were highly skilled (groups 6 and 7). Overall, holders of a PhD degree were less mobile than the medium skilled. This seems to contradict the widespread believe that the higher the education, the more likely one is to move. But as we will show below, despite the majority of movers being medium skilled, education increases the probability to move significantly. The patterns that show up in figure 8 are a result of age- and family- effects. The higher the level of a degree, the older one gets until one reaches it. People who enter an academic career tend to be more mobile than low skilled, but they also move (e.g. for educational purposes) more in their younger ages, before they obtain their PhD degree. Becoming older and especially having a family increases immobility considerably. In view of these factors, education is an important determinant of immobility. The higher the educational level of a person, the higher is her propensity to move, ceteris paribus. These relative differences, however, do not change the fact that the overwhelming majority of people is immobile once they are older than 35.

The importance of family variables shows up in figure 9. Married people are clearly more immobile than single ones. 81 per cent of all movers in 1994 were not married. In total population, only 55 per cent were single. While 98.8 per cent of all married individuals were immobile, this percentage was 96.3 for the single ones. Somewhat surprisingly, gender does not matter for propensities to stay in Swedish regional labour markets. Both, in this descriptive overall analysis as well as in the partial statistical analysis later on, no significant differences in mobility patterns between females and males could be detected.

Not only being married, but also having children makes people immobile. However, figure 10 illustrates a somewhat surprising effect that our analytical models cannot reveal because the individuals concerned are too tiny a minority. While the propensity to stay increases considerably when people have one to three children instead of none, individuals who have four or more children obviously are the more mobile the more children they have. It is unclear where this arises from (may be having so many children means that most insider advantages are family-specific or may be having many children captures a specific selection of people that is mobility prone on other grounds). There are several family variables that are related to immobility and which partly related to each others. Another important one is houseownership, as figure 11 illustrates. House owners are less mobile than people who life in rented property.

Finally, figure 12 provides support for the main premise of the labour market theory. Labour market status influences propensities to move. We define all people aged 19-64 who record an annual income from work larger than SKr 50,000 as employed, all those who receive unemployment benefits larger than SKr 20,000 as unemployed and the rest as being out of the labour force. With 4.1 per cent of the unemployed moving as compared to 1.7 per cent of the employed, unemployed are more mobile than employed. But with 6.1 per cent the mobility rate was actually highest for those who were out of labour force (and aged 19-64). Obviously, mobility has not just to do with direct labour market reasons. Even with the unemployed, 95.9 per cent remained immobile during 1994. 39 per cent of all movers were employed. However, all these effects are better analysed within a model that allows us to isolate partial effects. This will be provided in section four below.

Immobility, Insider Advantages and Duration Dependence: A Bivariate Probit Model of the Determinants of Immobility
4.1 The model
We start with a traditional neo-classical utility maximising migration function and assume that individuals decide whether to move or not by comparing expected utilities at alternative destinations with their present situation:

_ (1)

Individuals are likely to move if the net present value NPV(mij)to at time t=0 of migrating from i to an alternative destination j is larger than zero. The NPV is made up by the difference over time t=({0...x} between expected utilities at an alternative destination (uj) and at the

º_ respect to the determination of UWO and ULO individual time horizons and discount factors are obviously dependent on age. The standard migration theory implies that UWO and ULO are furthermore conditional on individual micro-economic characteristics XMIC like education and experience as well as on an individualís family situation (marital status, number of children etc.). Also we expect region-specific macro factors XMAC like labour market conditions (unemployment and vacancies rate) or general regional attributes (wage level, local opportunities) to be important for the determination of place-specific utilities. Provided that the life-event theory of mobility proves relevant, we should find utilities to be further dependent on life course events LCE like the formation or disintegration of households, engaging in educational activities, getting unemployed or having a baby. Last but not least, the insider-advantage hypothesis suggests that the amount of location-specific insider advantages LA an individual has accumulated will be important in defining the work and leisure oriented utilities a person can expect to achieve at a certain place. Given our insider advantage hypothesis, we expect LA to be a function of duration of stay and transferability of skills and abilities. Thus,

_ (3).

It is obviously impossible to observe any continuous proxy for actual utilities. What we can observe is the result of the decision making process only; people either stay or go. However, if individuals decide to move when the expected net present utility of staying at another location exceeds the one at the present place of residence, we have

_ (4).

What kind of quantitative model one gets is thus entirely dependent on the assumption one makes about the distribution of _. For convenience in inference we will work with the probit specification:_

_ (5)

Recalling our previous discussion, we have to estimate a relation of the form

_ (6).

As it obviously often is difficult to know whether regressors proxy work or leisure oriented insider effects, distinguishing between the two will remain to some extent a question of interpretation.

4.2 Econometric issues
As usual when estimating a non-linear model, the estimated coefficients _in the probit model do not equal the marginal effect bí of a change in an explanatory variable x on the probability of staying y. Indeed,

_ (7)

The partial impact of a marginal change in an explanatory variable on the probability of staying thus depends on the actual value of the other explanatory variables. If nothing else is mentioned, we are going to calculate marginal effects bí for all other characteristics being at the mean value of all observations.

Unlike in OLS-regressions it is impossible in binary choice models to calculate an R2 that sets true and estimated y into proportion and explains the part of the total variance an estimated model explains. For discrete choice estimation, several goodness of fit indicators have been proposed (for a discussion of the most prominent alternatives see Amemiya 1981 and Veall and Zimmermann 1992). Most of them aim to approximately mimic the properties of the OLS-R2 and are therefore also called pseudo-R2. In a comparative simulation study, Windmeijer (1995) compared the performance of different goodness of fit indicators. He suggests to use the pseudo-R2 that best mimics the correlation coefficient of the true conditional mean with the predicted conditional mean but concludes that ëit would not make much difference which measure is used in evaluating modelsí (Windmeijer 1995; 112).

Many goodness of fit indicators in binary choice models rely more or less directly on log-likelihoods. In what follows we will calculate the pseudo-R2 proposed originally by McKelvey and Zavoina (1975) and developed further by Laitila (1993). It did not only perform best in the Windmeijer study but also seems least sensitive to the proportion of 1 in the sample (a problem particularly evident in investigating moving or staying were roughly 98 per cent of all individuals stay, i.e. have a y variable of 1). McKelvey, Zavoina and Laitilaís pseudo- R2 (R2-ZML) calculates as:

_ (8)

Because our Y variable is bivariate and the underlying actual probabilities are unknown, the pseudo-R2 puts the model.ís estimated variance in proportion to the underlying theoretical variance which consists of the estimated variance of y plus the assumed variance of the error term (which is standardnormal in the case of probit estimation, i.e. _=1). Alternatively, we will also provide the pseudo- R2 developed by Mc Fadden (1974) which has become popular as likelihood ratio index (R2-MF). It puts a model with k-explanatory variablesí maximum of the log-likelihood into proportion of a model with only a constant term as right hand variable:

_ (9).

Both pseudo-R2, (8) and (9), are constrained to the range 0 to 1. Indicating goodness of fit they should move in the same direction, but there is no theoretical reason for them to be identical at the absolute level.

4.3 Estimation Results
Table 1 shows the estimation results for all people aged between 19 and 64 who have stayed for at least 12 months prior to moving and for whom the necessary information is available. The sample comprises 271,806 observations, which allow for a relatively detailed combination of explanatory variables. Results are provided for five different models derived from equation (6). These five are all nested._ They are built in order to allow checking for the relative importance of our arguments and the stability of the different parameters. Model (1) includes duration variables linked to our insider-advantages hypothesis only. Model (2) adds the ëclassicí socio-economic micro and region-specific macro factors. Model (3) also includes ëlife-course eventsí, model (4) checks for the importance of different places of birth and model(5) introduces occupational dummies. A ë*í at the end of the Log-L term indicates that a Log-likelihood ratio test rejects the hypotheses that the model is not significantly better than the less comprehensive nested one with a probability larger than 0.99. It thus indicates that despite the absence of improvements in the pseudo-R2 the more comprehensive model includes additional explanatory factors that are jointly significant.

4.3.1 Duration of stay and insider advantages
Estimation results provide support for the insider advantage hypothesis that people are immobile because they accumulate location-specific insider advantages which are sunk in case of a move. The longer people stay at a certain place, the less likely they are to move. In model (1) we regress the probability of staying on just duration of stay in months, number of previous moves and for immigrants years since immigration. Immigrants should be more mobile (because they already lost their insider advantages), but their behaviour should approximate natives' overtime. To control for the possibility that the expected duration effects occur only because one gets older and is therefore less likely to move, we control also for age and allow the effect of age being non-linear by including age square into the regression. For duration of stay we have checked various specifications, but the most simple linear one applied here proved best._ In the estimation of model 1 duration of stay proves a highly significant explanatory variable and all regressors show the right sign. Although model (1) obviously suffers from omission bias, its goodness of fit is quite remarkable. the ëyears since immigrationí variable produces the expected effect clearly. Having just recently immigrated, foreign born people are significantly more mobile than native born. The number of previous moves still increases the likelihood of moving significantly, but the significance level of duration of stay is higher and already two to three additional years of staying usually increases the probability of staying by more than a previous move decreases it._

Comparing parameter estimates in model (1) to (5) reveals that model 1 suffers from omission bias and therefore overestimates the importance of the duration effects. But once we account for individual socio-economic characteristics and regional differences in model (2), the duration parameters continue to be among the most important explanatory factors of mobility behaviour. Even if we add further significant additional explanatory factors in subsequent models the duration effects remain very robust.

Supposing it is a relatively fair approximation of reality we use estimates obtained from model 3 to calculate some illustrative predicted probabilities of staying. Figure 13 shows how duration of stay changes the probability of staying. In probit estimations the probability of staying of an

Table 1: Estimation Results
Probability of staying in the Same Swedish Labour Market Region in 1994

> than 0.99.

**see appendix for further remarks on the definition of variables.

Table 1: (continued: results for model 3)

_EINBETTEN Excel.Sheet.8___

Table 1: (continued: results for model 4 and 5)

_EINBETTEN Excel.Sheet.8___(results continue on the next page)

Table 1: (results for model 4 and 5 continued)

_EINBETTEN Excel.Sheet.8___

source: own estimations.

Figure 13: Duration Dependence of the Probability of Staying -
Marginal effects by age and marital status

_Source: own simulations using estimation results of model 3.

sample (here about 0.98). This is what we find in figure 13 for individuals with mean characteristics and an average duration of stay of more than six years._ Having stayed just 1 year makes a mean individual 3 per cent more likely to move than having stayed immobile during the last ten years. This seems quite small a difference. Indeed, it is as such misleading because the small effect of duration of stay arises mainly from the fact that at average characteristics and the mean age of 41 all people are very, very unlikely to move. Therefore, all marginal effects in table 1 are very small. At mean

Figure 14: Duration Dependence of the Probability of Staying ñ
Some typical cases


Source: own simulations using estimation results of model 3.

characteristics the marginal effect of having moved once previously or never or of being married or single are comparatively much smaller even than the effect of having stayed for 1 year only or for five or ten years.

The small magnitude of marginal effects is first of all because they are each calculated at immobile prone mean characteristics, but it also results from several different explanatory variables simultaneously moving away from mean characteristics if somebody is less immobile. Remember graph 5 illustrating the unconditional effect of duration of stay on the probability of staying in our data. It has shown that while about every fifth of those individuals who have migrated a year ago moved again in 1994, less than half a per cent of those who have stayed for ten years did move again.

To overcome the problems with marginal effects at mean characteristics and to illustrate the working of marginal effects in the model, we have simulated the probability of staying by duration of stay for some exemplary individuals._ Figure 14 shows the results. It demonstrates how duration of stay makes a more important difference for individuals who are ëat riskí of moving. Assume an individual who has just completed first degree university studies, became unemployed and then decided to continue with some further education. Our person is now 24 years old and lives as a single in a rented apartment. For this exemplary individual ëAí the probability of staying ranges between 0.64 and 0.83. Whether the person has stayed for one or for ten years at the same place of residence makes a difference of 19 per cent in the probability of staying, while for the 40 year old, less skilled married person C who has two kids and owns an own house the difference amounts to less than 0.3 per cent.

In figure 14 the probability of staying is also given for a foreign born individual A. Having immigrated 5 years ago, the highly skilled foreign born is still between nine and four per cent (depending on the duration of stay) less likely to stay than the native born ëtwiní. According to our estimations it takes him 17 years of staying in the country of immigration before the years since immigration effect dominates the foreign born effect. Therefore, having become 40 years old, the probability of staying as calculated from our model and depicted in graph 14 is greater for the foreign born individual ëAí than for the native born. How fast the years since immigration effect dominates the foreign born effect depends on the level of education, because higher education increases the propensity to move less for foreign born than for native born. In our estimations the propensity to stay of a highly skilled foreign born equals the one of a native born after roughly 17 years while for medium skilled it may take 21 and for the unskilled more than 25 years to line up with their native born ëtwiní._ At the mean duration of 14 years since immigration (for which estimation results are most reliable) foreign born individuals are thus still somewhat more mobile than the native born population even if we account for differences in the other explanatory variables. However, compared to general socio-economic criteria, being foreign born proves to be of relatively marginal importance for the probability of staying._

To turn back to figure 13, the graph also illustrates the (small) marginal effect of a previous move on the probability of staying. At mean individual characteristics a previous move increases the propensity to move after a duration of stay of one year by less than half a per cent, after ten years by less than one pro mille.

Partial marginal effects of tje duration of stay are enforced by the age-specificity of immobility. While one stays, one gets older. Figure 13 draws the probability of staying for individuals with mean characteristics and zero previous moves at the age of 20, 40.7 (mean age) and 55. These differences by age are much larger than by marital status. Also, for younger ones duration of stay makes a greater difference to the probability of staying than for older ones. Figure 4 adds the qualification, however, that being 24 or 40 makes a large difference for the working single named ëAí, while it is of considerably less importance to the married exemplary individual ëBí. All in all, duration of stay proves to be an important determinant of the probability of staying which becomes enforced by changes in age, family formation etc. that typically occur while people stay immobile. Altogether they explain the strong overall relation between duration of stay and probability of staying shown in graph 5.

4.3.2 Traditional determinants of migration
In model (2) we add to the duration variables the factors commonly referred to in traditional migration theory, namely individual socio-economic characteristics and regional differences in macro variables. Although this increases the goodness of fit only moderately, the increase in the validity of the model as measured in terms of the Log Likelihood is highly significant._

In general, socio-economic variables show the expected signs and prove significant. Immobility is a very age-specific phenomenon. Apart from age, education turns out to be an important determinant of immobility. Presumably the skills of less educated people are less transferable locationally than those of the highly skilled and the return on moving is lower. Also in additional tests using census information on the socio-economic position of individuals we find less skilled blue collar worker as well as farmers to be particularly immobile while white collar highly skilled are particularly mobile (Fischer and Malmberg 1998). For immigrants, the mobility enhancing effect of education is significantly less pronounced than for natives. Again, marginal effects are very small, partly due to the fact that at mean characteristics all individuals are very unlikely to move.

Houseownership obviously contributes significantly to explaining immobility. Owning a house probably entails individual and place-specific rents that are sunk in case of a move and relevant enough to keep people from moving._

Contrary to what one would expect from traditional migration theory, regional differences in macroeconomic variables performed very poorly in explaining immobility. We constructed and tested various indicators of regional differences in employment and unemployment intensities, average income and wage levels as well as vacancies rates. They proved usually insignificant and regularly even show the wrong sign. Because regional employment patterns and wage levels are highly correlated, we had to chose a few particular indicators. As a crude proxy for the possibility that individuals value opportunities in the proximity different than more remote ones, we constructed regional indicators relative to the average in the neighbouring region on the one hand and relative to all other, more remote regions on the other hand.

Our estimations suggest that (im)mobility patterns do not seem to be dependent on regional employment intensity differences. People tend to be more immobile in regions where workerís mean income level is relatively higher. The variable we used measures mean income in thousands of Swedish Kroner for those who earned at least SKr 50,000 a year (as a proxy for regional development). But the (very small) effect is statistically significant only at the 10 per cent level (and insignificant at the usual 1 or 5 percent significance levels)._ The effect of the difference between the average income level in the labour market in which one was resident at the beginning of 1994 and the average in the surrounding labour markets turns out to be insignificant at the 10 per cent level. It also shows the wrong sign (indicating that people would be more mobile if the income difference were positive with respect to neighbouring regions). This is the more surprising as differences between the levels of average regional income from work were not totally neglectable: the lowest was 50,000 Swedish Kroner or 40 per cent less than the highest average. The difference between average work incomes in the ëhomeí and neighbouring regions ranged between SKr ñ32,000 and SKr +19,000.

The ëbestí performance of all the macroeconomic indicators scrutinised recorded regional vacancies rates relative to the average of non-neighbouring regions. But though remaining relatively stable in all models, the size of the effect again turns out very small and significant only at the 10 per cent level.

One regional difference turns out to be clearly significant and noteworthy, however: People living in the three large metropolitan regions of Sweden (Stockholm, Gothenburg and Malmà) are significantly and considerably more immobile than individuals living in the rest of the country. We have tested whether people living in larger, more urban labour markets are generally less mobile because they have more alternative opportunities nearby (we included the size of a regionís population in the estimations). We also checked whether the metropolitan effect could be primarily related to regions incorporating university towns. Both hypotheses we had to reject. We are thus inclined to conclude that metropolitan regions offer more scope for the accumulation of location-specific insider advantages and thereby keep people staying.

Concerning micro variables, it has proven very fruitful to include the individualsí family situation. Not only whether one is married or not but even more so whether one has a working partner affects the probability of staying strongly. The more a partner earns, the larger the potential costs or insider advantages that are sunk in case of a move._ Having children increases the probability of staying significantly too. In line with the premises of economic labour market theory people are found to be more mobile if unemployed. Less expectedly, being out of the labour force (i.e. not earning an income from (self)employment and not receiving unemployment benefits of more than Sr 20,000 a year) decreases the probability of staying by almost as much as being unemployed.

4.3.3 Life-course events
The observation that individuals who are out of the labour force are almost as mobile as the unemployed has probably to do with the fact that our results strongly support the hypothesis that (im)mobility is a phenomenon highly specific to certain life course situations. These life events may be labour market related but are also often not. Model (3) adds to the specification of model (2) life-course events. This again improves its validity in terms of log-likelihood very significantly (although the goodness of fit measures improves only slightly). From model (3) it becomes clear that being in education is one of the most important causes for mobility._ It increases the probability of moving by more than being unemployed. That individuals engaging in education are often also out of the labour force enforces the effect even further and probably explains much of the high relevance of the out of labour force variable.

Notwithstanding the fact that most people are young when they move, building a new or splitting from a previous household are also highly significant factors that induce people to move over labour market borders. Having a baby may also increase mobility, but for that the estimated coefficient is insignificant at any standards.

With respect to labour market issues we have also tested for the importance of becoming (and not just being) unemployed in 1994, for both, oneself and the partner, irrespective of the length of actual unemployment and the sum of benefits received. Both ëlife eventsí increase mobility and the coefficient of becoming unemployed oneself was highly significant._

Labour market issues provide an evident opportunity to illustrate the interaction between labour market issues, family situation and life course events. Figure 15 results from simulations of the probability of staying by duration of stay for an exemplary individual conditional on her employment and family situation. It shows that the 25 year old single who became unemployed during 1994 is most likely to move, although this will still be highly dependent on her duration of stay. Having just moved a year ago or living at the same place for ten years makes a difference of 11 per cent to her probability of staying. If the same individual is married to a partner who became unemployed in 1994, the probability of staying increases only moderately, by about one to two per cent. If on the contrary she is married to a partner who earns a substantial income from employment, then this

Figure 15: Duration and Probability of Staying
by employment status and family situation

_EINBETTEN Excel.Sheet.8___

source: own simulations using estimation results of model 3.

decreases her probability of moving by a remarkable 3 to 8 per cent. Whether the partnerís net work income amounts to SKr 80,000 or 150,000 is less important._ Indeed, the 25-year-old unemployed who lives with a working partner who earns SKr 80,000 a year turns out just as likely to stay as the employed single.

Being employed and married to a partner who also earns an income of SKr. 80,000 increases the likelihood to stay for the employed by another two to five per cent to between 94.8 and 98.7 per cent, depending on the duration of stay. Having a child makes then the minor difference of less than half a per cent additional increase in the probability of staying.

In brief the mobility effect of unemployment is not only dependent on individual socio-economic characteristics, but also on whether one became unemployed recently, is single or married, has children or not and last but not least whether the partner earns a substantial own income. Becoming unemployed is likely to be a relevant incentive to inter-regional labour market mobility in certain life course situations mainly, namely when one is young and single or young and married to a non-working partner.