Internal Migration in Great Britain: Causes and Effects

Transcription

1 Internal Migration in Great Britain: Causes and Effects A thesis submitted to the University of Manchester for the degree of Ph.D. in the Faculty of Humanities WILLIAM RODNEY WHITTAKER SCHOOL OF SOCIAL SCIENCES

2 Contents Chapter 1: Introduction... 7 Outline Chapter 2: Regional Inequality in Great Britain Regional Differences Since 1980: Is there a North-South Divide? The Effects of Regional Inequality on Individuals and Regions Summary Chapter 3: The Determinants of Regional Migration in Great Britain: A Duration Approach Introduction Background Data Methodology Results Conclusion Chapter 4: The Employment and Earnings of Migrants in Great Britain Introduction Background Data Methodology Results Conclusion Appendix Chapter 5: The Pursuit of Happiness: Internal Migration and Subjective Well-being in Great Britain Introduction Background Data Methodology Results Conclusion Appendix Chapter 6: Conclusion References Word Count: 34,074 2

3 List of Figures and Tables Figure 1 Regional Unemployment Rates Figure 2 Regional Vacancy Rates Figure 3 Regional Labour Market Tightness Figure 4 Regional Real Wage (Male) Figure 5 Regional Real Wage (Female) Figure 6 Productivity Per Hour by Region Figure 7 GVA Per Capita by Region Figure 8 Types of Spell Figure 9 Male Hazards Figure 10 Female Hazards Figure 11 Regional Labour Market Tightness Figure 12 Regional Real Wage (Male) Figure 13 Regional Real Wage (Female) Figure 14 Weibull Age Hazards Figure 15 Weibull Year Hazards (no aggregate variables) Figure 16 Male GHQ Score Figure 17 Female GHQ Score Figure 18 Male Life Satisfaction Score Figure 19 Female Life Satisfaction Score Figure 20 Histograms of the GHQ Questions Table 1 Individual Migration Rates Table 2 Results Table 3 Migration Rates Table 4 Reasons for Migration Table 5 Average Real Hourly Wage Rates Pre and Post Migration ( ) for the Employed Sub-Sample Table 6 Average Employment Rates Pre and Post Migration Table 7 Wage Estimates Table 8 Employment Estimates Table 9 Male Wage Results Table 10 Female Wage Results Table 11 Male Employment Results Table 12 Female Employment Results Table 13 Migration Rates Table 14 Migration Reasons Table 15 Total Number of GHQ Score Changes Table 16 Transition Matrix of Life Satisfaction Score t,t Table 17 Total Number of Life Satisfaction Score Changes Table 18 Raw Subjective Well-being Differentials Table 19 GHQ Score Estimation Results Table 20 Life Satisfaction Estimation Results Table 21 Male GHQ Estimation Results Table 22 Female GHQ Estimation Results Table 23 Male Life Satisfaction Estimation Results Table 24 Female Life Satisfaction Estimation Results

4 THE UNIVERSITY OF MANCHESTER ABSTRACT OF THESIS/DISSERTATION submitted by WILLIAM RODNEY WHITTAKER for the Degree of Ph.D. and entitled INTERNAL MIGRATION IN GREAT BRITAIN: CAUSES AND EFFECTS Month and Year of submission September Using nationally representative, longitudinal data from the first 15 waves of the British Household Panel Survey, this thesis investigates the determinants of internal migration between the regions of Great Britain, and the effect of migration on employment, earnings, and subjective well-being. There are three substantive empirical analyses. The first analyses the determinants of regional migration conditional on duration spent in a region. This is important because previous models not controlling for this are prone to biased estimates on other calendar time variables such as age and year. Further, the comprehensive treatment of duration is more accurate than past studies, which, rather than include duration in the region/local Authority District, measure duration at current address. This will tend to underestimate the true duration in a region and thus affect the estimation results. We find significant negative duration dependence on the hazard to migrate. Thus the longer a person resides in a region, the less likely they are to migrate. This is particularly prominent in the first few years spent in a region. The second analysis of the thesis estimates the impact of migration on individual earnings and employment. Controlling for endogeneity, heterogeneity and selfselection, we find substantial long-run wage premiums associated with migration for both males and females who move for job-related reasons. There is, however, no evidence that moving across regions increases the probability of employment for males and females; in fact, some female movers experience a long-run employment penalty. Migration should be the response of rational individuals aiming to maximise expected lifetime utility, as such, the absence of earnings or employment gains for some migrants found in this thesis and in the wider literature suggests other factors may be influencing the decision to migrate. The final study in the thesis takes a unique perspective on measuring the returns of migration by analysing the effect migration has on individual subjective well-being. Subjective well-being as a measure of welfare for an individual encompasses all aspects of ones life, allowing a more meaningful yardstick by which to measure migrant success. We find significant gains to subjective well-being for males and tied females, in particular, these gains are primarily amongst those not found to benefit from migration in previous studies. 4

5 DECLARATION AND COPYRIGHT DECLARATION No portion of the work referred to in the thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning. COPYRIGHT i. The author of this thesis (including any appendices and/or schedules to this thesis) owns any copyright in it (the Copyright ) and s/he has given The University of Manchester the right to use such Copyright for any administrative, promotional, educational and/or teaching purposes. ii. Copies of this thesis, either in full or in extracts, may be made only in accordance with the regulations of the John Rylands University Library of Manchester. Details of these regulations may be obtained from the Librarian. This page must form part of any such copies made. iii. The ownership of any patents, designs, trade marks and any and all other intellectual property rights except for the Copyright (the Intellectual Property Rights ) and any reproductions of copyright works, for example graphs and tables ( Reproductions ), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property Rights and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property Rights and/or Reproductions. iv. Further information on the conditions under which disclosure, publication and exploitation of this thesis, the Copyright and any Intellectual Property Rights and/or Reproductions described in it may take place is available from the Head of School of Social Sciences (or the Vice-President) and the Dean of the Faculty of Life Sciences, for Faculty of Life Sciences candidates. 5

6 Dedication This thesis is dedicated to my Mum and Dad. Mum, I miss you and hope this makes you proud. Dad and Irene, my inspiration and driving force in making me believe that anything is possible, from humble beginnings to Ph.D. your boy done good! With thanks to Darren, for putting up with losing your partner for four years! Acknowledgements Financial support was provided by the Economic and Social Research Council. Supervision of the PhD was provided by Ken Clark and Martyn Andrews, who provided valuable contributions to the thesis, in particular, to Chapters 3 and 4. All models and statistics were estimated using STATA. Data on regional house prices were obtained from the Halifax Building Society. Claimant Count unemployment data, Job Centre Plus vacancy data, and Gross Value Added data were provided by nomisweb and the Office of National Statistics. Regional wage rates were obtained from the New Earnings Survey and the Annual Survey of Hours and Earnings. The data (and tabulations) used in this thesis were made available through the ESRC Data Archive. The data were originally collected by the ESRC Research Centre on Micro-social Change at the University of Essex (now incorporated within the Institute for Social and Economic Research). Neither the original collectors of the data nor the Archive bear any responsibility for the analyses or interpretations presented here. Useful comments were received from participants, (particularly Mette Christensen) at the University of Manchester Econometrics and Applied Economics Workshop 2007, the 2007 WPEG Annual Conference at the University of Manchester, the 2007 BHPS Conference at the University of Essex, and the 2007 EALE Conference in Oslo. The Author William s first course of study at the University of Manchester, was for the degree of B.Sc. in Economics, in which he passed with first class honours. Funding was then obtained by the ESRC for his second and third courses, an M.Sc. in Economics and Econometrics ( ), and PhD in Economics ( ). William would like to express his extreme gratitude for the persistence, motivation, and invaluable guidance provided by his supervisors, Martyn Andrews and Ken Clark; throughout the PhD. 6

7 Chapter 1: Introduction 7

8 This study is motivated by the presence of economic differentials between the eleven standard statistical regions of Great Britain. Such differentials, which will be reviewed in Chapter 2, ought to be arbitraged away by between-region migration. Indeed, each year approximately 1.6% of the population migrate between the eleven standard statistical regions of Great Britain and some evidence suggests these flows are in the direction expected, with migrants moving to better performing regions (Jackman & Savouri 1992). The apparent market failure suggested by the persistence of inter-regional differences has been attributed to a lack of internal migration. Net migration, out of high- and into low-unemployment regions could help reduce regional inequality by lowering unemployment and wage differentials. However, such flows are small, and are particularly low during periods of recession when most needed. The presence of regional differentials in Great Britain is a significant problem. A Treasury report highlights the magnitude of the problem: If all lagging UK countries and regions improved their productivity performance to at least that of the current average, the UK s overall growth rate over the last twenty years would have increased by approximately an additional 0.4 per cent per annum. (HM Treasury 2001, p.1). As part of its strategy to improve national productivity, the UK government has considered ways of boosting labour mobility in order to narrow productivity gaps between different regions of the country. If the government wish to address the 8

9 migration issue, successful policy would need to be informed by good, solid, micro evidence on both the causes and the effects of regional migration. Migration is viewed within the human capital framework as a response (and potential solution) to regional inequality. Individuals will migrate to a different region if there is a net gain to expected utility from doing so. In this framework, regional differentials play a significant role, where regions with low unemployment rates (high vacancy rates) should be seen as more attractive to individuals in high unemployment (low vacancy) areas since this may signal a stronger demand for labour. Wage differentials will represent higher potential earnings whilst house price differentials could act as potential barriers to entry amongst regions. Knowledge of migrant characteristics will help our understanding on the causes of migration and the types of individuals flowing into regions, and would also highlight those individuals who may be constrained in moving, enabling better informed policies that impact on migrant flows. Further, the degree to which migration may help reduce regional differences relies on the individual outcomes of migrants. Do migrants find employment and higher earnings following a move? If not, are there other benefits from moving? The answers to these questions are crucial to our understanding of the effects of migration on individuals, households, and Great Britain as a whole. In this thesis we use the first fifteen waves of the British Household Panel Survey ( , henceforth BHPS), to analyse internal out-migration between the eleven 9

10 standard statistical regions of Great Britain. 1 We model out-migration rather than net-migration since the effects on origin and destination region are difficult to measure, primarily since changes in regional population can occur for many reasons, fluctuations in international migrants or birth and death rates, for example. Secondly, migrant flows are particularly small, which causes empirical and measurement problems for studies of net-migration. The inter-regional migration specification was chosen to capture transitions between different labour markets. However, as with any regional specification, there is a possibility that we include some migrants we do not wish to model, namely, those moving regions but continuing to work in their original labour market. Modelling moves between regions rather than residential or Local Authority Districts reduces the likelihood of this occurring. In addition, in the proceeding analyses we exclude those migrants who move and remain with their existing employer. Though this does not guarantee we have completely removed the problem migrants (someone may move region to attain better housing and commute back to the region, albeit with a new employer, and be measured in our analysis as a migration), this is the best we can do to ensure our migrants are entering a new labour market. The BHPS was designed as an annual survey of each adult (16+) member of a nationally representative sample of more than 5,000 households, making a total of approximately 10,000 individual interviews. The same individuals are re-interviewed 1 These are London, South East, South West, East Anglia, East Midlands, West Midlands, Yorkshire and Humberside, North East, North West, Wales and Scotland. 10

11 in successive waves and, if they split-off from original households, all adult members of their new households are also interviewed. Children are interviewed once they reach the age of 16. Thus the sample should remain broadly representative of the population of Britain as it changes through the 1990s (Taylor et al. 2001). The BHPS contains four booster samples. The first began in wave seven, this was a sub-sample of the United Kingdom European Community Household Panel (UKECHP). Together with a low-income sample of the Great Britain panel, this sample was generated for the European Community Household Panel (ECHP). The ECHP sample additions tend to overstate low-income households and individuals. The remaining new samples were added in waves nine and eleven, these were (extended) samples in Scotland and Wales, and a new sample from Northern Ireland. The Northern Ireland, Scotland, and Wales new samples overestimate the respective population by 7, 2.5, and 4 times that of the United Kingdom population distribution respectively (Taylor et al. 2001). We exclude these additional samples as these may bias the results in the analyses later. This has the effect of excluding all respondents from Northern Ireland. The panel nature of the data enables us to construct profiles of individuals over time, recording region of residence, as well as labour market and personal characteristics, and is thus ideal for migration analysis. In particular, we can compare earnings, employment status, and subjective well-being scores before and after a move between regions. Further, the panel structure allows intra- as well as inter-personal 11

12 comparisons to be made: in particular, this means that controlling for unobserved heterogeneity is possible. The thesis contains three, largely self-contained, empirical investigations. In Chapter 3, we analyse the determinants of migration. Using survival model techniques, we model duration spent in the current region. The detailed, retrospective questions contained in the BHPS, allow us to construct a measure of duration at current residence ( elapsed duration ). To our knowledge this is the first study to place regional migration in Great Britain within a duration model and one of few studies of any country to additionally control for unobserved heterogeneity. Whilst previous studies have controlled for time at current address, this may underestimate the actual time spent in a region (and thus affect the estimates), since there may be multiple moves within, but not between, regions. Our treatment of duration dependence, leads to a more comprehensive model of migration. Chapter 4 of the thesis estimates the impact of migration on individual earnings and employment. Unlike most of the previous literature, which has focused on youths in the US and either males or couples for the UK, our sample comprises males and females of working age. Furthermore, in contrast to previous studies using the BHPS, we focus on migration across the boundaries of the eleven standard statistical regions of Great Britain. As well as being of relevance from a policy perspective, the regional focus allows a cleaner test of human capital theories of migration. This is because, compared to short-distance movements between Local Authority Districts (LAD), an inter-regional migrant is much more likely to be changing labour market 12

13 and will therefore incur more substantial costs of migration. In this scenario, the existence of small or insignificant labour market returns to migration challenges the validity of a pure human capital approach to explaining migration decisions. In our investigation of post-migration outcomes we distinguish between men and women, and for the latter, whether they can be considered as tied movers or not. For both genders we also study whether the returns to migration differ if migration was explicitly for job-related reasons or was motivated by other considerations. Furthermore, our use of panel data enables us to distinguish between the long-run and short-run impact of migration on labour market outcomes. This reflects the idea that migrants may take time to adjust in the destination labour market. Finally, our empirical approach addresses three specific reasons why migration might be endogenous. These concern unobserved heterogeneity, reverse causality and sample selection. The results in Chapter 4, and previous studies, show that the returns to migration in terms of higher earnings and better employment prospects are lower for women and vary according to the reasons for migration. Significant gains to wages have been found for job-related movers only. Whether or not migration leads to better job prospects depends on whether migration is specified as regional, as in Chapter 4, or between LAD (Boheim & Taylor 2007; Rabe 2006; Taylor 2006). A significant problem with the use of wage and employment as measures of utility, is the large majority of migrants experiencing no benefits having moved. The problem 13

14 arises because wage and employment effects on their own, reflect only part of the overall change in utility. To capture a more direct measure of utility, in Chapter 5 we measure the effect of migration on subjective well-being and life satisfaction scores. The BHPS includes several questions on subjective well-being that allow one to correlate changes to subjective well-being scores with changes in circumstance. To the best of our knowledge, no study has estimated the effect of internal migration on subjective well-being in Great Britain. Migrants may exhibit important differences to non-migrants, not least since migrants have necessarily severed friendship and familial ties having moved and are typically in a position of having to build up knowledge of the new area. To this extent, it would seem migrants may have lower subjective well-being levels, at least until they are fully integrated in the new region. This is consistent with the human capital approach where individuals weigh up the effect of migration on their net expected lifetime utility. We therefore investigate the short- and long-run effects of migration on subjective well-being scores. Taken together, Chapters 3, 4, and 5 provide a comprehensive empirical analysis of the causes and consequences, at the individual level, of inter-regional migration in Great Britain. The findings of this thesis will help inform policies attempting to increase regional productivity and reduce regional differentials. The introduction of subjective well-being to the literature opens new avenues for research in the migration literature and suggests how future work could adopt a more fundamental approach to modelling migration and its impact on individuals and regions. 14

15 Outline Chapter 2 provides an overview of the trends in key economic indicators across the regions of Great Britain. Chapter 3 models the determinants of migration, and duration dependence as detailed above. Chapters 4 and 5 model the returns to migration on employment and earning, and subjective well-being respectively. Detailed reviews on the relevant literature are provided in each Chapter. The main conclusions of the three studies and areas for future research are given in Chapter 6. 15

16 Chapter 2: Regional Inequality in Great Britain 16

17 In this chapter we describe the extent to which there are regional differences. This is important because we construct regional-level covariates that are potentially correlated with both migration and the returns to migration in Chapters 3 and 5. The chapter begins by discussing whether there exists a North-South divide amongst the regions of Great Britain. Trends in regional unemployment, vacancies, wages, productivity, and Gross Value Added figures are then presented. The chapter concludes with an analysis of how these differences may influence, and be influenced by; migration. Regional Differences Since 1980: Is there a North-South Divide? The poorer performing regions tend to be situated in the North of Great Britain. This North-South divide emerged during the 1920s where many traditional industries in the northern regions declined. Unemployment differentials persisted for the rest of the twentieth century, with a tendency for these to widen during periods of recession and narrow with prosperity (Pissarides & McMaster 1990). The 1980s saw the widest dispersion in regional unemployment rates since the 1930s. This was because an appreciation of sterling, and tight monetary policy in the early 1980s had reduced manufacturing competitiveness, and hence resulted in disproportionately higher job losses in those (northern) regions heavily reliant on the manufacturing sector (Evans & McCormick 1994). The gap between the northern and southern regions narrowed towards the end of the decade due to labour shortages in the south (Johnes & Taylor 1996). The recession of the early 1990s caused a temporary increase in unemployment differentials, but since then the differences 17

18 have declined continuously, prompting many to question the existence of a North- South divide by the end of the century (Hughes & McCormick 1994; Jackman & Savouri 1999; McCormick 1997). Jackman and Savouri (1999) find the variance of unemployment rates across regions has reduced substantially in the 1990s, suggesting a possible decline in the level of imbalance. They note however, that this can be explained by the structure of the recession compared to those of the past. The variance was low since the recession was neutral across regions ( neutral in the sense that southern regions were now hit as hard as the northern regions, hence performing relatively worse than in previous downswings). They claim Britain has to some extent solved the regional problem and that this has been due to manufacturing accounting for a smaller proportion of the workforce and on the expectation that shocks will not hit the economy as hard as they previously have. It is unlikely future recessions will have the same characteristics: whilst unemployment differentials have reduced in the last decade, demand-side shocks to the manufacturing sector could widen the differentials once more. Figure 1 shows the convergence of unemployment rates across the UK since 1991: following the early 1990s recession, unemployment rates have narrowed and stabilised. Figure 2 plots Job Centre Plus vacancies by region, the data show that northern regions have relatively higher vacancy rates. To highlight the differences in job opportunities between regions, Figure 3 graphs labour market tightness by region. A clear explosion in regional differences is evident, one may expect the asymmetric 18

19 opportunities to be short lived, being mitigated by the movement of potential employees to regions with better prospects for finding work. This does not seem to be the case. Figure 1 Regional Unemployment Rates Claimant-count unemployment rates Data Source: Nomisweb 19

20 Figure 2 Regional Vacancy Rates Job-centre plus vacancies Data Source: Nomisweb Figure 3 Regional Labour Market Tightness Tightness defined as job-centre plus vacancies divided by claimant-count unemployment rates Data Source: Nomisweb 20

21 Although unemployment rates have converged during the 1990s, a balanced rate of unemployment across regions disguises the actual employment position of individuals. It may still be the case that areas in the North suffer lower living standards due to the dependence of local labour markets on specific occupation and industrial sectors. Figure 4 and Figure 5 reveal a widening of real wage differentials for both males and females. The real wage is measured here as the average wage divided by the average house price in a region, and was calculated in this way to control for regional differences in costs of living. The decline in the real wage from 1995/1996 is a reflection of increasing house prices over the period, differentials have widened, with wages in southern regions not compensating the increasing disparity between regional house prices. Figure 4 Regional Real Wage (Male) Real Wage defined as average regional wage rates (by manual status) divided by average regional house price Data Source: New Earnings Survey, Annual Survey of Hours and Earnings, and Halifax Building Society 21

22 Figure 5 Regional Real Wage (Female) Real Wage defined as average regional wage rates (by manual status) divided by average regional house price Data Source: New Earnings Survey, Annual Survey of Hours and Earnings, and Halifax Building Society The Effects of Regional Inequality on Individuals and Regions The negative effects of regional differences are most prominent during periods of recession. There is some evidence, of the different effects recessions have on individuals in different regions. A study by Clark, Leslie and Symons (1994), using the Family Expenditure Survey to estimate the impact of the 1980s recession on individuals expected utility, suggests the impact of northern regions was almost double that of southern regions. Whittaker (2004) extends the methodology of Clark, Leslie and Symons (1994) to the 1990s recession, and found that: although the recession was relatively more severe in the southern regions in comparison to the previous recession; northern 22

23 regions still faced higher loses in expected utility borne from the slump. However, the differences in costs of recession between the northern and southern regions had narrowed. Northern regions now experienced approximately 50% higher costs than the southern regions, there was also evidence of a marked increase in London costs to a level matching those in the North. This was largely because financial liberalisation in the 1980s enabled vast amounts of borrowing, particularly for owner-occupiers of which the southern regions had a relatively higher rate. The sudden increase in interest rate caused the downswing which particularly harmed those with credit. The asymmetric costs of recession should, in the human capital approach to migration, encourage migrant flows. However, migration tends to be pro-cyclical with the business cycle (Champion 2005; Jackman & Savouri 1992). This suggests any removal of the constraints facing potential migrants could help reduce the costs of recession on the individual, and further, were migrants to gain from better job prospects having moved, this may generate gains at the regional level and for Great Britain as a whole. Regional inequality has a direct impact on the performance of regions. The differences in productivity between the regions of Great Britain can be seen in Figure 6. Only London and the South East have above average productivity (base UK=100). Of those regions below average, the southern regions are typically above the northern regions. Figure 7 plots regional GVA per capita at basic prices for the same period (1996 to 2006). Many southern regions, namely London, the South East, and East 23

24 Figure 6 Productivity Per Hour by Region Base UK=100 Source: ONS Figure 7 GVA Per Capita by Region Base UK=100 Source: ONS 24

25 Anglia have remained stable in being those regions with higher per capita GVA than the UK average (UK=100). One may expect regional differences to be reduced over time, with the incentives for firm and individual to relocate in those more prosperous regions thus alleviating high unemployment in the poorer regions and dampening wage pressures in the better performing areas. Figure 6 and Figure 7 suggests this is not happening in the UK; indeed, the differences tend to be increasing. This is particularly worrying since the Treasury estimate the UK economy would have grown by an addition 0.4 per cent per annum had all poorer performing regions had productivity levels at least that of the UK average (HM Treasury 2001). Summary Since the late 1980s, the UK economy has experienced a reduction in regional unemployment differences. However, large differences exist in real wages and vacancies between regions. There is evidence of the asymmetric costs individuals face due to regional differences during periods of recession, and a suggestion by the Treasury that regional inequality in terms of productivity differences has led to large losses for the UK economy. Increasing the performance of lagging regions in the UK would thus reap gains to individuals, regions, and the UK on the whole. The aim of Chapters 3, 4, and 5, is to investigate the potential of internal migration to help reduce regional differences, and in doing so, foster a higher level of regional performance. 25

26 Chapter 3: The Determinants of Regional Migration in Great Britain: A Duration Approach 26

27 Introduction We begin our analysis by assessing the determinants of migration. This is important because we need to know what factors drive or constrain migration, in order to understand the impact migration may have regions, and how policy could influence migrant flows. In Chapter 1 we placed migration in a human capital context, in this framework, we would expect migrants to be those facing lower costs and/or higher benefits from moving. Our analysis may also be viewed as a test for this approach. Whilst there has been a large volume of research in this area, our study makes a number of contributions. We model migration using survival model techniques, this enables us to model duration in a region. To the best of our knowledge this is the first migration study to control for the years spent within a region. This is important since past studies have used time spent at the current address to control for duration effects. This is not the same as time spent in a region, and as such, will underestimate the true effect of duration dependence on regional migration. Our study models migration from , and as such, includes several waves of the BHPS that were conducted since the completion of previous studies. Unlike the majority of the literature, we do not limit our study to the migration of males and/or heads of households. This, we feel, provides a more modern approach to migration, particularly given the rise in female participation and changing family dynamics towards the end of the last century. 27

28 The chapter begins with a survey of past studies that analyse the migration decision. This is followed with a discussion of the data used, which contains a detailed description of how we construct our duration in a region variable and migration dummy. The methodology section sets out the estimation techniques used, this is followed by an analysis of the results and concluding comments. Background To model migration, one needs information on individuals through time, we need to observe changes in both region of residence, and other personal and regional characteristics that may influence and/or be influenced by migration. The BHPS is the most appropriate dataset to model migration in Great Britain. Since 1991 the survey has followed respondents, monitoring changes in region of residence as well as changes in a wide range of personal and labour market characteristics. We are able to create detailed time profiles of individuals and use this to determine those factors that influence the decision to migrate. Boheim and Taylor (2002) was one of the first studies to model migration behaviour in the BHPS. Using the first seven waves of the BHPS ( ), the paper investigates the reasons to move house and the extent and determinants of house moves. Three models are estimated, a random-effects probit (for all household moves, between and within region), a bivariate probit (again, all household moves), and a 28

29 multinomial logit (a comparison of the difference in migrant characteristics of local and regional moves). The random-effects probit model expresses the decision to move as a function of employment status, family characteristics, housing tenure, moving preferences, and regional characteristics. They find individual unemployed increases the propensity to migrate, thus this model supports the Classical economic hypothesis that movement is instigated by unemployment experience. Like Jackman and Savouri (1992) below, the results show a negative relationship between migration and duration of unemployment. Unlike a previous study of household moves by Henley (1998), Boheim and Taylor find that local unemployment does increase the likelihood of moving, and attribute the period (early 1990s) used by Henley to be the cause for his results. Individuals tend therefore, to move from low local labour demand. The random-effects probit also contains duration at current address. The estimates suggest that with each year spent at the current address, the likelihood of migration falls by 0.3%. These estimates give some idea of the effects of duration dependence, though are not directly comparable with the specification we use later. Our argument stems from the idea that migration will be affected by the ties to a region (rather than address), as such intra-regional movements, in our framework, do not constitute an end of spell in the region. The simplest example would be the movement of a household to an adjoining street. The same measure of duration was used in a study by Buck (2000) who used the BHPS ( ) to model duration spent at the 29

30 current address. This measure of duration was used irrespective of the definition of migration (residential, intra-regional (LAD), or inter-regional). The bivariate probit model estimates the relationship between moving residence and changing employment status. This was used since the decision to move may influence the employment status of an individual (the reverse also being a possibility). Our migration specification omits those moving and remaining with their present employer. Further, using regional moves ensures these are only a small portion of migrants. A multinomial logit model was estimated to identify differences between the influences on the decisions for moving short and long distances. The unemployed appear to be the most mobile, particularly between regions. Wanting to move for job reasons has the largest quantitative impact on the likelihood of moving between and within regions. The results suggest there are key differences in the types of migrants (and their motives) with the distances moved. Until the introduction of the BHPS in 1991, the National Health Service Central Registrar (NHSCR) was the only available panel data to monitor individuals through time. Data from the NHSCR comes from the registration of individuals with family practitioners, we can observe migration where an individual registers with a new practitioner in a different region. Only information on the gender and age of individuals are reported, this limits any microanalysis of migration. Further, everyone registered with a practitioner is contained in the data, thus, whilst the data is by far the largest available for Great Britain, it is impossible to distinguish between 30

31 movements of students and movements for retirement reasons with the main subject of economic analysis, the movement of the labour force. Jackman and Savouri (1992) used the NHSCR for the period to analyse the causes of time series variation in bilateral flows of regional migration. They estimate, via Pooled OLS and fixed-effects, the effects of regional characteristics on the regional migration rate. They find unemployment, vacancy, and house price differentials work in the expected direction, with individuals moving from areas of low/poorer labour markets to more prosperous regions. For wages, a perverse effect was found, though the authors claim this may be due to the dampening effect of migrants on destination wages in better performing regions, and/or migrants being more likely to be high-earners and thus less responsive to wage differences. The large volume of observations in the NHSCR is a key benefit over other data sources. Smaller sample sizes necessarily reduce the number of migrations under study, this has the effect of lowering the numbers of net-migration to a level unsuitable for econometric modelling. Cameron and Muellbauer (1998), use NHSCR data for the period , estimate the effects of regional labour market characteristics on net migration. They find negative relative unemployment and house price effects, suggesting rising house prices (unemployment) tend to reduce inmigration (out-migration) to a region. A positive earnings effect was found, suggesting migration flows to regions with higher relative earnings. Because of the low numbers of net-migration observations available in other sources of data, the bulk of the literature focuses on estimating out-migration. Hughes and 31

32 McCormick (1994) are an exception, they estimate the decision to migrate (via a nested logit model), using a pooled LFS dataset for 1981 and , and are able to analyse the effects of regional labour market characteristics on net migration. The majority of studies use pooled cross sections of the General Household Survey (GHS) or Labour Force Survey (LFS). These studies typically model migration as a binary variable (equal to one if the individual migrates), at the regional level, and almost exclusively, restrict migration to movements of heads of household or males. The main findings suggest those that migrate are typically: males, those without children, the better educated, and younger generations. Housing constraints faced by council tenants and mortgage holders are consistently found to hinder migration (Hughes & McCormick 1981, 1985, using the GHS). There has been mixed evidence on the effect of individual unemployment. Pissarides and Wadsworth (1989) studied the likelihood of migration with unemployment and found: individual unemployment increases the likelihood of migration, unemployment differentials had seemingly no effect but higher unemployment overall reduces the likelihood. Due to the lack of available panel data, their analysis on overall unemployment rates were made by comparing migration in the LFS for 1984 to that of There is also mixed evidence on the effect of regional differences on migration propensities (Hughes & McCormick 1994; Jackman & Savouri 1992; McCormick 1997; Pissarides & Wadsworth 1989). The reliance of previous research on cross-sectional or aggregate (regional level) data, has resulted in the inability to control for two factors which are likely to impact 32

33 upon an individual s decision to migrate. The first of these is duration dependence it seems likely that a longer spell in a region will alter the probability of migration. The omission of this variable is likely to cause bias since time in a region may be correlated with other explanatory variables. The second problem is unobserved heterogeneity. Some individuals, given other observable characteristics, may be more footloose than others and hence more likely to switch regions at any point in time. The omission of this from the analysis is likely to bias the estimates. Data To investigate the determinants of migration, our analysis uses the first 14 waves of the BHPS ( ). 2 We exclude students, the retired, and those in the armed forces, because migration for these groups is unlikely to be for labour market reasons. Our sample consists, therefore, of those aged who are either employed or unemployed/inactive. Each individual is interviewed once a year (at a date that varies from year to year). The data form an unbalanced panel comprising 6,266 females and 5,986 males observed over 44,366 female-years and 39,569 male-years. Duration in each region (a spell ) is measured in integer years. 3 2 We use wave 15 (2005) of the BHPS to distinguish those moving in wave 14 (2004). We do not know which individuals migrate in a wave without information of region of residence in the proceeding wave. 3 When we say an individual was observed in 1993 we mean that he was observed by the interview date in Even though latent durations are continuous, actual durations are observed only at 14 discrete points in time. 33

34 Because we are interested in the labour market effects of migration, our definition of migration excludes those who change region but remain with their existing employer (for example, those moving to be closer to their place of work, or relocations for internal promotion reasons). Thus, where an individual is employed post-migration, we analyse those who change job following migration. There are 511 male migrants in our sample out of 5,986 males, and 544 female migrants from 6,266 women (see Table 1), converting to ever-migrated rates of 8.68% for females and 8.54% for males. On the other hand, the annual migration rate is 1.63% for females and 1.71% for males. These migration rates are lower than those found in previous studies of the UK/GB since we are using movements across the eleven standard statistical regions of Great Britain as the definition of migration rather than relatively smaller movements across local authority boundaries. Note that Census data indicate that two-thirds of migrants move less than 10km while only one if fifteen move more than 200km (Champion 2005). 4 The BHPS contains essential retrospective information without which our analysis is not possible. In their initial interview, respondents are asked how long they have resided at their current address. This provides us with elapsed duration at their current residence. From this we are able to construct information about each individual's spells of residence in a given region. 4 Our rates are similar to the regional rates of 1.6% and 1.8% found by Jackman and Savouri (1992) using NHSCR data, and Boheim and Taylor (2002) using the B HPS, respectively. 34

35 Table 1 Individual Migration Rates Males Females Number Migrations (a) (m=1,c=1) Number Migrants (b) Sample Size Person-Year (c) 39,569 44,366 Sample Size Persons (d) 5,986 6,266 Migration Rate (Person-Year) (a/c) Migration Rate (b/d) Number Completed Spells (e) (c=1) 4,603 4,821 Number Censored Spells (f) (c=0) 2,359 2,642 Number of Spells (e+f) 6,962 7,463 Left-truncated First Spell (inds) 3,381 3,771 Flow Sample (inds) 3,581 3,692 We start the duration clock when an individual turns 16 years old, or, if in education, when the individual finishes her studies. Thus where someone has never moved home, her elapsed duration is measured from age 16 or time completed education. For someone who has moved since turning 16 or completing her education, we observe duration as the number of years at the current address. We define the elapsed duration of individual i's first spell when first observed in the BHPS as a i1. Thus, for many individuals, their first spell is left-truncated, as it starts before 1991, and a i1 1. Individuals whose first spell starts in 1991 or later are not left-truncated, defined by a i1 1. The econometric methodology below needs to distinguish between these two sub-samples of individuals. Of 6,266 females and 5,986 males, 3,771 females and 3,381 males have left-truncated first spells. The sample of left-truncated first-spells is referred to as a stock sample and lefttruncation is referred to as late-entry. The remaining spells form a flow sample. The unit of observation is our analysis is a spell: there are 14,425 spells in the data, giving the average number of spells per individual as 14,425/12,252=

36 Figure 8 Types of Spell Date Interviewed ( ) Late entry, right censored Late entry, retires a 1 4, a1 17, m1 0, c1 0 a 1 11, a1 20, m1 0, c1 1 Migrant with late entry Mature migrating student Education a 8, a a a a , a 2, a 1, a , m 7, m 6, m 2 1 6, m 2 1 1, c 0, c 1, c 1 0, c Mature student Education a, a 9, m 0, c year old entrant a 1 1, a1 11, m1 0, c1 0 36

37 Figure 8 gives six stylised examples of individuals in our data. The first four individuals have left-truncated first spells. The other two are new to the labour force, starting either in 1991 (the mature student ) or later than that (the 16-year-old entrant ). The duration of the individual's first spell is denoted a i1. Thus, for an individual whose first spell starts in 1991, and finishes in 1992 (maybe because they migrate), then a i1 1, a 2, and so the number of years the first spell is observed in the i1 BHPS is two a a 1 2). For the second individual in Figure 8, his first spell ( i 1 i 1 ends in 2001 and so his duration is 10 years. It is important to note that our duration variable measures time spent in a region. As individuals may move within a region, our derived duration variable differs from the years at address variable given in the BHPS. For example, the second individual in Figure 8 could have changed address but remained in his current region numerous times over the 11 years in the region. Using years at current address will therefore understate the time spent in a region. Years at current address has been (incorrectly) used in numerous past studies analysing inter-regional moves, in particular; Hughes and McCormick (1985) using the GHS, and Boheim and Taylor (2002) and Buck (2000) using the BHPS. Some individuals have more than one spell. The duration in spell s is denoted a is. Consider the migrant in Figure 8. The last year observed in his first region is 1998 and so a a 1 8. The duration of his second spell is a 7 i 1 i 1 i2. In 37

38 general, ais ai1 1 m, where m is the number of Waves of BHPS that the s individual is observed in. There are three reasons why spells end: an individual can migrate, leave the sample ( attrit ), or the spell is right-censored in Attrition occurs because an individual leaves the labour force (retires or dies) or leave the BHPS. We define two binary indicator variables: m 1 if an individual migrates and c 1 if the spell is i completed. A completed spell can either end in migration ( c 1, m 1) or attrition ( c 1, m 0) ; otherwise the spell is right-censored ( c 0). In Figure 1, five spells i i i i i i are right-censored. Attrition is denoted by and migration by. In the data, there are 2,359 censored male spells and 2,642 censored female spells. From the completed spells, we can compute the raw hazard rate to completing a spell, which is 4,603/39,569=0.116 for men and 4,821/44,366=0.109 for women. Of the 4,603 completed male spells, 677 finish as a migration (14.7%); for females, these figures are 4,821, 722 and 15.0% respectively. As 14.7% out these 4,603 completed male spells end up in a migration, the raw hazard to migration for men is 677/39,569=1.71%. For women it is 722/44,366=1.63% because 15.0% of completed spells end up as a migration. The movement of individuals may cause problems of attrition within the BHPS. This is unavoidable but does not seem to be too problematic. Buck (2000) gives a response rate of 72% for migrants between waves one and two, with an overall response rate of 87%. Response rates for all respondents decreased wave on wave to 90% and 95% for waves two and three, and three and four respectively. Taylor (2006) 38

39 claims at least one household member could be interviewed in 80% of all moving households over the first thirteen years of the BHPS. Defining spells for mature students is potentially problematic. We stop the duration clock during the years they are being educated. When an individual has moved region during their studies and not returned, we reset our duration variable but do not record the change in region as a completed spell since migration here was for nonlabour market reasons (the fourth individual in Figure 8). For other individuals we do not pause the clock because they might leave the labour force for labour-market reasons. These include women in domestic production, the long-term sick and disabled, and those on government training schemes. To summarise, spell s for individual i is characterised by the following vector of information: 5 ( is is is is is a, a, m, c,x ) s 1,...,7. For all spells apart from the first, a 1, and a 1 if the individual enters the sample in 1991 or later. is i1 x is represents all observed covariates, which can potentially vary over the elapsed duration of the spell and by calendar time. Each spell comprises a a 1 rows (years) of data. is is 5 Just one male has 7 spells. 39