Exploring the drivers of internal labour migration for the regions of Great Britain

ABSTRACT The role of internal migration in reducing regional inequalities is a common feature of classical economic theory and urban economics models. If regional migration is important in reducing spatial disparities, then understanding its causes, and barriers, is crucial. This paper explores the drivers of regional migration behaviour in Great Britain. Findings point to rigidities in housing that deter mobility across regions; and regional differences in the drivers and effects of regional migration. Our paper supports greater focus on spatial disaggregation, since migration studies conducted at the national level ignore important spatial differences in migration behaviour.


INTRODUCTION
Great Britain is one of the most inter-regionally unbalanced countries in the industrialised world, with longstanding inter-regional imbalances in key economic areas such as productivity, incomes, employment and wellbeing.As yet, the reasons why these inter-regional inequalities have been so large and persistent remain unclear (McCann, 2016).These deep-seated regional economic problems are not unique to Great Britain.Territorial imbalance-related problems are evident in emerging countries as well as much of Europe, and this rise in territorial polarisation has manifested itself in a 'geography of discontent', in which the fear of being left behind, and having no future, has led to serious political, social and economic consequences (Rodríquez-Pose, 2018).Most notably, the places that 'don't matter' are increasingly using the ballot box to rebel against this feeling of being left behind, and of lacking opportunities and future prospects (for example, the recent rise in populism in the Western world as evidenced by the 2016 UK Brexit vote, and the elections in the USA, Austria, France, Germany and Italy since 2016).
The long-standing regional inequalities in Great Britain fit into the policy debates occurring at the international level regarding the relative importance and efficacy of placebased versus space-blind development policy.The 2009 World Bank Development Report argued that development policy should be 'space-blind', in that national economic policies should be designed without explicit consideration to space.This space-blind argument presupposes that people and investment should be allowed to move freely according to market forces to where they are most productively employed, and any spatially targeted policies would simply result in restricting inter-regional migration, and thus the overall economic efficiency of the national economy (Gill, 2010).However, more recent evidence has argued for a 'place-based' approach to economic development, where policy has an explicit geographical logic to it and is aimed at overcoming the under-development traps in certain regions (McCann, 2016).
A key indicator of the importance and effectiveness of space-blind versus place-based policies is the extent of labour mobility.The role of migration in alleviating economic imbalances has long been debated within the economics literature.In classical economic equilibrium models, labour and capital respond to inter-regional economic disparities through undertaking inter-regional migration and, in doing so, act to reduce these disparities.Classical economic theory would therefore predict regional wage convergence over time as a result of the mobility of labour across geographical areas.
However, the persistence of wage differentials across regions in many Western economies points to the reality that transition processes in labour markets, especially regional wage convergence, may be slow and not a forgone conclusion.Whether regional migration acts as a labour market adjustment mechanism is not an inconsequential issue.If regional migration does arbitrage regional wage disparities then, from a policy perspective, measures aimed at increasing regional migration are appropriate.Thus, people-based policies that favour inter-regional mobility (as the best way to promote economic development) may be more effective than place-based policies that emphasise economic restructuring and redistributive efforts in order to reduce regional wage gaps.
This paper speaks into the ongoing debate on the important role played by regional migration.If regional migration is a primary mechanism in facilitating regional convergence, then identifying and understanding the drivers of regional migration is crucial for the design of policy that seeks to promote labour mobility as a device for equalising regional economic differences.Specifically, this paper explores two key aspects of regional migration: first, what are the drivers of individual migration behaviour (and are there regional differences in the determinants of individual migration); and second, is regional migration wageenhancing as predicted by economic theory (and are there regional differences in the returns to migration)?These questions are important for the design of migration policy: if the drivers of inter-regional migration are regional (i.e., spatial in nature), then migration strategies set at the national level may not be appropriate compared to devolved, or spatially disaggregated migration policies.
To explore these questions, we purposely identify a period of stable economic growth in the British economy and use longitudinal data to estimate both the determinants of between-region migration and the returns to regional migration for six broad regions of Great Britain.In the econometric analysis, we explicitly account for the possibility of selectivity bias in the estimates of the returns to migration.
The paper is organised as follows.Section 2 provides a brief overview of the theoretical and empirical literature on individual migration decisions.Sections 3 and 4 outline the data and methodology used.Section 5 discusses the results, and Section 6 concludes.

THE MIGRATION LITERATURE ON INDIVIDUAL MIGRATION BEHAVIOUR
2.1.Theoretical approaches One of the greatest challenges in 'evaluating research on migration theory lies in the diverse and fragmented nature' of the literature, which derives from various disciplines, and which is 'rarely bound together into an integrated treatment' (Molho, 1986, p. 396).Within economics, there is a rich variety of theoretical perspectives on migration that focus on individual migration behaviour, as well as models of aggregate behaviour.In this paper, we focus primarily on individual migration behaviour.
Early economic literature on migration stressed the importance of wages as the main cause of migration.
Differences in wages across geographical areas would lead to a process of migration out of low-wage regions and into high-wage regions, resulting in the elimination of regional wage differentials.Models of urban/rural shift in developing countries added to the classical approach by linking migration to both wages and the probability of being able to find employment (e.g., Harris & Todaro, 1970).These early approaches to migration assume rationality in decision-making, as individuals choose to migrate based on the earnings-enhancing benefits of migration.
Extending this view of migration for the purposes of personal enhancement, the human capital approach suggests that migration can be modelled as an individual investment decision (Sjaastad, 1962).Potential migrants base their migration decision on an assessment of the anticipated future stream of benefits and costs that result from migrating.The individual evaluates the expected benefits, less the movement costs for each area, and selects to live in the area with the highest net outcome.One of the strengths of the human capital model is that it allows for a variety of factors to influence migration decisions, including economic, social and environmental factors, and consequently, departs from the traditional emphasis on wages as the primary determinant of migration.The human capital approach also explicitly incorporates a spatial dimension and temporal factors as individuals evaluate the benefits and costs of migration over geographical space and time.The main limitation, however, is the assumption that individuals have information on all available opportunities in all regions.
Both the traditional approach and the human capital model suggest that all migrants will move in the same direction: in the classical approach from low-wage to high-wage regions, while the human capital model predicts broadly unidirectional moves for any identifiable homogeneous population group (Molho, 1986).This, however, does not explain the heterogeneity in observed migration behaviour.Random utility models capture the reality that not all migrants move in the same direction.By partitioning an individual's utility function into two components, one that reflects the behaviour of a representative individual and another which reflects the unobserved individual characteristics and situational factors that cause the individual's behaviour to deviate from the representative individual's behaviour, the random utility model incorporates the characteristics of individuals that systematically influence migration.
Search theory, building upon the limitations of the human capital model, concentrates on the underlying search strategy in the decision to migrate.This literature provides an important distinction between speculative migration and contracted migration: speculative migration is undertaken in the hope of finding a suitable opportunity in the destination area, whereas contracted migration is undertaken after an opportunity has already been secured.Speculative migration is therefore an integral part of the search process, whereas contracted migration is the outcome of the search process.Contracted migration is the more likely form of migration for most migrants as speculative migration would involve sacrificing current wages for the uncertain prospect of getting a better job after an indefinite period of search activity.An individual's decision to engage in speculative migration may be modelled in a standard risk theoretical framework, involving a comparison of the expected relative payoffs to search in different areas, and individual decisions are likely to vary according to current income and employment status, as well as the perceived variance of opportunities in the destination area.Migration is therefore viewed as the outcome of receiving and evaluating opportunities as they arise.In this view, individuals who have already migrated once as part of the search process may engage in further migration as other opportunities arise (Molho, 1986).
In response to criticism of the neoclassical theory, mostly based on observations of sluggish migration, the option theory of migration (Burda, 1995) posits that individuals may not migrate immediately in response to wage differentials due to uncertainty over future wage levels and this may hinder migration.'O'Connell (1997) extends this analysis to show how uncertainty surrounding the evolution of future conditions in both the origin and destination regions may deter migration and how it is optimal for the potential migrant to postpone migration until some or all of the uncertainty can be removed' (Drinkwater, 2003, p. 8).
The theory of family migration originates from Mincer (1978), who extended the human capital model of migration to families.The migration decision in this context entails the family weighing up the present value of expected changes in future family income, net of discounted migration costs.A positive net family gain will result in migration occurring, even though the net gains may differ between spouses (Nivalainen, 2004).While the human capital model of family migration is gender neutral, in the household context, moves often occur to fulfil the career aspirations of the main breadwinner at the expense of the partner and so household migration is typically viewed as improving the career prospects of the husband with less regard to those of the wife (Taylor, 2007).

Empirical analyses
The empirical testing of migration behaviour at the microeconomic level faces several problems, most notably the inability to capture, at the individual level, the full range of characteristics and factors that explain the complexity of individuals' migration decisions.Consequently, numerous empirical studies restrict their scope to analysing the movement/non-movement decision (Molho, 1986).The decision to migrate is modelled empirically using the discrete choice framework to identify those characteristics that are important in an individual's decision to migrate.
With respect to personal characteristics, empirical studies typically find that the probability of migration is higher for young people (Bover & Arellano, 2002, for Spain;Greenwood & Hunt, 2003, for various countries), and for men (Long, 1973;Mincer, 1978, both for the USA).The presence of children in the household has a negative impact on the probability of migration, especially school-age children.Young couples tend to have higher probabilities of moving, and as the duration of marriage increases, the probability of migrating decreases.The employment status of wives is important, with a lower migration probability when the wife is employed (Mincer, 1978).With respect to human capital, educated individuals are more likely to move (Bover & Arellano, 2002, for Spain;Faggian & McCann, 2009, for Great Britain).There is also a strong negative relationship between the length of time an individual has spent in an area and their willingness to move (Drinkwater, 2003, in the context of Central and Eastern European migrants).
For tied migration, two predominant findings emerge from the empirical literature.First, families migrate in response to economic motivations on the part of the husband; and second, that wives' employment considerations are of minor importance in migration decisions (see Nivalainen, 2004).Family migration has large negative effects on the labour market outcomes of women in terms of labour market participation, employment, working time and wages (Taylor, 2007).
This paper explores individuals' decisions to migrate at the regional level in Great Britain.Much of the empirical literature focuses on the national level, but there is good reason to suspect that the determinants of migration will vary spatially across a country such as Great Britain, which contains regions with distinct regional economies, industrial structures and labour market characteristics.Specifically, the paper explores; (i) whether there are regional differences in the determinants of migration, and (ii) whether the returns to migration differ across regions.These research questions have direct policy relevance in regard to whether regional policies should be people-based or place-based to reduce spatial disparities, and whether migration policy should be set at the national level or at a regional/state/city level.
These questions are explored using longitudinal data over the years 1991-2007.Despite this period not being up-to-date, we have purposely chosen these years as they represent a prolonged period of sustained economic growth within Great Britain with few, if any, shocks to the national economy.We restrict our attention to the period prior to the Great Recession, as this downturn had a pronounced effect on migration in the UK (in 2008, the UK entered its worst recession since the Second World War).Furthermore, in 2016, the pattern of migration in the UK experienced another shock when the UK voted to leave the EU.By restricting our attention to the years prior to these structural breaks in the UK economy, we aim to investigate the determinants and effects of regional migration during a period of economic stability and no major shocks to migration behaviour or patterns.This should provide a cleaner and more accurate picture of the drivers and outcomes of regional migration without the distorting effects of major economic shocks.
The study period covers nearly two decades of Britain's recent economic history.The early 1990s were a recessionary period that had a greater adverse effect on the southern regions.This was followed by a period of GDP growth in the mid-1990s, fuelled by export growth and recovering consumer expenditure, and then sustained growth from the late 1990s to the late 2000s.
The decades preceding the 1990s were characterised by deindustrialisation and its impact on the distinctive industrial geography of Britain.Manufacturing employment fell sharply in the 1980s, and the subsequent shift from industrial to service sector employment had a greater negative impact on the northern regions (historically characterised by narrow industrial bases and an over-dependence on declining heavy industries) compared to the southern regions that are more concentrated in service industries.In particular, London emerged as a global financial centre with a concentration of jobs in banking, finance and business services.

DATA AND DESCRIPTIVE STATISTICS
This paper uses data from the British Household Panel Survey (BHPS), which provides a nationally representative sample of the population of the UK living in private households.It is an annual survey of each adult member (aged 16 years and over) of a nationally representative sample of more than 5000 households, amounting to a total of approximately 10,000 individual interviews.
An advantage of the BHPS is that it covers a wide range of socio-economic characteristics of respondents and is comparatively rich in terms of information that may help to explain both the earnings of individuals and their migration decisions.The sample consists of fulltime male and female employees aged between 16 and 65 years, and the earnings variable is specified as the log of real hourly earnings (deflated to 1991 prices using the Retail Price Index).The analysis focuses on six broad regions of Great Britain: Greater London, the Rest of the South, the Midlands, the North, Wales and Scotland (this is the same regional classification and dataset used in Dickey, 2014).This broad regional classification is based on the recognition that the standard UK regions are largely administrative in nature, and don't necessarily map well onto regions in terms of economic coherence.It also reflects the wider spatial effects that operate across broader regions, e.g., the operation of broader regional labour markets, the historical impact of industrial structure, cultural effects and geography (Bishop & Gripaios, 2010).To establish whether this regional classification is a sensible way of breaking down the national economy, Kruskal H Wallis tests of independent samples were conducted, and these indicate that the six broad regional wage distributions are significantly different from each other and should therefore be treated separately.
Table 1 provides summary statistics for the whole sample, and for migrants and non-migrants separately. 1  The objective of Table 1 is to explore whether there are differences between migrants and non-migrants with respect to their personal, household and job characteristics (t-tests of differences in means are reported in column 7).This comparison is a first step in describing the possible nature of selection of individuals into migration.On average, migrants are younger, less likely to be married, have a smaller household size and a lower prevalence of home ownership.Migrants also tend to have higher levels of formal education; e.g., 3.6% of migrants have no qualifications compared to 12.5% for non-migrants, whereas 34.5% have university qualifications relative to 19% for non-migrants.
The influence of these characteristics on migration can be interpreted in the context of the potential benefits and costs of migration as predicted by the migration literature.The longer time horizon and the likely lower costs of migration due to lower community and family attachments may explain why younger workers are more likely to migrate.Household size and housing tenure can also be considered as proxies for the costs of migration, with both larger household size and home ownership expected to increase the pecuniary and non-pecuniary costs of migration.Overall, the descriptive statistics support the predictions of theory and previous empirical evidence, and reveal that migrants differ from non-migrants along numerous observable dimensions.
Table 2 shows average hourly real wages by region for selected years across the time period.Average hourly real wages increased by 28% for Great Britain as a whole.At the regional level, average hourly real wages have been consistently higher in Greater London and the Rest of the South than at the national level, and consistently lower in Wales and the North.Table 3 reports the proportion of individuals in each education group by region.Consistent across this period, Greater London has the largest proportion of individuals in the highest education group.Over the period there has been an upgrading of education levels in all regions.
Table 4 shows the migration flows between the six broad regions over the period.These are reported in relative terms, e.g., 77.6% of those who migrated out of Greater London moved to the Rest of the South.There is a clear pattern that most of the internal migration occurring across Great Britain is to neighbouring regions, with those in northern regions migrating to other northern regions, and those migrating from southern regions moving to other southern regions.The prevalence of moves to neighbouring regions is in line with predictions from migration theories where distance between origin and destination regions matter since migration costs are lower for shorter-distance migrations.Table 4 also highlights the role of the southern regions, in particular the Rest of the South, in attracting migrants from the other regions.

Model and estimation strategy
The primary objective is to estimate the determinants of the decision to migrate between the regions of Great Britain.These regional equations aim to explore how personal, household and job characteristics affect an individual's decision to migrate, and whether there are regional differences in the determinants of migration.
The probability of migrating between regions is estimated using a standard probit model, where the dependent variable is equal to 1 if the individual migrated between regions (migrant), and 0 if the individual did not migrate (stayer).The regional probit models are estimated using the pooled data across the period 1991-2007 and estimated with clustered standard errors.Estimating the migration equations separately for each region does not constrain the parameter estimates to be equal, allowing the rates of return for each characteristic to vary by region.This is a more flexible specification that provides valuable regionspecific information on the determinants of migration.
We further estimate the regional migration models separately by gender since the literature on tied migration suggests that migration behaviour may differ between men and women (if migration within the family occurs to improve the career of the husband/male partner).We would hypothesise that human capital and labour market characteristics are more important for male migrants than female migrants.
The second objective of the empirical analysis is to empirically test the theoretical prediction that individuals migrate due to the anticipated returns from migration, and whether there are regional differences in this return to migration.To explore this, regional wage equations are specified where the dependent variable is the log of hourly real wages.These regional equations are estimated by ordinary least squares (OLS) using pooled data across the period.The aim is to: (i) empirically test for evidence of the rationality assumption that individuals decide to migrate based on anticipated earnings-enhancing benefits of migration; and (ii) to identify whether the returns to regional migration differ at the regional level.
The third objective of the econometric analysis is to investigate the potential presence of self-selection in regional migration decisions.Both the theoretical and empirical literatures suggest that regional migration will be non-random (Greenwood, 1975;Pekkala, 2002), and the summary statistics provided in Table 1 suggest that there are systematic differences between migrants and non-migrants.The aim is to empirically test for the presence of self-selection in migration decisions in Great Britain.For instance, we might presume that selection into migration is present and similar in all regions; however, it may be the case that there are differences in whether, and how, individuals select into migration at the regional level.Despite the prevalence of self-selection in the migration literature, there has been very little attention given to empirically testing whether there are regional differences in selection into migration.
The self-selection problem is typically resolved by adopting the joint decision/outcome model of Heckman (1976) and Lee (1978).In the first stage of the Heckman model, the probability of migration is estimated using a reduced probit model.In the second stage, the wage equation is estimated using linear regression including the selectivity term from the first stage as an explanatory variable in order to adjust for self-selection.The coefficient on the inverse Mills' ratio represents the covariance between the errors in the migration and wage equations and is used to test for the presence of selection bias.In particular, the coefficients on the inverse Mills' ratio provide information on whether there is positive or negative selection bias.A typical definition of positive selection is that, on average, migrants are more skilled than stayers, and therefore the expected earnings of migrants exceed the expected earnings of stayers (Borjas et al., 2019).
The issue of self-selection into migration is not a new concern.However, a recurring theme in the empirical migration literature is the lack of natural experiments  Exploring the drivers of internal labour migration for the regions of Great Britain 1395

REGIONAL STUDIES
and the scarcity of good instruments to deal with the identification problem.As Kaestner and Malamud (2014) point out, addressing the selection problem in a compelling manner is extremely challenging.In our analysis, we replicate three approaches from the empirical literature to draw tentative conclusions about the nature of self-selection in regional migration in Great Britain.First, we follow Kaestner and Malamud (2014) by estimating the first stage probit model including the main human capital variables and gender, along with the number of children which is excluded from the second stage model.The two-stage model is estimated at the national level with regional dummies included.Second, we specify the two-stage estimation separately for movers and stayers (following Agesa, 2001;Brücker & Trübswetter, 2007;Robinson & Tomes, 1982).Third, we use the predicted values from the first stage probit model as an instrument for migration status in the second stage wage equations (e.g., Axelsson & Westerlund, 1998).
The migration and wage models include a range of personal, household and job characteristics that reflect the theoretical and empirical literatures on both the decision to migrate and the determinants of individual-level wages.Education is classified into four categories: (i) no education qualifications; (ii) low levels of formal education (GCE levels or equivalent, Commercial qualification, CSE grade, O levels, Apprenticeship); (iii) semi educated (Other higher qualifications, Nursing qualifications, GCE A levels); and (iv) high levels of formal education (Higher degree, First Degree, Teaching Qualifications).Age is a proxy for labour market experience and is commonly found to play an important role in both migration decisions and earnings determination.To allow for the expected non-linear effect of age on migration and earnings, the age variable is classified into three groups: (i) 16-26 years; (ii) 27-36 years; (iii) 37-65 years of age.
In the UK, the nature of housing tenure has been found to reduce spatial mobility.Housing tenure is incorporated using three categories: (i) house ownership; (ii) public housing (property owned by a government authority); and (iii) private rental housing.Household size, defined as the number of people living in the household, is included due to its influence on family migration decisions (through its effect on migration costs).To investigate whether employment in the public sector reduces mobility relative to private sector employment, a dummy variable denoting whether an individual is employed in the public or private sector is included.A dummy variable denoting employment status (full-time or part-time) is also included.The skills variable is classified into three categories: (i) highly skilled; (ii) semi-skilled; and (iii) low-skilled, and is derived from the Standard Occupational Classification (SOC) Code.

The migration decision
The first objective of the empirical analysis is to model the determinants of regional migration (results shown in Table 5).In line with human capital theory, age has a negative effect on migration, with the most mobile group being those aged 16-26 years (those mainly moving for study or first jobs in the labour market).Those in early career stage (mid-20s to mid-30s) are also more mobile than those who are mid-and late-career (late-30s onwards).This is particularly the case in the southern regions (Greater London and the Rest of the South), where the probability of migrating is much lower for older workers.Interestingly, the probability of migrating does not differ across the two younger age groups in Scotland, Wales and the Midlands.For Scotland and Wales, it may be the case that those aged 16-26 years stay in their home region (nation) to attend higher/further education and then remain there to continue their careers.Marital status and gender appear to be less important in regional migration decisions.For education, those with higher education are more likely to migrate in the southern regions, the Midlands and Wales.
Job characteristics have a limited effect on individual migration decisions.Sector of employment does not influence the decision to migrate and employment status is only influential in the North.Being in a highly skilled occupation, compared to a lower-skilled one, positively affects regional migration in the Midlands and the northern regions, but skill level is less relevant to individuals' migration decisions in the two southern regions.Perhaps this reflects differences in the flexibility of labour markets across the regions, with greater mobility across all skill levels and occupations in the labour markets of the southern regions.
Housing tenure has an important influence on regional migration.Those who rent public housing are less likely to migrate, especially in Scotland, while those who live in privately rented accommodation are more likely to undertake regional migration in all regions.
The next step is to explore whether the drivers of regional migration differ between men and women (Tables 6 and 7).The negative effect of age on migration is greater for women than for men.One interesting difference between men and women is the effect of education on migration: education level does not appear to influence women's migration decisions, whereas the expected relationship is observed for men (especially in the southern regions).This may reflect the greater prevalence of tied migration amongst women, where migration is more likely a result of the husband's economic motivations, and therefore migration is more dependent on the husband's human capital characteristics.Another important gender difference is the role of housing tenure: in particular, renting public housing hinders regional migration for women but not for men.
Table 8 repeats the migration model with the inclusion of interaction terms between age and education, which provides a test for the hypothesis that the age effect on migration has a differential effect depending on education level.For instance, if younger individuals migrate to other regions for the purposes of higher education or early career jobs, then younger, Heather Dickey and Maire Carroline Magante REGIONAL STUDIES more educated individuals may have higher probabilities of migrating compared to younger, less educated individuals.For the Rest of the South, the North, and Great Britain as a whole, the age effect on migration does not depend on education level.For Greater London, the Midlands and Scotland, the results do suggest that the probability of migrating does increase for young individuals as their education level rises, i.e., younger, more educated individuals are more likely to migrate to these regions than younger, less educated individuals.In contrast, older workers' probabilities of migrating decreases with higher education levels.
To investigate whether the determinants of migration exhibit regional differences, we apply t-tests for significant differences, and these allow us to reject the hypotheses of no significant differences across the regional coefficients (particularly for housing tenure, household size, and the education and skills variables).This finding has important consequences for analyses of migration behaviour.It suggests that geographical variation should be accounted for when modelling migration decisions.This has important implications for policies designed to promote migration as a means of reducing regional economic inequalities.

The effect of regional migration on wages
The secondary objective of the empirical analysis is to estimate the effect of regional migration on individual wages (Tables 9, 10 and 11).Exploring the drivers of internal labour migration for the regions of Great Britain 1397

REGIONAL STUDIES
Table 9 demonstrates the impact of personal characteristics on individual wage levels.Age, as a proxy for labour market experience, exhibits the expected effect on wages in all regions, with the returns to age being highest in Greater London and lowest in Wales, suggesting interesting differences in regional age-earnings profiles.Education has the expected positive coefficients with higher levels of education being associated with higher wages.The results suggest interesting differences in the returns to education.For example, the premium for having higher levels of education is relatively low in Greater London but is much higher in Wales.
Similarly, higher skill levels have the expected positive impact on individual wages.Low-skilled workers are more disadvantaged in the Rest of the South, whereas both low-and semi-skilled workers are less disadvantaged in Greater London.With respect to job characteristics, there is a wage premium for being employed in the public sector in all regions except for Greater London.This wage benefit from public sector employment is highest in the Midlands and the North.There is a positive association between wages and home ownership.The wage difference between home owners and public renters is largest in the devolved regions of Wales and Scotland, whereas the wage difference between home owners and private renters is larger in the two southern regions.
Estimation of the regional wage equations provides an opportunity to empirically test whether there is a positive return to regional migration, as predicted by numerous migration theories.However, we find mixed results.Migration has a positive influence on wages in only three regions (the North, Wales and Scotland), whereas it has a negative wage impact in Greater London.This counterintuitive result for Greater London may be explained by the fact that, while Greater London does 1398 Heather Dickey and Maire Carroline Magante REGIONAL STUDIES consistently have higher average wages than the rest of the country, it also has the highest level of wage inequality.The higher level of inequality in the wage distribution may serve as a negative signal that acts as a deterrent to prospective migrants from other regions.Overall, there is limited empirical support for the wage-enhancing benefits of regional migration.
Tables 10 and 11 repeat the regional wage equations for men and women separately.For women, there is less dispersion in the returns to age across regions, suggesting more similar regional age-earnings profiles for women compared to men.In contrast, the wage differential by education is greater for women in most regions.One explanation is that the wage premium for human capital acquired through formal education is higher for women, whereas the wage premium for human capital acquired through labour market experience is greater for men.
Migration continues to have little impact on individual wages for men and women.

Self-selection in the migration decision
The final objective is to investigate the presence of selfselection in individuals' migration decisions, and whether there are systematic differences in how individuals select into migration across the British regions.There has been very little focus on potential geographical differences in terms of whether and how individuals select into migration.
Table 12 displays the results for the first approach (using the number of children as the exclusion restriction).For the first stage (column 1), we find the expected results for the main personal and human capital characteristics.The second stage is reported in columns 2-4.In columns 3 and 4, the migration variable is negative (significant at Exploring the drivers of internal labour migration for the regions of Great Britain 1399 the 1% level), suggesting a wage penalty for undertaking regional migration.The selectivity term is negative (and significant at the 1% level), indicating the presence of negative selection in regional migration in Great Britain.
Interacting the regional dummies with the selection term further enables the effect of selection bias to differ regionally.The interaction terms are significant for the Midlands, the North and Scotland, implying that the direction and scale of self-selection varies for these regions relative to Greater London.A comparison with columns 5 and 6 allows us to see how the estimates of the wage equations are affected by the inclusion of the selectivity term.In general, the coefficients are qualitatively and quantitatively similar, but the effect of regional migration on individual wages is bigger in the OLS estimation, suggesting the returns to migration are overestimated when selection bias is not controlled for.
The second approach to selection bias is to re-estimate the two-stage estimation process for migrants (movers) and non-migrants (stayers) separately.
Table 13 displays the results of the first stage probit model (column 1) and the wage equations by migration status (columns 2 and 3).For non-migrants, the selectivity term is positive, implying that stayers are more skilled than movers, and the expected earnings of stayers exceed the expected earnings of movers.Testing for equality in the coefficients across the two wage equations (column 4) further highlights that some of the explanatory variables differentially affect wages, reinforcing the hypothesis that there are some systematic differences between movers and stayers.
The third approach comprises instrumenting the migration variable with the predicted values from the    2) and ( 3) are estimated with the number of children as the exclusion restriction for identification purposes.
Exploring the drivers of internal labour migration for the regions of Great Britain 1405

REGIONAL STUDIES
first stage migration model.The results in Table 14 reveal that there is a positive return to migrating to Greater London, but a substantial wage penalty attached to migrating to Scotland, followed by the North, and no returns to migrating to Wales.The Hausman F test reports that the null hypothesis of exogeneity can be rejected for all regions, confirming that the two-stage estimation is appropriate.Taking these results altogether, we find little support for the wage-enhancing motive for regional migration.Given the positive selection of stayers, and the negative returns to migration in most regions, these findings are a tentative challenge to the traditional models of migration and their predictions that regional labour mobility acts as an equalising mechanism to reduce regional disparities.

CONCLUSION
Much of the theoretical literature on migration focuses on the important role that regional migration has in mitigating regional economic disparities.If shifts in the geographical distribution of labour supply are an important device in reducing regional differentials, then understanding the causes, and barriers to, regional migration is crucially important.This paper has explored the drivers of individual migration behaviour between the regions of Great Britain during a period of economic stability and growth.Several interesting findings emerge from the empirical analysis.First, we find that education appears to have a lesser influence on migration than skill level, suggesting that skill level may be a more significant determinant of migration.If education is interpreted as an indicator of general human capital (including general transferable skills), then occupation-and firm-specific human capital may be a more important predictor of migration decisions and potential success compared to broader education levels.Thus, potential migrants may give more weight to the assessment of their skills rather than their education level in their migration decisions (Biswas et al., 2009).Another potential explanation is that unobserved characteristics are more important than observed characteristics in determining migration (Borjas et al., 2019), and that an individual's skill level is a better proxy for unobserved abilities than formal education.As long as unobserved characteristics are sufficiently transferable across regions, then migrants are positively selected when the rate of return to unobservable skills is higher in the destination region compared to the home region.Thus, migration may be more correlated to skill level than education attainment.
Second, our analysis demonstrates the importance of housing tenure in inter-regional migration.In all regions, home ownership and public housing reduce mobility relative to private renting.The nature of housing tenure has long been blamed for discouraging spatial mobility in the UK, most notably public renting in the 1980s and home ownership in the 1990s.'The relative immobility of public renters may stem from public housing rents being below market rates, the restricted transferability within public housing, high waiting lists, and the security of tenure' (Battu et al., 2008, p. 311).Hughes and McCormick (2000) explicitly link the relatively low level of labour mobility in the UK with the relatively unusual structure of housing tenure, with its small private rental sector, declining social rented sector, and prevalence of home owneroccupancy.
Our findings, therefore, point to the important effect of institutional rigidities in regional housing markets that act to lower mobility.For instance, policies that aim to encourage home ownership or increased public housing may act to constrain labour mobility across regions.This highlights the importance of recognising the trade-offs faced by policymakers.While calls to extend public housing provision in the UK have been increasing in recent years, policies that aim to increase the public housing stock in the UK will have the unintended effect of reducing inter-regional labour mobility and prolonging persistent spatial economic inequalities.Government policies that favour owner-occupancy will have similar effects.As argued by Cameron and Muellbauer (1998), housing concerns remain a priority in the context of inter-regional migration in Great Britain.
Third, the paper confirms that the determinants of both migration and wages vary at the regional level in Great Britain and offers tentative evidence that selection into migration also exhibits regional differences.Much of the empirical literature lends support to the hypothesis of self-selection in migration decisions, but we add to this by pointing to the possibility of spatial differences in how individuals select into inter-regional migration, which has implications for the effectiveness of migration policies set at the national level versus those set at a more disaggregated level.
Fourth, our paper speaks into the international debate around the role and nature of development policies for addressing long-standing spatial inequalities.To obtain efficient labour markets, workers should be able to locate where the relevant jobs are found, and thus inter-regional migration is an essential element for obtaining wellfunctioning labour markets (Haas & Osland, 2014).Both urban economics and classical migration theory emphasise the role of individual mobility and perfect or costless migration is a general assumption in urban economic models.However, these assumptions stand in stark contrast to the realities of migration, where real barriers to mobility include rigidities in the housing market, deep emotional attachments to place, lack of skills and education to migrate to the urban centres and legal restrictions that prevent mobility within countries (Rodríquez-Pose, 2018).
The rise in populism driven by the 'left-behind' places presents an opportunity to openly debate and address the problems of spatial inequalities.So far, the traditional space-blind approach of relying on labour mobility, and the willingness of individuals to move to places where there are opportunities (as purported by the World Bank, 2009) has not succeeded, and is unlikely to in the future given the declining trend in inter-regional migration in many countries.While the UK has predominantly focused on people-based policies, the sheer persistence of deep-seated regional imbalances (and the historical inability of inter-regional migration to remove these spatial inequalities) points to the limitations of the space-blind approach to address long-standing spatial imbalances in key economic outcomes.It may be the case that an innovative regional strategy that combines both place-based policies and people-based policies is the most appropriate solution for redressing deep and long-standing geographical divides.
This paper has added to the empirical literature on regional migration by focusing on the drivers and effects of regional migration in Great Britain.However, there are several limitations to the study.First, the time period has been purposely chosen to study regional migration in the absence of major economic shocks to the UK economy and their impacts on migration behaviour.Consequently, the analysis is not up-to-date.However, the determinants and effects of migration behaviour during economic shocks are less informative to policymakers who seek to design policies that will encourage regional labour mobility Exploring the drivers of internal labour migration for the regions of Great Britain 1407

Table 2 .
Average hourly real wages by region.
Source: British Household Panel Survey.Unit of currency is pound sterling (£).

Table 1 .
Descriptive statistics by migration status.

Table 3 .
Education levels by region.
Source: British Household Panel Survey.
Source: British Household Panel Survey.

Table 5 .
The regional migration decision.

Table 6 .
The regional migration decision for men.

Table 7 .
The regional migration decision for women.

Table 8 .
The regional migration decision: age and education interactions.Greater London Rest of the South Midlands The North Wales Scotland Great Britain Notes: Clustered standard errors in parentheses.***p< 0.01, ** p < 0.05, * p < 0.1.1400HeatherDickey and Maire Carroline Magante REGIONAL STUDIES

Table 9 .
Regional wage equations: OLS.Exploring the drivers of internal labour migration for the regions of Great Britain 1401

Table 10 .
Regional wage equations for men.

Table 11 .
Regional wage equations for women.

Table 12 .
The Heckman selection model.Exploring the drivers of internal labour migration for the regions of Great Britain 1403

Table 13 .
Wages equations for migrants and non-migrants.

Table 14 .
IV estimation of regional wage equations.