Regional Labour Market Adjustment and the Movements of People: A Review

Maré, David C; Choy, Wai Kin Working Paper Regional Labour Market Adjustment and the Movements of People: A Review New Zealand Treasury Working Paper, No. 01/08 Provided in Cooperation with: The Treasury, New Zealand Government Suggested Citation: Maré, David C; Choy, Wai Kin (2001) : Regional Labour Market Adjustment and the Movements of People: A Review, New Zealand Treasury Working Paper, No. 01/08, New Zealand Government, The Treasury, Wellington This Version is available at: https://hdl.handle.net/10419/205449 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ ACKNOWLEDGEMENT: The authors would like to thank Jason Timmins for his assistance in preparing the data for analysis. Comments from Benedikte Jensen, Jacques Poot, Philip Morrison, Richard Bedford, James Newell, Simon Chapple, Paul Callister, Kerry Papps and other participants at the seminars at the Treasury and the Labour, Employment and Work (LEW) conference were also gratefully received. DISCLAIMER: The views expressed are those of the authors and do not necessarily reflect the views of the New Zealand Treasury. The Treasury takes no responsibility for any errors or omissions in, or for the correctness of, the information contained in this paper. TREASURY WORKING PAPER 01/08 Regional Labour Market Adjustment and the Movements of People: A Review David C. Maré 1 and Wai Kin Choy 2 Abstract : This review paper examines the link between internal migration and regional labour market adjustment. It outlines the motivation and scope of our enquiry, discusses the three key questions that we plan to pursue, reviews relevant international and New Zealand literature, and outlines proposals for future research. The first key question examines whether migration helps regional labour market adjustment. The second question investigates how important migration is as a regional labour market adjustment mechanism. The final question looks at who is moving and whether it matters for regional labour market adjustment. JEL Classification: R23 - Regional Migration; Regional Labour Markets; Population; J61 - Geographic Labour Mobility; Immigrant Workers Keywords: regional labour market, internal migration, regional labour market adjustment. 1 Motu Economic and Public Policy Research Trust, Email: [email protected] 2 Policy Co-ordination and Development, New Zealand Treasury, Email: [email protected] 2 TABLE OF CONTENTS Abstract: .....................................................................................................................................1 Executive Summary ....................................................................................................................4 List of Figures, Tables and Boxes ..............................................................................................6 1 Introduction.........................................................................................................................7 1.1 Introduction ........................................................................................................7 1.2 Motivation ...........................................................................................................7 1.2.1 Policy environment ..........................................................................................7 1.2.2 Overview of Literature ...................................................................................10 2 Structuring Our Thoughts...............................................................................................14 2.1 Scope ................................................................................................................14 2.2 Key Questions ...................................................................................................15 2.3 Theory and Concepts .........................................................................................16 2.4 Empirical Issues and Data Sources .....................................................................20 3 New Zealand Patterns of Internal Migration and of Regional Labour Market Adjustment.................................................................................................................................27 3.1 Internal Migration .............................................................................................27 3.2 Regional Labour Market Adjustment ...................................................................36 4 Key Question 1: Does Migration Help Regional Labour Market Adjustment?.........44 4.1 What? ...............................................................................................................44 4.1.1 As revealed by actual mobility patterns ...........................................................44 4.1.2 As revealed by people when asked about the connection ..................................46 4.2 How? ................................................................................................................47 4.2.1 As revealed by actual mobility patterns ...........................................................47 4.2.2 As revealed by people when asked about the connection ..................................64 4.3 Findings ............................................................................................................66 4.3.1 As revealed by actual mobility patterns ...........................................................66 4.3.2 As revealed by people when asked about the connection ..................................75 4.4 Concluding Remarks ..........................................................................................76 5 Key Question 2: How Important is Migration as a Regional Labour Market Adjustment Mechanism?..........................................................................................................78 5.1 What? ...............................................................................................................78 5.2 How? ................................................................................................................78 5.2.1 Short-time series approaches ..........................................................................79 5.2.2 Longer time series approaches ........................................................................80 3 5.3 Findings ............................................................................................................90 5.3.1 VAR approach ...............................................................................................90 5.3.2 Case studies approach ....................................................................................94 5.3.3 Structural modelling approach ........................................................................95 5.3.4 Other approaches ...........................................................................................95 5.4 Concluding Remarks ..........................................................................................96 6 Key Question 3: Who Moves and Does It Matter for Regional Labour Market Adjustment?...............................................................................................................................97 6.1 What? ...............................................................................................................97 6.1.1 Identifying heterogeneity ................................................................................97 6.1.2 Implications of heterogeneity on regional labour market adjustment ..................98 6.2 How? ................................................................................................................99 6.2.1 Identifying heterogeneity ................................................................................99 6.2.2 Implications of heterogeneity on regional labour market adjustment ................103 6.3 Findings ..........................................................................................................103 6.3.1 Identifying heterogeneity ..............................................................................104 6.3.2 Implications of heterogeneity on regional labour market adjustment ................110 6.4 Concluding remarks .........................................................................................111 7 Research Options............................................................................................................112 7.1 Introduction ....................................................................................................112 7.2 Choosing which of the options to proceed with ..................................................112 7.3 Option 1. Identifying the link between migration flows and regional labour market adjustment.....................................................................................................114 7.3.1 Introduction .................................................................................................114 7.3.2 Methods ......................................................................................................114 7.3.3 Specification issues to be resolved ................................................................116 7.3.4 Data ............................................................................................................118 7.3.5 Resources & Timing ....................................................................................119 7.4 Option 2. Identify relative strength of migration as a regional labour market adjustment mechanism..............................................................................................120 7.4.1 Introduction .................................................................................................120 7.4.2 Methods ......................................................................................................120 7.4.3 Data ............................................................................................................121 7.4.4 Resources & Timing ....................................................................................122 7.5 Option 3. Who moves and does it matter? ....................................................123 7.5.1 Introduction .................................................................................................123 7.5.2 Methods ......................................................................................................123 7.5.3 Data ............................................................................................................124 7.5.4 Resources & Timing ....................................................................................124 References................................................................................................................................126 4 EXECUTIVE SUMMARY Introduction (Chapters 1 and 2) There are strong links between policy, internal migration and regional labour market adjustment • Policy issues include regional development, regional inequalities and the spatial impact of non-spatial policies This paper focuses the enquiry by specifying the scope, key questions, relevant theory as well as empirical and data issues • We focus on the intersection of internal migration, labour market and regional adjustment We present some theory that shows how people move out of the region in response to shocks • Migration is potentially a key regional labour market adjustment mechanism Patterns (Chapter 3) We summarise patterns and trends of internal migration and regional labour market adjustment • Internal migration is significant in itself and as a source of regional population change, and future empirical work should account for international migration • There is regional labour market adjustment occurring but there are still disparities that are persistent. These patterns in New Zealand are consistent with international evidence The next three chapters examine whether there is a link between internal migration and regional labour market adjustment, and if so, whether the link is large, as well as whether it depends on who’s affected by the shocks. Key Question 1 (Chapter 4: Does Migration Help Regional Labour Market Adjustment?) We distinguish between direct questioning and indirect inferences to reveal the factors influencing the decision to move On balance, the evidence supports a link between labour market attractiveness of locations and migration flows, but: • Current indicators are not necessarily good indicators of attractiveness • Observed differences in labour market indicators do not necessarily represent differences in attractiveness that induce migration flows. There may be differences in amenities, costs of living and other sources of omitted variable bias. • There are other forms of adjustment, such as in the housing market • There are differences across individuals that may not be obvious from aggregate flows We highlight the implications of these complications and different insights from using aggregate studies, microdata studies and longitudinal studies • Evidence of the labour market story is mixed from aggregate studies, particularly for the unemployment variable • There is more consistent (stronger) evidence from microdata studies • Longitudinal studies take into account life-cycle considerations that are potentially important in migration decisions 5 Key Question 2 (Chapter 5: How Important is Migration as a Regional Labour Market Adjustment Mechanism?) We look at how rapidly the migration adjustment process takes place, and its importance compared to other forms of adjustment Methods vary according to length of time series used and structure imposed • Short time series approaches limit the extent of analysis, and do not explicitly model other forms of adjustment • Longer time series approaches allow one to assess the contribution of migration to the full adjustment mechanism. These can be divided into studies that impose prior structure on the data, and those that impose little or no structure • Different approaches tell us about different parts of the story − no one method is always better than another Vector Autoregression (VAR) is particularly common and we include an extensive discussion of VAR studies • The results vary across countries • Internal migration is a significant regional labour market adjustment mechanism in some countries like the US and Australia, but not in most of the European countries Case studies can paint a richer picture because they are region-specific • The results suggest that external pressures and shocks impact upon different local labour markets quite differently Key Question 3 (Chapter 6: Who Moves and Does It Matter for Regional Labour Market Adjustment?) We discuss two key issues – identifying heterogeneity and the implications for regional labour market adjustment • There is clear evidence of heterogeneity in migration flows. Some characteristics have been found to affect migration in a more systematic way (e.g. age and education) than others (e.g. ethnicity). Generally, there is evidence that younger, highly skilled, single males, who do not own homes, are relatively more mobile • We find few existing studies on the implications of who moves on the origin and destination communities Research Options (Chapter 7) We present three options for future research: • that can help answer the three key questions • that are feasible research options, in terms of data availability and plausible ways of addressing the identification problem • that are likely to generate findings that are robust and of relevance for public policy Option 1: Spatial interaction model of migration-labour market links Option 2: VAR modelling of regional labour market adjustment Option 3: Case study approach to get behind the average relationships The choice of whether to proceed with each of these options depends on available resources, and a judgement about the importance of answering each key question, relative to other possible demands on research resources 6 LIST OF FIGURES, TABLES AND BOXES Figure 2.1 The scope of our work.........................................................................15 Figure 2.2 Defining our key questions ..................................................................15 Figure 2.3 A simple model of regional labour market adjustment..........................17 Figure 3.1 Number of Moves between Area Units (1991–1996) and the Cumulative Percentage of AU Moves, by Distance..............................................................30 Figure 3.2 International vs Internal Inflows (1991-1996).......................................34 Figure 3.3 International Migration Flows as a Proportion of Total Inflows and Total Outflows, by Regional Council between 1991 and 1996......................................36 Figure 3.4 Patterns of Convergence and Persistence...........................................37 Figure 4.1 Migration and Labour Market Outcomes..............................................46 Figure 4.2 Correlations of migration flows with labour market indicators...............67 Figure 5.1 Impulse Response to a 1 percent Negative Employment Shock (in the US) ............................................................................................................86 Figure 5.2 Impulse Response to a 1 percent Negative Employment Shock (in Australia) ............................................................................................................93 Figure 6.1 Incidence of internal migration by age in selected OECD countries, for the latest available year....................................................................................105 Table 2.1 Forms of regional labour market adjustment........................................19 Table 3.1 Summary of Some Common Migration Measures................................27 Table 3.2 New Zealand Internal Migration Rates (1991-1996).............................28 Table 3.3 Proportion of the identifiable population that moved within and between different (administrative) zones, 1991-1996........................................................29 Table 3.4 Internal migration in selected OECD countries: gross flows and net flows ............................................................................................................31 Table 3.5 Persistence of population and labour market indicators (Correlations over time)............................................................................................................37 Table 3.6 Correlation between recent and historical values of regional unemployment rates ...........................................................................................40 Table 3.7 Convergence indicators.......................................................................41 Table 4.1 Correlation between net migration flows and unemployment by region, for OECD countries.............................................................................................67 Table 5.1 Overview of results from the VAR studies ...........................................90 Table 5.2 Decomposition of the response to a one-standard-deviation shock in employment growth (percentages of the change in log employment)..................92 Table 6.1 Proportions of migrants by age, for the 1996 total population............104 Table 6.2 Propensities to migrate inter-state in the US, by age and education (1980-1985)......................................................................................................105 Table 6.3 Internal migration rates for the 15-29 year age group, by ethnicity in Auckland ..........................................................................................................106 Table 6.4 Internal migration rates for New Zealand’s regions by ethnicity, 1991- 1996 ..........................................................................................................107 Box 1 Gravity models ....................................................................................53 Box 2 Illustrative VAR equations....................................................................80 7 1 INTRODUCTION 1.1 Introduction This review paper is part of a broader work programme in the Treasury that raises the profile of regional issues in public policy debate. Earlier work has set out the key concepts of economic geography, made progress on the database front, and illustrated some applications of the data (see Box (2000), Kerr, Maré et al. (2001), Kerr and Timmins (2000) and Maré and Timmins (2000)). This paper builds on these previous papers, and on an extensive international and (a smaller) New Zealand literature, to investigate the role of migration as a regional labour market adjustment mechanism. In particular, it provides a foundation for further empirical work. The outcome of this review document is a recommendation of a series of potential projects that can contribute to policy making and filling research gaps. Against this background, the paper has four aims. The first is to identify the key questions that will help inform policy makers and contribute to the New Zealand research literature. Secondly, it reviews relevant international and New Zealand literature to identify the range of methods and models that are used to shed light on the key questions. The third aim is to briefly summarise key evidence in relation to each key question 3 . Finally, this paper will suggest possible approaches for further research aimed at improving our understanding the role of migration in regional labour market adjustment. The remainder of this paper is organised as follows. The next subsection outlines the motivation for this paper in more detail, covering both the policy environment and research gaps. Chapter 2 outlines a framework for thinking about the underlying issue – migration as a regional labour market adjustment – and thus, provides a map to guide the reader through this document. Chapter 2 also highlights some data and empirical issues that have to be addressed. Chapter 3 then presents some summary statistics on patterns of internal migration and of regional labour market adjustment in New Zealand. Chapters 4 to 6 form the core of this paper and discuss each of three key questions in turn. For each key question, we discuss what researchers look for, how they go about finding it, and what they have found. Finally, Chapter 7 proposes three possible empirical projects for further work. 1.2 Motivation This subsection covers the factors that led us to embark on this review document. These are discussed under two headings - the policy environment and the research literature (what we do know, what we don’t know, and other current related work). 1.2.1 Policy environment Government policies can affect outcomes in regions within New Zealand in various ways. These can be policies specifically aimed at intervening in a region, or policies that are not specifically aimed at particular regions (non-spatial policies) but that have regional implications. Both are considered below. 3 In this paper, we survey the literature and draw inferences from our reading of it. We are not undertaking a formal meta-analysis although this would provide a more structured and systematic basis for drawing such inferences. 8 1.2.1.1 Regional Development Policy It is clear that there is a resurgence of interest in the regional economic development debate. New roles in the current government were created, such as the Minister for Economic Development and Minister for Industry and Regional Development. The Ministry of Commerce was renamed the Ministry of Economic Development, and given the responsibility for regional development policy advice. Industry New Zealand was established to enter partnerships with the private sector and with local communities to transform the economic base of New Zealand. Our first focus in this paper is the link between migration and developments in regional labour markets. Migration patterns are relevant for designing spatially targeted policies. Attempts to improve the prospects of people in particular regions may be confounded if the migration response is large. Assistance could, in this case, end up benefiting new entrants to the region rather than the initially targeted population or community, or could create concentrations of poverty. Also, even if a particular policy does lower unemployment in one area, from a national perspective, these benefits could be offset by increased unemployment in other areas. Does out-migration from less well-off areas increase or decrease the welfare of those remaining? How do we address declining regions? Differences in unemployment across New Zealand regions have been remarkably persistent over the post war period (Morrison (1999)). Therefore, there is a need for more geographically based research. In particular, we need more knowledge of the differences between places and the way in which the local economy and associated opportunities interact with the presence of vulnerable groups. The focus of active labour market programmes in New Zealand has traditionally been on the characteristics of vulnerable groups. There is little focus on the characteristics of the local labour markets in which many currently seek employment. However, there has been an increasing interest in devolution. For example, the Minister of Employment and Social Services has recently announced that he intends to introduce greater "flexibility" in the delivery of employment services in the regions 4 . 1.2.1.2 Addressing Inequalities There has been a significant amount of attention on the gaps between Maori and non- Maori recently (see for example, the Closing the Gaps 2000 report by Te Puni Kokiri). There is a clear regional dimension to disparities between Maori and non-Maori. To what extent does migration ameliorate or worsen the disparities? Do the migration patterns of Maori differ from those of non-Maori? Are Maori more or less mobile than non-Maori? Do they vary with iwi area? The more important that iwi affiliations are in determining Maori migration patterns, the more policies will need to explicitly account for iwi. One could also investigate regional disparities for other population groups, or in other dimensions, such as education and health. In terms of education, what effect does 4 The Community Employment Group (CEG), a service of the Department of Labour, is physically located in nearly 30 local communities across New Zealand, and works in many more, creating local opportunities for employment and economic self-sufficiency. The service was transferred from the Department of Work and Income to the Department of Labour in early 2000 as part of the Government's strategy to strengthen community employment development capacity. Also, a formal memorandum of understanding recently reached between the Mayors' Taskforce for Jobs and Government Ministers fleshes out the Government's commitment to work in partnership with local communities to tackle unemployment. 15 below. However, it is important that our research is eventually placed into a larger context, linking to the areas that are not our primary focus. Figure 2.1 The scope of our work Internal Migration Regional Adjustment Labour Market 2.2 Key Questions Within this scope, there are still potentially many questions that can be asked. Our interest is in how regional labour markets adjust when there is a shock, and the role of internal migration. We further focus down to three key questions, as Figure 2.2 illustrates. Figure 2.2 Defining our key questions Regional labour market shock Region needs to adjust + or - (1) Regional labour market adjustment > or = or < (2) (1) Does migration help regional labour market adjustment? (2) How important is migration as a regional labour market adjustment mechanism? (3) Who moves and does it matter for regional labour market adjustment? When a region experiences a shock, the region must adjust in one way or another (e.g. via the labour, capital, housing, goods and services markets). We are interested in how the regional labour market adjusts. There are different forms of regional labour market adjustments - internal migration is one of them. The first key question asks whether internal migration responds to labour market factors, and if so, whether it helps the particular region to adjust. The second key question compares internal migration to other forms of regional labour market adjustment. The final question looks at who is migrating and whether it matters for regional adjustment. Researchers do not generally use the same methods, models or approaches to investigate these three questions. To do so would require an approach that explicitly deals with many complex interactions. What is more common is to choose an Other regional labou r market adjustmen t mechanisms Firm migration Labour force changes Job creation/destruction Worker migration Who? (3) 16 approach that enlightens us about one of the questions, while controlling for the more important sources of heterogeneity or bias that arise due to the partial nature of the analysis. Therefore, for each key question, we will identify the sort of evidence that people look for (the “what”), the way they identify/isolate patterns/relationships (the “how”), and what evidence they have found (the “findings”). With the focus and key questions already defined, what theory can and should we use to provide a way of perceiving the world? This is covered in the next subsection. 2.3 Theory and Concepts Another way of imposing structure is to adopt some theory and concepts. The purpose of a theoretical framework is to foster understanding of phenomena encountered in the real world. Any such framework necessarily abstracts from details of the complex true state of the world in order to develop an explanatory model able to provide insight into these complexities. Different explanatory models are designed to help us understand different facets of the world. Where do we begin? A good starting point is to consider an economy with only two regions. Suppose that a region experiences a local labour market shock. This shock can be either a supply shock (e.g., workers do not work anymore for whatever reason, a sudden change to the labour force participation rate) or a demand shock (e.g., a firm closes down, a new firm opens, or an increase in the demand for the underlying product/service, hence more workers needed to accommodate the higher production). We are interested in how regions respond or adjust to these shocks. For ease of exposition, we shall look at one shock consistently throughout. Given there is some interest in regional development, let’s suppose that a new policy (e.g. wage subsidies, tax breaks) is put into place to attract companies to set up in a particular region (region A). As a result, the demand for labour in region A rises. Now let’s assume that there is only one other region in the whole economy (region B) and that regions A and B are not closed (i.e., labour and capital are allowed to flow across regions). We allow amenities in regions A and B to be different; hence the absolute level of wages in regions A and B are not expected to be equal. In other words, we allow for some permanent differences in wage levels between the two regions. This can be the case for various reasons. For instance, if region A is a nicer place to live in (e.g., better views, more fresh air, less traffic congestion, etc.), then workers would be willing to accept lower wages in region A than in B. Conversely, workers need to be compensated for working in the less attractive region B. This compensation is to ensure that in the equilibrium state, utility is equalised across regions, from both consumers’ and the producers’ points of view. There is then no incentive for people or firms to change their location. In such circumstances, the demand for and supply of labour curves (in the initial equilibrium state) would be as depicted below (where the W-axis for the two regions have different starting points). We are interested to see the effects of the rise in the demand for labour in region A. The initial impact of this positive shock will be a rightward shift in the labour demand curve in region A (annotated as (1) in Figure 2.3 below). This labour demand shift would result in a higher level of wages (W) and employment (E). What would follow this rise in labour demand in region A? For now, let’s assume that region A can adjust in only one way (worker migration). At least some workers from region B will be induced to move into region A, because of better employment opportunities and 17 possibly higher wages (this depends on how flexible wages and employment are) in the affected region. Therefore, there is a rightward shift in the labour supply curve in region A (annotated as (2) in Figure 2.3). This in-migration will offset the wage increase and reinforce the rise in employment in region A. Similarly, region B will experience a leftward shift in the labour supply curve (annotated as (2) in the Figure 2.3). This will lead to higher wages and lower employment in the region where migrants are coming from (i.e. region B), as firms there are forced to compete with region A for workers. This adjustment will continue until once again nobody wishes to move, at which point the favourable impact of the initial labour demand shock has been spread across both regions, with higher wages and/or lower unemployment everywhere. Note that in the initial and final equilibrium state, utility is equalised across regions, from both consumers’ and the producers’ points of view. Figure 2.3 A simple model of regional labour market adjustment Region A Region B W W (2) Worker in-migration (2) Worker out-migration (1) Positive shock E E For simplicity reasons, the diagrams above have focused only on internal migration as a regional adjustment mechanism. There are other regional labour market adjustment mechanisms, as explored further below. This simple model can also be used to examine the effects of a negative shock. The simple model above assumes that there are no costs of adjusting for workers and firms alike (i.e., no mobility costs). Unfortunately, in real life, things are not that simple. Where there are adjustment or mobility costs, two possible scenarios are possible. If mobility costs are very high, there will be very little migration or none at all. Otherwise, migration will take place until the difference in utility from living in one region rather than the other region equals the mobility cost. In practice, there are plausible reasons to expect only slow or partial adjustment. These include: fixed mobility costs: If there are costs of moving between locations, people will not move unless the differences in the attractiveness of local labour markets are large enough to outweigh the costs. There may therefore be sustained differences in attractiveness, and hence in indicators such as unemployment and wage rates. convex adjustment costs: If adjustment is more costly if it is done all at once rather than gradually, mobility flows in response to regional shocks may have an extended impact. Such costs may arise, for instance, because of the high costs of rapidly expanding housing or infrastructure in the receiving region. 18 uncertainty: Adjustment may be slow because people want to be sure that differences in attractiveness will last. Adjustment in response to a transitory shock will be much smaller than adjustment to a permanent shock 11 . productive amenities: There may be sustained differences in labour market indicators because regions differ in dimensions other than those measured. These could arise for consumption reasons or for production reasons. For instance, people may be willing to stay in a region with a favourable climate, despite the fact that it has high unemployment. For the high unemployment rate to be maintained, it must be the case that firms choose not to create jobs in the region. There must therefore be a difference in regionally-specific productivity across regions – in this case lower productivity in the high unemployment region 12 . demographics: Some demographic groups are less mobile than others. Adjustment by means of outflows will therefore be slower for regions that have immobile populations. The stronger are these effects, the less we can rely on regional migration to aid regional adjustment to labour market shocks. Furthermore, in reality, there are many more regions, and many different reasons why people may move from one place to another (not merely job related or amenities). However, the general principles above still apply. As mentioned earlier, migration is not the only regional labour market adjustment mechanism. The paragraphs below elaborate on the range of potential adjustment mechanisms to a regional labour market shock. In a well-functioning labour market, geographical unemployment differences resulting from past shocks should be reduced, if not eliminated, relatively quickly. There could be two scenarios – firstly, suppose there is a positive region-specific shock due to a regional development program, and secondly, suppose there is an adverse regionspecific shock on the labour market. The potential adjustment mechanisms for each of these scenarios are 13 : 1) Creation or destruction of jobs – With a positive shock, existing or new local firms could create new jobs to take advantage of or as a result of the regional development program. On the other hand, a negative shock could result in local firms destroying jobs and retrenching employees. 2) Migration of firms 14 – A positive shock could attract firms from other regions to relocate to the region to take advantage of the regional development program. Conversely, a negative shock will motivate existing firms in the region to move to other regions. 3) Changes in the labour force participation rate – A positive shock will result in an increase in the relative demand for workers in the particular region. There may be eventually an outward shift in the relative supply curve for labour either because of changes in the education and training decisions, or through nonlabour force participants re-entering the labour force. On the other hand, if 11 Sjaastad (1962) emphasised the intertemporal nature of migration decisions by analysing migration decisions in the context of human capital investments. 12 For a fuller discussion, see Roback (1982). 13 The effects do not need to be symmetric, as discussed later on. 14 This concerns employers’ decisions about the location of their plants and facilities. 19 there is an adverse shock on the local labour market, some unemployed workers will remain in the region, but become “discouraged” and drop out of the labour force. 4) Migration of workers – A positive shock will attract workers from other regions 15 . Conversely, with a negative shock, workers will migrate to seek jobs elsewhere. Unemployment itself is an incentive to migrate, but the incentive is even greater if there is a wage difference. 5) Wage adjustments – Unlike all the four quantity adjustments above, we could also have a price adjustment, that is wage differentials, to provide a further incentive for the quantity adjustments to occur. A positive shock will raise the region’s wage relative to the rest of the country. A negative shock, on the other hand, results in the wage in the region falling relative to the other regions. By separating different margins for adjustment, we are not suggesting that adjustment will take only one of these forms. We would expect there to be some adjustment on several margins simultaneously. The form and extent of adjustment will depend on relative demand and supply elasticities. Furthermore, short-run and long-run adjustments may take different forms. Table 1 summarises the range of regional labour market adjustment mechanisms discussed above 16 . Table 2.1 Forms of regional labour market adjustment Quantity adjustments Within region Between regions Labour Demand (1) Job creation or destruction by local firms (2) Firm migration Labour Supply (3) Labour force participation changes (4) Worker migration (5) Price adjustment (Wage changes) Our review focuses on worker migration. One might ask how appropriate this is. In some countries, for example in the United States, worker migration is an important adjustment mechanism (see Blanchard and Katz (1992)). However, in a few European countries (see Decressin and Fatas (1995), Mauro and Spilimbergo (1998) and Mauro, Prasad et al. (1999), worker migration is not as important. Therefore, we need to see how New Zealand compares with other countries in terms of the size and composition of its migration flows. In addition, migration is an important consideration in the design of regional development policy, as mentioned in section 1.2.1.1. This can be summed up by the “people versus place” debate. Should policies be targeted at disadvantaged people, or at the places where disadvantaged people live? Place-based strategies can fail if the in-migration response is large. Therefore, it is important to understand the dynamics of worker migration. Also, focusing on worker migration is an appropriate first step given that there is lack of data for the other three forms of quantity adjustments. As mentioned earlier, in a well-functioning labour market, geographical unemployment disparities resulting from past shocks should be reduced, if not eliminated, relatively quickly. Therefore, if there are disparities in regional labour market outcomes and they are persistent, one might ask whether such disparities constitute a problem. It is possible that regional labour market disparities represent an optimal equilibrium where a region with an attractive environment and other amenities encourage people to 15 There could be an international flow as well. 16 Nijkamp, Rietveld et al. (1986) provides a diagram summarising the all the links in regional labour markets. 20 remain or move there, despite a lagging regional growth and high unemployment. To the extent that persistent regional labour market disparities reflect people’s lifestyle choice and preferences, then the issue need not be a matter of great concern. Alternatively, this persistence can be due to regional differentials in productive amenities, such as better roads, better infrastructure, etc. Large and persistent disparities (due to a slow or partial adjustment process) may also be the result of market distortions or barriers to adjustment mechanisms. In this case, the issue of regional labour market disparities is more of a concern. As discussed earlier, impediments to regional labour market adjustment include information gaps and uncertainty, fixed mobility costs, rigid wage-bargaining systems, and imperfect competition in product and labour markets. These are all plausible reasons to expect only slow or partial adjustment. We are interested in one particular factor in such persistence in regional labour market disparities – the direction, size and composition of internal migration flows. If a region experiences an adverse local labour demand shock, we expect to see, among other things, a movement of labour away from weaker to stronger regions. Does internal migration respond to labour market differentials in the direction predicted by theory (e.g. from high unemployment area to low unemployment area)? If the answer is yes, then worker migration is said to be helping regional adjustment. Then, we would expect to see the effects of any shocks dissipating over time. We would not expect to see persistence or divergence between locations (of whatever geographical scale). On the other hand, if migration fails to promote regional adjustment, shocks to locations will be permanent (i.e., not dissipate over time). In this case, locations with low unemployment, for example, will continue to have low unemployment, or will have even lower unemployment. The first key question examines whether internal migration helps regional labour market adjustment or otherwise. Alternatively, migration might be playing a very small role. An examination of migration and commuting flows across OECD countries shows that the scale of labour mobility is limited in many countries, and therefore it is probably not sufficient to act as an adjustment mechanism (see OECD (2000)). This is our second key question. In addition to the direction and size of internal migration flows, the heterogeneous characteristics of migrants could well have implications for the origin and destination regions. The third key question examines the question of who moves and how this matters for regional labour market adjustment. Equipped with some theory and concepts above, and how they relate to the three key questions, we should consider the next set of issues for our journey towards gaining an understanding of the role of internal migration as a regional labour market adjustment mechanism – data and empirical issues. 2.4 Empirical Issues and Data Sources When we do empirical work, we inevitably have to deal with some data and empirical issues. This can be considered another way to impose structure on the world, although it can be a constraint rather than an explicit choice on our part. This subsection highlights a few common empirical issues that research work in this area will encounter, and summarises the range of data sources that will be useful for our projects. 21 In the literature, there is a distinction between regional shocks and aggregate shocks. Regional shocks are likely to trigger different adjustments of labour, capital, and output than do aggregate shocks. One reason may be that some regions are specialised in the production of particular goods and services, and thus national markets are somewhat arbitrary constructs. If regional dynamics are different, policymakers may need to identify regional differences and/or imbalances in formulating policies and assessing their consequences. Similarly, there is a distinction between regional shocks and sectoral shocks. The two are not unrelated concepts. For example, an aggregate shock may have different region-specific impacts because the regions have different sectoral compositions. Therefore, whether we use the concept of regional mobility, or sectoral mobility, depends on what structure we impose on the complex true state of the world. It depends on how we identify shocks. Our paper here focuses on the regional dimension 17 . There is also the issue of what the appropriate unit of analysis is when considering migration flows. This is essentially defining the parameters of what constitutes a migration flow. The first dimension is a temporal one – what the appropriate time period for the analysis is. In New Zealand, we only have census data to track internal migration flows. As a result, there is a five-year cut-off period, which is determined by the date of the previous census. The scope, limits and reliability of our census data for internal migration flows have been well documented (see for example, Poot (1986a)). The second dimension of concern is a spatial one. Studies have shown that alternative calculations of distance between origin and destination areas can have very significant effects on the migration measures. In particular, place-to-place migration declines with distance. Put in another way, as we move from an aggregate unit of analysis (e.g. regions) towards a more disaggregated level of analysis (e.g. territorial local authorities (TLA), area units (AU), or meshblocks (MB)), the number of counted moves will presumably increase. Distance is also important because reasons for moving may vary by distance moved, that is, local movers (within area units or labour markets) and longer distance migrants (across area units or outside labour markets) move for different reasons. There is evidence that economic and employment factors are much more important in long distance than local movements (see Green (1994) and Lichter and De Jong (1990)). How do we decide which level of analysis is appropriate for our purpose? A logical spatial unit of analysis could be areas whose boundaries are “delimited on the basis of socio-economic criteria”, as adopted in Australia (Duke-Williams and Blake (1999)). How do we operationalise this in real life? One common strategy is to work with administratively defined areas within which labour market policies can be taken by planning authorities, such as regional councils and territorial local authorities. Alternatively, one can use functional labour market areas, which are usually preferred on theoretical grounds, although this strategy has several drawbacks in practical modelling situations (see Isserman, Taylor et al. (1986)). Definitions about what constitutes a local labour market vary considerably in the literature (see for example Box 1 in (OECD 2000). Otherwise, one could also define boundaries strictly by distance. The concept of distance can simply be the distance 17 See for example, the book edited by Padoa-Schioppa (1991) for a starting point on references examining sectoral mobility and mismatch. 22 over the Earth’s surface between points (e.g. 20 kilometres), or a distance through a transportation network (e.g. road distances or travel times). One might then ask what the most appropriate spatial framework is. The answer is that it depends on the purpose of analysis. This question has to be asked whenever we conduct any empirical work. It might well vary according to the key question at hand. Earlier migration theories treat migration decisions as though only individuals make them. Some have noted that migration is often a family or household decision (see Greenwood (1997) and Snaith (1990)). A move takes place only if the net gain accruing to some members exceeds the others’ net loss (see e.g. Mincer (1978), as cited in Lichter and De Jong (1990)). Therefore, some recent approaches model migration as a risk-sharing behaviour of families. In contrast to individuals, households can diversify their resources such as labour, in order to minimise risks to the family income. Therefore, by sending some family members to work in other labour markets (where wages and employment conditions are negatively or weakly correlated with those in the local region) and through remittances by family members, the family can secure their economic well-being in the face of an adverse shock to the local region. With this type of model, one can explain migration flows in the absence of wage differentials. Unfortunately, in New Zealand, there is limited data on migration at the household unit. Another issue is how we actually impose structure on what we observe and thus draw inferences. The key source of statistical analyses in migration research in previous decades has been aggregate data due to a lack of available micro data sets or insufficient computer power. These data are either time series observations or cross section data, or some combination of both. Over recent decades, however, there has been an increasing amount of research using individual data. Each of these is discussed briefly below. Time series analyses assume that the relationships among a set of variables (e.g. migration, wage and unemployment rate) stays the same over time. This can of course be challenged if one believes, for example, that economic reforms in the past 15 years have altered the behaviour of people in the economy. A particular issue is the problem of spurious regressions, whereby common trends can be mistaken for a strong relationship. Alternatively, one can use cross-sectional data, which is data on regions (regional councils, territorial local authorities, area units, or meshblocks) at the same point in time. Cross-sectional studies have their own problems – heterogeneity. Different regions have different amenities, different levels and types of economic activities, different social structures, etc. Studies that utilise cross-sectional data implicitly assume that the unit of analysis is a group of people with a common set of characteristics. Also, cross-sectional studies estimate an average relationship among a set of variables (e.g. migration, wage and unemployment rate) across regions. We need to control for some of these sources of heterogeneity in the cross-sectional units in order to make sense of the relationships of interest. One can also use elements of both time series and cross-sectional data, resulting in a pooled data set. In other words, we are imposing structure by inferring from variation over time and across areas. This is less restricted since we are not imposing the assumption that relationships are homogeneous either over time or across areas. 23 The problems of using aggregate data are already well known in the literature 18 (see for example, Bauer and Zimmermann (1998) and Greenwood (1997)). The alternative is to use individual data. However, the provision of individual data remains a problem in many cases, particularly in the New Zealand case, for a few reasons. Firstly, there is a confidentiality issue in the provision of statistics by Statistics New Zealand. Secondly, it is expensive to track the same units over time, particularly at the national level. There is also a debate about equilibrium versus disequilibrium modelling, which is far from settled (see Hunt (1993), Greenwood (1997), Goetz (1999) and references cited therein). Disequilibrium models emphasise spatial variations in economic opportunities, and de-emphasise amenity factors. They assert that the migration process and land and labour markets are relatively inefficient. Since there is a significant time lag for the adjustment process to occur, one expects to observe persistent differences in economic opportunities between areas and across time. Therefore, one may choose to model variables such as employment opportunities as being integrated of order one (i.e., nonstationary). On the other hand, the equilibrium view argues that the migration process is mainly driven by consumption amenity factors. Economic opportunities differentials across areas may exist but are adjusted fairly quickly by adjustments in efficient land and labour markets. Therefore, at any one point in time, the assumption of spatial equilibrium is reasonably accurate. If this were the case, one can resort to modelling variables proxying for economic opportunities as being integrated of order zero (i.e. stationary). Rather than adopting explicitly at the outset an equilibrium or disequilibrium view (two polarised scenarios), we believe that it is more appropriate to ask the question of how long it takes for a system to adjust back to equilibrium, assuming that it is stable. As noted by Greenwood (1997), the speed of adjustment to equilibrium is an important issue about which little is known. This in fact is similar to our second key question, as explained in section 5. These are just a few empirical issues that are important in migration research 19 . The key message from this discussion on empirical issues is that there are important decisions and judgements to be made on how to treat these empirical issues and that we may want to vary these treatments according to the key question at hand. Before we move on, it is worthwhile giving a general flavour of the potential New Zealand data sources that might be relevant to this area of research and the tradeoffs between them. As in most other countries, the main source of data for our research will be the New Zealand Census of Population and Dwellings. Using census data, we can only track movements over a five-year period. The census asks for current and previous census (5 years ago) residential addresses. From these two questions, a change of address, and therefore a move, can be identified 20 . An origin-destination table, produced by Statistics New Zealand (SNZ), provides gross movements of people in and out of Area 18 For example, there are substantial problems in identifying the determinants of the migration decision, due to the inaccuracy of the aggregated measures to account for the important factors in the individual decision making (i.e., selectivity bias). 19 Maré and Timmins (2000) also provides a discussion of some of these issues. 20 Fixed period questions (as above) tend to underestimate geographic turnover and are unable to capture the following return and repeat migration (see Poot (1986a)). 24 Units within New Zealand 21 An Area Unit (AU) is a SNZ defined spatial unit roughly equivalent to a city suburb and normally contains 3,000–5,000 people, but AUs in rural areas can be considerably larger and contain fewer people. There are 1,766 area units defined for New Zealand. Area units can be aggregated into Territorial Local Authorities (TLAs) and Regional Councils (RCs) to allow examination of flows between larger administration zones. Significant changes were made to the boundaries of New Zealand’s local government regions in 1989 and so it is not possible to construct a time-series of migration flows that goes back to the early 1970s (as cited in Bedford, Goodwin et al. (1997)). Therefore, we shall restrict our empirical analysis of inter-regional migration to the last two inter-census periods: 1986-1991 and 1991-1996. Census data contains mobility information of everyone in New Zealand at the time of the census. We therefore observe people who entered New Zealand in the five years prior to a census, but not those who left New Zealand over that period. The mobility data that we use contains information on movements only of people who were in New Zealand at the times of both the current and previous censuses (1986 and 1991 for the 1991 census, and 1991 and 1996 for the 1996 census). The analysis therefore excludes flows arising from international migration. This exclusion is discussed further below. The origin-destination table provides previous residence information at Area Unit level and for overseas countries. Current residence, however, is only provided for New Zealand (at Area Unit level). Besides providing data on migrant status, the census provides a vast amount of demographic and labour market data down to meshblock level, which can be combined to create AUs, TLAs and RCs. This allows the data to be matched with the migration data described above and provide characteristics for each administrative zone. Key variables include age, sex, ethnicity, religion, income, employment status, occupation, child dependency, family relationships, living relationships, living arrangement, education, religion, industry, and unpaid work. The census generally has good geocoding. The scope, limits and reliability of using the census data on internal migration have also been discussed elsewhere (e.g. Poot (1986a)). The latest data set is for the 1996 census. Provisional 2001 census data will be available around mid-2002. We will need to supplement census data with other potential data sources. These are briefly highlighted below. The Household Economic Survey (HES) 22 collects information on household annual income and expenditure, as well as a wide range of demographic information on individuals and households, for the years 1983/84 to 1997/98. However, post 1998, the survey is three yearly (i.e., the next will be for 2001). The survey covers approximately 3000 private households living in permanent dwellings and resident in New Zealand. Different households are surveyed each year. The survey data are available at a national level and for 4 broad regions (Auckland, top of North Island, bottom of North Island and South Island). There is information on around 3000 expenditure items (e.g. food, housing, household operation, apparel, transportation, insurance, medical and leisure). Variables include household/family type, employment, 21 Only 82.8 percent of individuals in the 1996 census can be traced to an area unit at the time of the previous census: 0.7 percent and 4.4 percent respectively could be traced only to regional council or TLA; 6.5 percent were living overseas in 1991, and 5.5 percent did not respond to the question about prior location. 22 The Household Economic Survey (HES) used to be called the Household Income and Expenditure Survey. 31 movement in New Zealand are relatively high – similar to high rates in Australia, Canada, and the United States. On the other hand, Belgium, Ireland, and the Netherlands have the lowest rates of residential mobility (as cited from Greenwood (1997)). In particular, Long’s measures indicate that around 1980 or 1981, the percentage of population who moved from one dwelling unit to another in one year was over 19 percent in New Zealand; 17 to 18 percent in the US, Canada and Australia; 9 to 10 percent in Great Britain, Sweden, Japan and France; 7 to 8 percent in the Netherlands and Belgium, and 6 percent in Ireland (see Long (1992)). To disentangle (labour market related) migration from residential moves, Long (1991) distinguishes between moves within and between local areas. Based on one-year measures, the US has very high internal migration rates. Based on five-year measures, the US and Canada have high rates. Unfortunately, there were no comparable data for other countries such as New Zealand and the United Kingdom (as cited from Greenwood (1997)). Similarly, OECD (2000, Table 2.12) illustrates developments in internal migration rates in selected OECD countries in more recent periods (around 1995) and finds that only Japan, Canada and the USA had higher gross migration rates than New Zealand. The OECD study also shows that New Zealand had a relatively low ratio of net to gross migration, suggesting that there are relatively large two-way flows between regions. This table is reproduced as Table 3.4 below. Table 3.4 Internal migration in selected OECD countries: gross flows and net flows a Source: Extracted from OECD Employment Outlook (2000, p. 53) 32 Comín (1999) presents a model of the relationship between the size of the country and the degree of inter-regional labour mobility. He concludes that if European economies were like the US (by rescaling them), the degree of inter-regional mobility in Europe would be as large as in the US. As far as we are aware, a similar exercise has not been done for New Zealand. This may or may not alter the conclusion earlier that New Zealand has a fairly significant degree of mobility, in comparison with other OECD countries. While we are mainly interested in the link between internal migration and regional labour market adjustment, it is important that we do not ignore the fact that people have the choice to, and some do in fact, leave or enter the country. Ignoring the international flows, particularly when they are significant, may well distort the estimated labour market story, similar to an omitted variable bias. The New Zealand Planning Council, Richard Bedford, James Newell and others have done some work in this area. Lidgard, Bedford et al. (1998) compares the components of the New Zealand population change over the period 1986-1996. The net gain from international migration during this decade was just over a fifth of the total population increase. When Gorbey, James et al. (1999) look at the breakdown of total population growth into natural increase and net international migration over the period 1950 to 1998, they find that net (international) migration has been the main cause of fluctuations in population growth, as the annual rate of natural increase has been a smoothly trending variable. Only twice since the early twentieth century (early 1970s and mid 1990s) has net (international) migration approached natural increase 30 . However, up to here, we have only been comparing different components of population change at the national level. We still have not assessed sources of population structure change at the regional level. This is considered next as we examine how internal migration compares to international migration. Although international migration is significant for population change at least in recent years, inter-regional flows, in aggregate, continue to be larger than international flows in most parts of the country (see Bedford, Goodwin et al. (1997)). Table 1 in Bedford, Goodwin et al. (1997) shows the respective contributions of inter-regional and overseas immigration to the usually resident populations of the 14 local government regions in New Zealand, as these were defined in 1991. The figures show that the number of inter-regional immigrants exceeds the number of overseas immigrants in all the regions except for Auckland, over both the periods 1986-1991 and 1991-1996. Aside from Auckland, Wellington and Canterbury, the internal in-migration figure is at least about twice the international in-migration figure. In some cases, the internal in-migration figure is more than 5 times the size of the corresponding international figure. For Wellington and Canterbury, the internal immigration figure is about 50 percent higher than the international immigration figure. The evidence clearly indicates the significance of inter-regional migration in determining the changing spatial distribution and composition of New Zealand’s population. One might ask whether the predominance of internal migration is only a recent phenomenon. NZPC (1989) provides estimates of migration into each local government region (as defined prior to 1980) from other parts of New Zealand and from overseas countries (see Table 12) over the periods 1976-1981 and 1981-1986. For all the regions, the number of in-migrants from other parts of New Zealand is greater than 30 See Figure 5.6 in Section 5.3 of the Statistics New Zealand Official Yearbook on the Web 1999 (http://www.stats.govt.nz/). 33 the number of international in-migrants. Aside from Auckland, the internal in-migration figure is at least about twice the international in-migration figure. In some cases, the internal in-migration figure is more than 10 times the size of the corresponding international figure. In Auckland, the internal in-migration is about 70 percent (25 percent) larger than the international figure in 1976-1981 (1981-1986). Meanwhile, Table 13 in (NZPC 1989) shows the estimates of migration from each local government region to other parts of New Zealand and to overseas countries over the periods 1976- 1981 and 1981-1986. The number of out-migrants to other parts of New Zealand exceeds the number of international out-migrants for all the regions, except for Auckland during the 1976-1981 period. In particular, the internal out-migration figure is at least about twice (again, up to more than ten times) the size of the international outmigration figure, in all regions except Auckland, Wellington and Canterbury. Therefore, the predominance of internal migration over international migration has been borne out even in earlier periods. Up to here, we have highlighted the results from partial analyses (i.e. any two of the three components – internal migration, international migration and natural increase). NZPC (1989) presents a broader view of the components of population change (see Table 8). Two interesting observations can be made. Firstly, the summary statistics indicate that net migration (the balance of internal and overseas in- and out-migration) offers much more variation across regions and between periods than does natural increase. Secondly, the contributions of natural increase and net migration to population change in the regions since 1976 have been quite variable and diverse across the regions. However, it is safe to conclude that there has been a tendency for net migration to become relatively more important in determining regional population change. As cited from NZPC (1989), a study by Lowe (1988a) covering the period 1956-1986 found that net migration was equivalent to more than half the natural increase in more than two regions out of three between 1981 and 1987. This was the case in only two regions out of four in the 1960s, and of fewer still in the 1950s. It is noted that net migration here is the balance of internal and overseas in- and outmigration. However, given the evidence discussed earlier that internal migration dominates international migration, one could infer that internal migration is indeed an important contributor to regional population changes. There is similar evidence overseas that internal migration is an important source of population structure change for regions. As Taylor and Bell (1996) puts it, “…in Australia, as elsewhere in the developed world, internal migration has become the primary determinant of regional population change” (p.154). This trend is also reflected by researchers such as Blanchard and Katz (1992). These authors treat the movement in the state labour force equation as characterising the migration of workers, because “most of the differences in average employment growth rates across states are due to migration, rather than to differences in natural population growth rates” (p. 17). With respect to international migration, it is worth noting that New Zealand census data excludes migration flows to countries other than New Zealand. This exclusion is due to the obvious lack of census data on people who had left New Zealand between consecutive censuses (i.e., international migration outflows) 31 , and the lack of consistent information on "regional" characteristics for other countries. Excluding international flows is potentially a weakness in empirical analyses if flows across New Zealand's border are a significant means of regional labour market adjustment. The open border between Australia and New Zealand makes Trans-Tasman migration a 31 James Newell of MERA has derived estimates of international outflows at the regional level but not yet at other levels of aggregation. 34 feasible option for people seeking better labour market prospects in the neighbouring country 32 . Figure 3.2 analyses the relative sizes of internal and international migration inflows, and the relationship between them. At the Area Unit level, overseas inflows do not exceed internal inflows (Figure 3.2a), i.e., internal inflows are a greater proportion of total inflows. However, the picture changes at the Regional Council level (Figure 2b). The Auckland region is unique because it is the only region where overseas inflows out−weigh internal inflows (Figure 2b). This is true at least in 1991-1996 when international inflows into Auckland were at historically high levels. The Wellington and Canterbury regions have the next largest international inflow component. However, for both these regions, internal flows still dominate. Figure 3.2 International vs Internal Inflows (1991-1996) Overseas Inflow Rate Internal Inflow Rate 025 50 75 100 0 25 50 75 100 External In Flow Rate (91-96) Internal In Flow Rate (91-96) 510 15 20 25 0 10 20 30 NORTHLAN AUCKLAND WAIKATO BAY OF P GISBORNE HAWKE'S TARANAKI MANAWATU WELLINGT WEST COA CANTERBU OTAGO RE SOUTHLAN TASMAN R NELSON R MARLBORO Aus. & UK In Flow Rate (91-96) Internal In Flow Rate (91-96) 510 15 20 25 0 10 20 30 NORTHLAN AUCKLAND WAIKATO BAY OF P GISBORNE HAWKE'S TARANAKI MANAWATU WELLINGT WEST COA CANTERBU OTAGO RE SOUTHLAN TASMAN R NELSON R MARLBORO Rest OS In Flow Rate (91-96) Internal In Flow Rate (91-96) 510 15 20 25 0 10 20 30 NORTHLAN AUCKLAND WAIKATO BAY OF P GISBORNE HAWKE'S TARANAKI MANAWATU WELLINGT WEST COA CANTERBU OTAGO RE SOUTHLAN TASMAN R NELSON R MARLBORO Note: The relationship between New Zealand internal and international inflows at (A) Area Unit, (B) Regional Council, (C) Regional Council (includes all internal inflows, but only international inflows from Australia and the UK), and (D) Regional Council (includes all internal inflows and all international inflows, excluding inflows originating in Australia and the UK). The two largest (gross numbers) overseas inflows are from Australia and the United Kingdom respectively. It is argued that due to weak perceived barriers between NZ and these two countries (both immigration and cultural), these flows are important. Figures 3.2c and 3.2d explore the international inflows in more detail (at the regional council level). Figure 3.2c suggests that on their own, international flows from Australia and the United Kingdom are not as important as internal flows (even in Auckland). This is not surprising as wrapped up in Australia and UK inflows will be many return Kiwi migrants (working holiday etc.), thus the inflows are absorbed across all NZ regions. However, when Australia and UK inflows are removed (see Figure 3.2d), we find that 32 The national border does, however, still appear to be an impediment to movement – see Poot (1995) A C D B 35 overseas inflows are heavily concentrated in Auckland (Wellington and Canterbury lose their importance as international destinations) 33 . The omission of external flows in the analysis becomes a concern when external migration flows are significant, particularly when they are larger than internal migration flows. The discussion here focuses on inflow and outflow rates separately. With regard to inflow rates, overall, international migration seems to be a fairly constant proportion of local population across regions, with the exception of Auckland (see Figure 3.2b). In areas with high internal inflow rates such as Nelson and Tasman 34 , the omission of international or external inflows would not constitute a significant problem. On the other hand, in areas other than Auckland where there are low internal inflow rates (e.g. Southland and Taranaki 35 ), the omission of external inflows would understate the extent of inflow rates. The bias is even more pronounced for Auckland, which not only has a low internal inflow rate, but also a high external inflow rate. Figure 3.3 shows the international migration flows as a proportion of total inflows (inflows from other parts of New Zealand and overseas countries) and total outflows (outflows to other parts of New Zealand and overseas countries 36 ) for the period 1991- 1996. For example, just over 60 percent of the total inflows into Auckland are from overseas countries, whereas just over 40 percent of the total outflows from Auckland have gone overseas. There are two messages to be inferred from Figure 3.3. Firstly, it shows that the pattern of external outflow rates is similar to that of external inflow rates, although outflows are less concentrated in Auckland than are the inflows. Secondly, Figure 3.3 indicates that for all the regions except Auckland, Wellington and Canterbury, internal migration (both inflows and outflows) accounts for more than 70 percent of total migration. The implication is that one should consider using modelled outflow data 37 for cases where international migration is significant, the extreme case being Auckland. For our purposes of detecting the relationship between migration and regional labour market adjustment, the relative size and volatility of international migration are not the only things we are interested in. As important is the relationship between international migration flows and regional labour markets. We expect international migration flows to respond to imbalances between New Zealand and other countries. It is an open question whether international migrants choose their location within New Zealand based on regional labour market conditions. The primacy of Auckland as an international migrant destination may dominate the role of relative regional labour market conditions. The determinants of location choices within New Zealand are probably different for internal and international migrants, and for international inflows as opposed to outflows. The bias that would be introduced by ignoring international migration depends on how different these determinants are, as well as on the relative size of the different flows. 33 Note however that Figure 3.2 uses data on inflows between 1991 and 1996, when there was an unusually high inflow of international migrants, especially from countries other than Australia and the United Kingdom. 34 These two regions are the two observations at the bottom right of Figure 3.2c, although their names are not intelligible from the figure. 35 These two regions are the two observations at the bottom left of Figure 3.2c, although their names are not intelligible from the figure. 36 The estimates of external outflows are provided by James Newell (see Newell (2001)). 37 Recall that international outflows are not available from the census data. 36 Figure 3.3 International Migration Flows as a Proportion of Total Inflows and Total Outflows, by Regional Council between 1991 and 1996 0% 10% 20% 30% 40% 50% 60% 70% NORTHLAND AUCKLAND WAIKATO BAY OF PLENTY GISBORNE HAWKE'S BAY TARANAKI MAN-WAN WELLINGTON WEST COAST CANTERBURY OTAGO SOUTHLAND MARL-NEL-TAS As a Proportion of Total Migration Flows Overseas Inflow Share Overseas Outflow Share In summary, this section has given a “contextualised” view of internal migration trends and patterns in New Zealand. The key message is that internal migration is a significant source of regional population changes, and that future empirical work should find ways to account for the international migration aspect. Before moving on to the key questions, it is important that we look at the other items of focus (see Figure 2.1) - labour market and regional adjustment. 3.2 Regional Labour Market Adjustment We wish to examine the relationship between internal migration and regional labour market adjustment. However, there is no universal agreement on what constitutes regional labour market adjustment. How do we know if regional labour markets are in fact adjusting and if so, how much regional labour market adjustment is there? To many commentators, regional adjustment occurs when differences between regions become less. This view implies that the equilibrium - when no further adjustment is needed - is one in which all regions are the same. Alternatively, we might accept that there are stable long-run differences between regions (a disequilibrium view, as discussed in Section 2.4). In this case, regional adjustment entails restoring long-run relativities after a regional shock. In this section, we consider two types of evidence on the amount of regional adjustment - persistence and convergence. The four panels of Figure 3.4 illustrate the range of patterns that we are looking for in the data. The horizontal axis represents time and the vertical axis could be any labour market outcome. Each of the three lines represents one of three regions. In the first row of the figure (labelled "persistence"), the relative order of outcomes is maintained, so that a region that performs relatively well in one period also does so in other periods. The two graphs labelled "no persistence" show a pattern where the order is not maintained. In the convergence column of Figure 3.4, outcomes are becoming more similar across time, whereas the two graphs on the right show divergence, where the gap between well performing and poorly performing regions widens. 37 Figure 3.4 Patterns of Convergence and Persistence Convergence Divergence Persistence No persistence What patterns do the actual data illustrate? The remainder of this subsection will cover the summary evidence for both persistence and convergence/divergence. First, we present results for how persistent relative local labour market outcomes are across time, as sourced from Maré and Timmins (2000). As an indicator of this, the authors use the correlation between labour market outcomes for locations over time. Table 3.5 presents such correlations for five labour market indicators. The three panels (A, B, and C) of Table 3.5 relate to three different levels of spatial aggregation (area units, territorial local authorities and regional councils). Table 3.5 Persistence of population and labour market indicators (Correlations over time) (A): Area Unit Relationships (n = 1,652 # ) Employed (E/WAP) Unemployed (U/LF) Participation (LF/WAP) Median Income Population (% of NZ pop) Levels 86–91 0.78* 0.73* 0.72* 0.87* 0.98* 91–96 0.88* 0.84* 0.82* 0.92* 0.98* 86–96 0.72* 0.70* 0.64* 0.78* 0.95* Growth Rates 8691–9196 -0.39* -0.18* -0.38* 0.18* 0.46* (B): Territorial Local Authority Relationships (n = 74) Employed (E/WAP) Unemployed (U/LF) Participation (LF/WAP) Median Income Population (% of NZ pop) Levels 86–91 0.86* 0.88* 0.80* 0.88* 1* 91–96 0.93* 0.92* 0.89* 0.93* 1* 86–96 0.74* 0.89* 0.63* 0.81* 1* Growth Rates 8691–9196 0.08 -0.15 0.07 0.00 0.73* 38 (C): Regional Council Relationships (n = 16) Employed (E/WAP) Unemployed (U/LF) Participation (LF/WAP) Median Income Population (% of NZ pop) Levels 86–91 0.75* 0.84* 0.62* 0.93* 1* 91–96 0.96* 0.95* 0.93* 0.99* 1* 86–96 0.59* 0.81* 0.45 0.91* 1* Growth Rates 8691–9196 0.29 0.29 0.16 0.09 0.86* Notes: # The analysis here only uses the 1,652 Area Units (1996 boundaries) which had a population of 100 or more. The 123 Area Units that were excluded accounted for 2,244 people, or 0.06% of the 1996 New Zealand total population. * = significant at 5%. Persistence in growth rates between 86–91 and 91–96. Source: Maré and Timmins (2000) The first panel of Table 3.5 (Panel A) shows high correlations for all indicators at the area unit level, between 1986 and 1991, between 1991 and 1996, and for the longer period between 1986 and 1991. Area Units that had high employment rates in 1986 tended to have high employment rates in 1991. A similar pattern is observed when comparing 1996 to 1991. The correlations of 1996 indicators with 1986 indicators are generally weaker than those between the shorter 1986-91 and 1991-96 periods, suggesting that persistence is less pronounced over longer time periods. The second and third panels (B and C respectively) of Table 3.5 show comparable results for TLAs and Regional Councils. One of the patterns that emerge is that persistence is weaker between 1986 and 1991 than in the following five years. This earlier period was a period of significant change and contraction, which appears to have altered the pattern of relative labour market outcomes across regions. The only correlation that is not high and significant is the 1991-96 correlation of participation rates across time for regional councils. The final row of each panel reports the correlation between growth rates in each of the variables. The correlation figures provide an indication of whether locations that had relatively high growth rates of an indicator between 1986 and 1991 continued to have relatively high growth in the 1991-96 period. In the first panel of Table 3.5, it appears that growth rates are not persistent for employment, unemployment and participation rates, but that median income growth and population share growth are 38 . It is likely that some of the apparent reversal of fortunes implied by the negative correlations of growth rates at the area unit level are due to random fluctuations due to the smallness of the areas. As the level of spatial aggregation increases, there appears to be stronger evidence of persistence in growth rates. Growth rates are positively correlated for regional councils, although the correlations are not statistically significant, except for population share. In summary, the evidence of persistence depends on the variable we are looking at, whether it is specified in levels or in growth terms, the level of aggregation, and the 38 A different relationship between the growth rates of 86–91 and 91–96 for income growth is observed when using per capita income. Median income did not present a particularly strong relationship, however, per capita income was found to be negative and the relationship strengthened as the spatial unit was increased from AU to RC. We do not currently have a full explanation of this difference, although it points to the importance of patterns of income change at the upper end of the income distribution. 39 time period. Although there is no overwhelmingly strong evidence of persistence, the evidence of such persistence is nevertheless strong enough to warrant further investigation. One might ask if the persistence of these labour market variables and population flows holds for other countries as well. This is found to be the case in the US during the 20 th century. Local areas which grow rapidly during one decade, tend to do so also over the following decades. Across US states, high persistence has been documented for employment growth over the period 1909-1953, for net migration over the period 1900- 1987, and for employment growth over the period 1950-1990 (see Barro and Sala-i- Martin (1991), Blanchard and Katz (1992), Rappaport (2000a), and references cited therein). Similarly, across US cities and counties, there is high persistence in population growth over the periods 1950-1990 and beginning from 1930 onwards respectively (see Rappaport (2000a)). Table 3.6 below, as reproduced from the OECD Employment Outlook (2000), shows that there is a high correlation between the previous and current unemployment rates in many OECD countries. In fact, unemployment is not the only labour market condition that is found to be persistent. According to the OECD Employment Outlook (2000), regional labour market disparities are important and persistent in many OECD countries. Such disparities are only partly explained by the composition of the labour force and the sectoral mix of regions. A specific regional dimension of labour markets clearly exists in many countries. 40 Table 3.6 Correlation between recent and historical values of regional unemployment rates Level 1 Level 2 Country Number of regions Correlation between 1997 and 1995 rates Correlation between 1997 and 1990 rates Correlation between 1997 and 1985 rates Number of regions Correlation between 1997 and 1995 rates Correlation between 1997 and 1990 rates Correlation between 1997 and 1985 rates Australia (a) 8 0.80 0.69 .. 60 0.70 0.58 .. Austria 3 0.99 .. .. 9 0.97 .. .. Belgium (b) 3 1.00 0.94 0.95 11 0.99 0.96 0.81 Canada 10 0.99 0.96 0.88 59 0.96 0.93 .. Czech Republic 1 — — — 8 0.82 .. .. Finland (c) 1 — — — 5 1.00 .. .. France 8 0.99 0.92 0.82 22 0.94 0.83 0.82 Germany (d) 16 1.00 0.96 .. 36 0.99 0.96 .. Germany (e) 11 0.98 0.8 0.72 31 0.97 0.87 0.78 Greece 4 0.97 0.99 .. 13 0.89 0.66 .. Hungary (f) 1 — — — 7 0.92 .. .. Italy 11 0.99 0.99 0.86 20 0.99 0.97 0.81 Japan (g) 10 0.92 0.92 0.82 47 .. 0.94 .. Korea 1 .. .. .. 15 0.82 0.92 .. Mexico 32 0.80 .. .. .. .. .. .. Netherlands 4 0.95 0.91 .. 12 0.89 0.8 .. New Zealand (h) 1 — — — 12 0.82 0.61 .. Norway 1 — — — 7 0.95 0.92 .. Portugal 3 0.29 0.06 .. 7 0.83 0.79 .. Spain 7 0.94 0.79 0.59 17 0.96 0.87 0.64 Sweden (c) 1 — — — 8 0.97 .. .. United Kingdom (c) 12 0.95 .. .. 37 0.95 .. .. United States (b) 51 0.86 0.51 0.54 172 .. .. .. EU (i) 76 0.95 0.87 0.76 201 0.95 0.86 0.77 OECD (j) 192 0.96 0.85 0.84 588 0.92 0.79 0.77 .. Data not available. — Not applicable. (a) Year 1992 instead of 1990 (b) Year 1996 instead of 1997. (c) Year 1996 instead of 1995 and 1998 instead of 1997 (d) Year 1991 instead of 1990. (e) Data refer to western Germany. (f) Year 1993 instead of 1995 and 1998 instead of 1997. (g) Level 1: year 1998 instead of 1997. Level 2: year 1995 instead of 1997. (h) Year 1998 instead of 1997. (i) The correlation rates have been calculated from the values of each region belonging to the European Union. Denmark and Luxembourg, which constitute one region each at both levels of territorial breakdown, as well as Ireland at Level 1 only) are included. (j) The correlation rates have been calculated from the values of each region belonging to the countries listed above, plus Denmark and Luxembourg, which constitute one region each at both levels of territorial breakdown and Ireland at Level 1 only). Source: See OECD Employment Outlook (2000, p. 47) 47 In the New Zealand context, such survey-based data are rather scarce, as discussed later. Therefore, the first approach (i.e. inferring motivations from actual mobility patterns) is likely to prevail. This is why the discussion in this chapter is predominantly on the first approach. Now that we have highlighted the sort of evidence researchers search for, the next question is how they actually isolate the relationships and evidence to answer the key question at hand, i.e. presence of a labour market story. 4.2 How? This subsection looks at the range of approaches that studies have used in shedding light on the key question. As before, the discussion here will be divided into two parts. The first part looks at techniques that infer motivations for migration from actual mobility patterns, while the second part covers the direct approach of determining motivations at the individual level. 4.2.1 As revealed by actual mobility patterns The labour market story is essentially a link between observed labour market conditions and observed migration flows. Let’s look at the first main ingredient - what qualifies as a labour market attractor? Labour market attractors The answer to that question would depend on the economic theory that one adopts 44 . Different theories highlight different reasons why observed labour market conditions are not a good guide to the differences in utility to which migrants are assumed to respond. Two approaches that have been used widely view migration either as an investment decision (the human capital approach), or a spatial job search process. The spatial job search approach extends the human capital approach by incorporating explicitly the joint determination of location and job choice in a world with uncertainty. As noted by Sjaastad (1962), from the worker’s point of view, migration is considered as an investment. Observed labour market conditions may be a very poor measure of how well potential migrants expect to fare in an area over their lifetime. People consider from time to time the income gain and other benefits which may result from a move and compare such benefits with pecuniary and non-pecuniary costs of the move. A worker would be expected to migrate when the present value of the benefits outweighs the costs. Ideally, one would like to have discounted values of costs and benefits, to capture the forward-looking concept. However, Gallin (1999) argues that most existing empirical studies of net migration ignore this aspect and thus do not properly identify the underlying parameters of the theoretical model. This can be thought of as an omitted variable bias since no explicit measure of expected future labour market conditions is included. Gallin (1999) proposes that future migration should be included in the migration model to control for expected future labour market conditions. The author in turn proxies the expected future net migration rate by using a one-step-ahead forecast formula. 44 See Goetz (1999), Greenwood (1997) and Vaithianathan (1995) for a more detailed discussion of the range of migration theories. 48 While the human capital approach facilitates a structural view of the migration decision, based on the rational decisions by migrants, a simple human capital approach fails to take into account the interdependence of the migration decision and other decisions. For example, a decision to migrate may well be taken together with the decision to change jobs 45 . This leads us to an extension of the simple human capital approach, as discussed below. Observed wages may also be a poor proxy for the attractiveness of a region if potential migrants take into account their chances of getting a job. Higher wages are not the only monetary gains from migration. It has been argued that differential economic opportunities consist of two parts – the expected wage rate and the probability of receiving that wage. Harris and Todaro (1970) argue that regional wages should be weighted by the respective probability of employment. Both the unemployment rate (Day (1992)) and the employment rate (Treyz, Rickman et al. (1993)) have been used as proxies for the probability of getting a job, in order to capture the trade-off between wages and the probability of receiving the wage. It is hypothesised that other things being equal, areas with high unemployment rates (low employment rates) should have both more (less) out-migration and less (more) in-migration. The spatial job-search theory views migration as a result of a job-search process. In other words, an individual decides to migrate when he or she accepts a job at a location different from his or her current location. Under this theory, a worker will continue searching for other jobs until a wage offer either equals or exceeds the reservation wage. The reservation wage is determined to equate the marginal cost of obtaining one more job offer with the expected marginal return from continued search. Reservation wages and thus search duration would depend upon one’s personal characteristics as well as attributes of local labour markets that affect wage distributions. Reservation wages and search duration are also likely to be affected by the extent of unemployment insurance, and the amount of labour market information. Greenwood, Mueser et al. (1991) argue that individual decision makers focus on a manageable subset of potential destinations, as well as on a manageable subset of characteristics. This idea of a spatial choice set provides a new direction in migration research both in its own right and because it has the potential to be implemented with laboratory experimental techniques. Herzog, Schlottmann et al. (1993) provide a literature review of empirical studies that considered labour migration as a spatial jobsearch. In summary, migration responds not only to income gains, but more generally to differential economic opportunities between regions. In fact, economic opportunities have been proxied by a wide range of labour market variables such as regional differentials in the employment rate, unemployment rate, wages, labour force participation rate, new-hire rates, lay-off rates, and various other labour market conditions (see Greenwood (1997) for examples). Although the migration model is usually formulated in the context of individual decisionmaking, the data employed in estimating the models were for many years aggregate in nature. For example, the influence of wages on migration is tested using average wages in origin and destination areas. However, this is really asking whether migration occurs from low-to-high wage areas. Alternatively, if one had access to individual data (in this case, migrant-specific wage measure), then one would be able to determine whether migrants themselves benefit by moving (consistent with the individual utility maximisation framework). The latter situation is important if we were interested 45 Vaithianathan (1995) provides a summary of the criticisms of the human capital approach. 49 particularly in the personal characteristics of migrants (as examined in the third key question). This aggregate versus individual data issue will be discussed again later, when we talk about how researchers control for the heterogeneity amongst people. A further problem from using current labour market conditions to explain current migration flows, is that causality may well run in two directions. Although some migration models include employment growth as one of the explanatory variables, employment growth can as easily be caused by migration. One way that has been used to address this issue is to include lagged measures of local labour market conditions in the model. For example, Maré and Timmins (2000) regress the migration flow over the 1991-1996 (1986-1991) period on the values of the explanatory variables as at 1991 (1986). Similarly, all the independent variables were lagged one period in two of the equations in Day (1992). She then compares the estimates from these equations to estimates obtained from equations using only current values of the independent variables, to test the hypothesis whether individuals do have fairly accurate information about conditions in other regions. Another approach is to model migration and local labour market conditions jointly, rather than modelling only migration in response to wages or unemployment (the latter being a partial equilibrium approach) (see for example Gallin (1999)). When choosing a measure of labour market attractiveness, it is important to note that only changes in relative attractiveness are expected to induce migration flows. Changes that improve the attractiveness of all locations will not necessarily change relative attractiveness. Although different regions may experience different regionspecific shocks at different times, they also face some aggregate shocks from time to time. An example of an aggregate shock is national slumps and booms. The distinction between the aggregate and region-specific shocks has been highlighted as an empirical issue in Section 2.4. Available evidence seems to indicate that migration declines during national slumps and rises during recoveries (see Molho (1984) and Gordon (1985), as cited from Greenwood (1997), and Milne (1993)). There are of course other fiscal and macroeconomic influences that may constitute an aggregate shock (see Milne (1993)). If aggregate shocks explain a large proportion of the variance in regional labour market variables, then it would make sense to control for such aggregate shocks by specifying the relevant labour market variables in relative terms – for instance, relative to the national average. This effectively “filters out” the effect of aggregate shocks (i.e. those that impact all regions across the board), so that what is left is the influence of regionspecific or idiosyncratic shocks. Examples of studies that specify regional labour market variables as deviations from national means include Blanchard and Katz (1992), Decressin and Fatas (1995), Jimeno and Bentolila (1998), Fredriksson (1999) and Debelle and Vickery (1999). Most of these authors do perform tests to examine the validity of the assumption that a significant proportion of the variance in regional labour market variables can be explained by aggregate shocks. For example, Blanchard and Katz (1992) ran a regression, for each state, of the change in the logarithm of state employment on the change in the logarithm of national employment 46 . This test basically tries to answer the question of how much the typical movement in state employment is common to all states (i.e. due to an aggregate shock) and how much is state-specific. Their results generally suggest that much of the year-to-year movements in state employment are 46 For a more detailed discussion of the test that Blanchard and Katz (1992) used, see pages 25 and 26 of that article. 50 accounted for by movements in aggregate employment. Therefore, it makes sense to analyse regional labour market variables expressed as deviations from the corresponding national means (so that the effect of aggregate shocks is “purged”). Interpreting labour market changes We note also that even where a plausible measure of relative labour market attractiveness exists, it is not a straightforward matter to interpret what is causing changes in those measures. Essentially, there is an identification problem in that changes may arise due to changes in labour demand, or labour supply. These changes have different expected impacts on migration behaviour, but are difficult to identify separately. Let’s discuss a labour demand shock first. Successful regional economic development policy (a positive local labour demand shock) increases labour demand in a local region. This increase is expected to reduce unemployment, increase labour force participation, and induce in-migration. As for a labour supply shock, there could be at least two important changes in a particular region’s labour supply. Let’s discuss a positive supply shock. One is rising educational attainment and skills of the labour force, amongst both new entrants and existing workers. The second is an increase in in-migration of individuals, possibly because the region has become perceived as a more attractive place to live (for whatever reason). This could be the case, for example, if a disproportionate number of overseas immigrants settle first in the Auckland region, from where they may then subsequently migrate to other regions. In the real world, both labour demand shocks and labour supply shocks may be occurring. How do we distinguish between them? These shocks induce different regional labour market adjustments. Therefore, it would be important to identify which type of shock is causing the changes. This identification process requires a combination of theory, judgements, and empirical patterns 47 . Most researchers have in fact focused on region-specific labour demand shocks, particularly in studies employing the Vector Autoregressive technique (see Section 5.2) (e.g. Blanchard and Katz (1992), Decressin and Fatas (1995), Jimeno and Bentolila (1998), Fredriksson (1999), Debelle and Vickery (1999)). These researchers associate unforeseen movements in employment with innovations in labour demand (i.e. equating a shock to employment to a labour demand shock). Blanchard and Katz (1992) argue that this assumption is approximately correct since they find that positive shocks to employment increase wages and reduce unemployment. Therefore, one can draw inferences from these studies (focusing on labour demand shocks) to provide some insights to the effects of regional economic development policies. There are of course studies that focus on local labour market adjustments in response to a labour supply shock (see for example, (Hanson and Slaughter forthcoming, 2001) and references cited therein). While we are on the topic of shocks, there is a distinction between a once-and-for-all (or temporary) shock and a shock with subsequent adjustments (permanent shock). To elaborate, Bartik (1991) considers a one-and-for-all shock to local job growth, with subsequent growth unchanged from what it would have been. For example, a local area has been continuously experiencing job growth of about 3 percent per annum, and suddenly is hit by a positive once-and-for-all shock, raising the job growth rate to 5 47 This is one example of the identification problem that has been highlighted previously at the beginning of Chapter 2. 51 percent for that year. The following year, the area’s job growth rate returns to the normal rate of 3 percent. On the other hand, Blanchard and Katz (1992) considers the effects of a permanent shock, that is, the one-time shock to local job growth is allowed to affect subsequent growth. In our numerical example just now, the area’s job growth rate would be 5 percent in the years subsequent to the one-time shock (i.e. it does not return to the normal rate of 3 percent). The effects on the dynamics of the variables of interest from the simulation exercise would be different under the two scenarios (see Bartik (1993)). This is an issue that researchers have to consider. Where there is a known labour market shock Sometimes, we can identify a labour market shock more clearly and easily. In such cases, one could use case studies or event studies to trace implications of the known labour market shock. Also, case studies are appropriate when a local area is unique in terms of what attracts migrants, and its regional adjustment process. A particular study worth highlighting here is by Klier (2000). This study looks at the migration response when auto plants open in a US county. In particular, they estimate the impact of auto plants on county-level net migration during the 1980s and 1990s. Meanwhile, shocks in New Zealand that are recognisable include industrial closedowns such as the Mosgiel Textile Mill and Southdown and Patea Freezing Works closures. Peck (1985) reviews published material on firm closures in New Zealand since 1980. Table A in Peck (1985) lists textile plant closures over the period 1980 to 1983 which have been reported through searches or in published reports. Peck (1985) highlights the main issues identified in the literature. It sets out the issues perceived as being important by the local community and outside agencies, and records ways in which individuals and communities adjusted to the closures. The kind and extent of local and central government intervention and support provided, and any known effects, are also recorded. Another background source for potential case studies would be Le Heron and Pawson (1996). Measure of migration The second main ingredient in uncovering a labour market story is the migration flow. What is the right measure of migration flows? As mentioned in Section 3.1, there is a range of alternative migration measures. The migration figure can be expressed in levels or as rates, gross versus net flows, migration effectiveness index, migration intensity index, etc. As alerted earlier, these measures are appropriate for different purposes. Therefore, for each of the key questions, the particular measure(s) used may vary. Different researchers have looked at modelling migration differently. For example, the gross-migration model focuses on the individual migrant’s decision-making or choice process (e.g. Poot (1986b) and Juarez (2000)). It attempts to directly identify the determinants of migratory behaviour. On the other hand, the net-migration model focuses on the equilibrium-seeking tendency of the population system. It abstracts from the processes that induce migration flows that cancel one another (e.g. Pissarides and McMaster (1990) and Treyz, Rickman et al. (1993)). Others have started to model individuals’ probability of making a move (e.g. Enchautegui (1997) and Molho (1987)), which is discussed further below when we talk about the use of individual data in estimating dichotomous choice models. Others have even used multiple choice models to allow for more than two alternative choices (e.g. Day (1992), Antolin and Bover (1997) and Böheim and Taylor (2000)). 52 Now that we have covered the two main ingredients to infer the labour market story – labour market attractors and migration flows, let’s look at how researchers “mesh” the ingredients together. In a perfect world, with identical regions and identical individuals In a perfect world where regions are the same (except perhaps in labour market conditions), and people are identical, then one could use descriptive techniques to uncover the labour market story, such as simple correlation analyses and cross tabulations. These descriptive analyses are able to control in a fairly crude way for the characteristics of migrants and/or of migration origins and destinations (“pull” and “push” forces) that make some individuals more likely to migrate. Descriptive analyses can be presented for homogeneous sub-populations (for example, by age) and show if there is a correlation between migration rates and labour market conditions defined for that subgroup (e.g. age). A general indication from studies of this type is that migration behaviour is far from homogenous and that there are a host of reasons for people’s moving decisions. For example, the young and more highly skilled tend to have higher propensities to move. Besides, some regions (e.g. those with high welfare benefits) may attract people with certain characteristics (e.g. the unemployed or low-skilled ones). Descriptive studies that ignore these forms of heterogeneity, and rely instead on aggregate patterns can be misleading. Descriptive studies, however, do not control for other influences, and do not provide precisely the quantitative nature of the relationships. Nevertheless, they help suggest explanatory variables that should be included in more formal multivariate analyses of migration behaviour. More formal studies basically control for a range of other influences, each of which are discussed below. Modelling the relationship between migration and labour markets Before discussing the way that researchers have dealt with non-labour market differences across regions, differences across individuals, and the impact of non-labour market forms of adjustment, we will first discuss a common way of modelling the systematic variation in migration rates that serves as a benchmark against which labour market influences can be measured. This common method is the use of a gravity relationship. If one were to simply look at the number of individuals migrating, it is almost certain that more populous areas such as Auckland will have a larger numerical migration flow than less populous areas such as Gisborne or the West Coast. Scale differences need to be controlled for. One possible treatment is to model migration rates (i.e. dividing the migration flow by the size of the population in either the origin or the destination), rather than migration levels. For example, Pissarides and McMaster (1990) models the net migration rate of region i, calculated as the number of persons that move as a percentage of the population in region i. Alternatively, some studies rely on the model itself to deal with this scale difference. One such model is the gravity model. The gravity model provides an appropriate functional form that treats the scale difference by including population size variables to proxy for the attractiveness of each area. At the heart of the model is a simple relationship derived from physics (see Box 1). Its attractiveness stems in part from its 53 simplicity, and in part from the fact that it requires only aggregate area data. In particular, gravity models can be estimated using data of a form that is available in many countries – a single cross-section with a retrospective question on location in a prior period (usually one or five years earlier). Box 1 Gravity models Basic Gravity Model The most simple gravity models relate the flow of people from area i to area j (M ij ) to the attractiveness of each area, as proxied by population size (P i and P j ), and the distance between the areas (D ij ). The form of the relationship is based on the formula for the Newtonian law of gravitation 48 , where force of attraction is proportional to the product of the masses of the two bodies involved and inversely proportional to the square of the distance between them: 321 α αα ijjiij DPGPM = Different definitions/measures of the size of the masses There are two masses in the model; one for the transmitting region, and the other for the receiving region. One interpretation of the former mass can be the size of the population in the transmitting region, or the migrating population. The mass of the receiving region can represent the population size there, the size of the region’s economy (Gross Regional Product), or some labour market variables such as the number of jobs (see the discussion on modified gravity models below). Different definitions/measures of distance Depending on the circumstance, different measures of distance may be appropriate. Some common measures or interpretations of the distance variable include the simple linear distance between two points, travel time, transport or travel cost, the cost of relocation (e.g. removal cost). For example, Day (1992) used the forgone wage cost of moving from region i to j, and the cost of transporting oneself and one’s belongings from i to j, while Bukenya, Schaeffer et al. (1999) or the actual distance variable between i and j. Other studies have included more general measures of distance: social distance, political distance, and psychological distance or psychic cost (e.g. the cost of leaving behind relatives and friends). The pure gravity model is nowadays rejected on numerous grounds (Poot 1986b). For example, it is not based on a behavioural theory, and does not predict very well. In addition, since the model explains a flow (i.e. migration) in terms of stocks (i.e. population size), the timing of measurement of the stocks is an important issue. Besides, it (unrealistically) assumes that the migration flow between regions i and j is not affected by characteristics of the other regions in the system (i.e. it ignores the substitution effect) 49 . Furthermore, there will be a simultaneous equation bias if the population masses are themselves very sensitive to changes in the migration flow. Some (but not all) of these criticisms against the pure gravity model are addressed by incorporating some modifications, as discussed below. Modified Gravity Model The basic gravity model is often extended by adding variables that have a behavioural content, typically characteristics of the origin and destination areas, ( 21321 ββ α αα jiijjiij XXDPGPM = ), and additional variables that are expected to influence the decision to migrate. Commonly added attributes include income or wage rate, unemployment rate, weather, degree of urbanisation, housing prices, heating costs, various measures of public expenditures and/or taxes and many others. The exponents of these additional variables in the model generally indicate the "attractiveness" of area attributes in the sense of making the area a good place to leave (β 1 ) or to move to (β 2 ). There have also been criticisms of the modified gravity model (see for example, Schultz (1982), as cited in Greenwood (1997)). For example, Schultz argues against the common use of the double logarithmic functional form to estimate modified gravity models. The argument is as follows. Since the regions of any country differ significantly in population size and land area, a larger share of all moves will tend to occur within the boundaries of larger regions. Therefore, more non-migration will appear to exist for such regions. The consequence is that non-migration is spuriously correlated with origin population size and land area. Other extensions to the basic gravity model include adding individual characteristics, amenities in the origin and destination areas, as well as public or national policies. These are discussed in the text. 48 The Newtonian Law of Gravitation replaces M ij with "gravitational force", and constrains α 1 =1, α 2 =1, α 3 =-2. 54 When researchers get to the empirical stage of their study, they need to decide on how they actually want to estimate their gravity models. The first and simplest way is to take the logarithms of the gravity model and use the OLS estimator. However, this method suffers from two drawbacks, as discussed in Bailey and Gatrell (1995, pp. 352-358) 50 . Alternatively, one can perform the maximum likelihood estimation of parameters under more realistic distributional assumptions. A commonly used distributional assumption used is the independent Poisson distribution (see Bailey and Gatrell, 1995). A more detailed discussion of the use of these procedures in gravity modelling can be found in Fotheringham and O’Kelly (1989), Sen and Smith (1995), and Isard, Azis et al. (1998). As elaborated in Box 1, the basic gravity model as used in the migration context depicts the relationship between the number of migrants from area i to j, the size of the population in areas i and j, and the distance between areas i and j. The basic gravity model is often extended by adding variables that have a behavioural content, typically characteristics of the origin and destination areas, as well as additional variables that are expected to influence the decision to migrate. The question of what variables to include in the migration equation have been reviewed extensively and thus, we do not have the intention of providing a comprehensive coverage 51 . For the purposes of answering our key question, we note that measures of relative labour market attractiveness can be included as variables in a gravity model. Other variables can be added to control for non-labour market differences across regions, for differences in demographic composition, and for other forms of adjustment. The sort of variables that are commonly used for these purposes are discussed in subsequent sections of the paper. In estimating the gravity relationship, one can use either aggregate (or area) data, or individual data. This distinction essentially separates migration studies into those that concentrate more on the characteristics of places (areas of origin and/or destination) and distance between areas as determinants of migration (a macro-level approach), and those that focus on characteristics of individuals 52 (rather than on streams of aggregate movement) that influence migration behaviour (a micro-level approach) 53 . We shall discuss first the studies that use aggregate data (i.e. the macro-level approach). Aggregate gravity studies In the New Zealand context, an early study by Giles and Hampton (1978) estimates an out-migration and in-migration simultaneous equation model of urban migration, using 49 Poot (1986b) incorporated such interdependencies in a system of spatially separated labour markets. 50 Firstly, there is no guarantee that flows predicted from the model, when summed over origins and destinations, would agree with the observed totals leaving origins and arriving at destinations, and that the total predicted ‘cost of travel’ would agree with that observed. Secondly, estimating parameters by the ordinary log-linear regression model would only be justified statistically if the flows were independent and log-normally distributed about their mean value with a constant variance. Such an assumption is patently not valid since flows are discrete counts whose variance is very likely to be proportional to their mean value. 51 See in particular Greenwood (1997) and Goetz (1999). 52 In practice, migration is often a family decision. This creates some complications, which are discussed in Section 2.4. 53 Empirical studies on migration can also be subdivided into time series studies (e.g. Pissarides and McMaster (1990) and Treyz, Rickman et al. (1993)) as well as cross-sectional ones (e.g. Saltz (1998) and Goetz and Debertin (1996), as cited in Goetz (1999)). More recently, some studies have used pooled time series and cross-sectional data (sometimes known as panel data) to obtain more reliable and precise coefficient estimates than either times series or cross-sectional models by themselves (e.g. Antolin and Bover (1997)). 55 aggregate data. The “push” and “pull” forces they included in their in- and outmigration equations include the road distance from Auckland, per capita income in the origin area, and a dummy variable for urban areas in the South Island. A few other variables, such as the average temperature and the population size of the areas, were initially included but were omitted later because they did not perform very well (as cited in Vaithianathan (1995)). A later study by Poot (1986b) used aggregate data to examine inter-urban migration for males between the main urban areas between 1971 and 1976. The dependent variable used was the number of individuals migrating from region i to region j, and the explanatory variables included a range of characteristics of region i and region j. These area characteristics can be thought of as “pull” and “push” forces. Examples of area characteristics he used include labour force participation rates, the percentage of people aged over forty years, average rainfall, average temperature, and the proportion of non-Europeans in the region. Maré and Timmins (2000) recently carried out a multivariate analysis of migrationlabour market links over the period 1986-1991 and 1991-1996. The data were obtained from the New Zealand census. They estimate three different forms of gravity models using gross flows between regions. The first one models a simple gravity relationship (population size in origin and destination regions, and distance), using regional data. A second model adds labour market and demographic attributes, as measured as at the beginning of the period in order to avoid problems of endogeneity. The labour market variables that are included are the unemployment rate, the employment rate, and median income. Demographic variables include the proportion of the region's population that are in various age, ethnicity, and qualification groups. Finally, the authors also add three variables capturing how similar each pair of regions is. The variables that they use are Duncan dissimilarity indices for ethnic composition, one-digit industry, and one-digit occupation. The coefficients indicate whether people are more likely to move to other areas that are similar to their own, or to ones that are different. One might expect people to move to similar regions because they would be more likely to fit in, because their knowledge, information and behaviours would be more in line with those in their destination. Individual (micro-level) data gravity studies Alternatively, one could also use individual data in estimating gravity models. The availability of individual data allows researchers to estimate gravity-type relationship using limited dependent variable methods such as the logit and probit models, which are designed to deal with models where the dependent variable takes on a small number of values – in this case an indicator for whether someone moves or not. This is discussed in more detail when we talk about how researchers control for composition effects, as a result of heterogeneous individuals. When we conduct empirical analyses to highlight the labour market story, we have to “filter” out the “nuisance variation” that is not part of the labour market story. These include (a) differences between regions in terms of their climate, amenities, and cost of living; (b) differences amongst people in terms of characteristics (e.g. age, education, etc.) and preferences (e.g. highly skilled people prefer regions with higher wages), and allowing for other (non-migration) forms of adjustment. The way researchers have gone about filtering out these “nuisance variation” is the gist of the rest of this section. 56 Dealing with non labour market differences across regions Now that we have covered the ways to address the scale differences between areas, let’s get back to our objective of uncovering the labour market story. As alerted earlier, there are differences across areas – in labour market and non-labour market factors. The non-labour market differences across areas may offset or reinforce the influence of labour market conditions on migration (i.e. the labour market story). These then are considered “nuisance variation” from our perspective. Cost of living differences One example is differences in the cost of living across areas. We discussed earlier that most studies examining the determinants of internal migration have used some economic price variable such as wages. Other things being equal, workers tend to migrate toward areas whose cost-of-living-adjusted wages are higher. However, some studies ignore the nominal income versus real income debate 54 , either because of a lack of data about cost of living, or because it is assumed that perceived differences in purchasing power are small (as cited from Poot (1986b)). Researchers can control for cost of living differences in at least two ways. One is to calculate a cost-of-living- adjusted wage or income level to be included in the migration equation (e.g. Poot (1986b)). This approach basically uses a measure of “real” wages based on full adjustment for cost of living. The second way is to include directly a proxy for the cost of living (usually the Consumer Price Index) in the migration equation itself (e.g. Juarez (2000)). This latter approach is a partial regression-based cost of living adjustment to wages or income, that lies between the no adjustment (i.e. using nominal income) and full-adjustment approaches (i.e. using “real” income) (as cited from Dumond, Hirsch et al. (1999)). In addition, including cost of living and income separately allows one to test for the significance of each variable. An important, and perhaps the largest component in the cost of living is housing costs 55 . The theory is that ceteris paribus, housing cost differentials between regions have a negative effect on migration. In other words, other things being equal, the higher the house prices or housing costs in a particular region, the more in-migration is discouraged. How do studies incorporate housing cost? Some studies use house price differentials (e.g. Antolin and Bover (1997), Cameron and Muellbauer (1998), Papps (2000), and Kennedy and Borland (2000)), while others use housing rent (e.g. Bukenya, Schaeffer et al. (1999)). Differences in amenities Another example of a non-labour market difference between regions that can affect the labour market story is amenities. The focus on the labour market story means that we are interested only in differences in regional labour market factors that induce migration flows. However, since people move for variety of reasons – labour market and nonlabour market reasons – we have to remove the effects of non-labour market differences between regions such as amenities, which may have an influence on migration flows. 54 Dumond, Hirsch et al. (1999) looks at whether cost of living matters for wage differential studies. In this paper, various versions of cost of living adjustments are made to wages and the results are then compared. 55 The discussion here focuses on housing costs as a proxy for the cost of living. The impact of housing tenure (e.g. home ownership versus rentals) on migration flows will be dealt with later, when we discuss about the other forms of adjustments. 63 important for migration decisions. Secondly, numerous studies have highlighted the connection between housing tenure and migration. For example, Hughes and McCormick (1987) find that other things being equal, council tenants experience much greater difficulty than owner-occupiers in actually fulfilling their migration intentions (as cited from Johnes and Hyclak (1994)). This relationship is not only being borne out by specific countries. Oswald (1999) finds a positive relationship across countries between levels of home ownership and unemployment rates, suggesting that the fixed costs of home ownership discourage migration. Other studies that have examined the link between migration and the housing market include Cameron and Muellbauer (1998), Böheim and Taylor (2000) and Johnes and Hyclak (1994). Commuting is often an alternative to migration. Cameron and Muellbauer (1998) argue that labour market conditions should influence commuting rates in the same direction as migration rates. However, the effect of relative house prices on commuting should operate in the opposite direction on migration. In other words, commuting is a way of overcoming housing market impediments. Therefore, Cameron and Muellbauer (1998) model regional commuting and migration choices in a common framework. Since these other forms of adjustments have either a direct or indirect effect on migration, it may be worthwhile modelling them in some way. There are at least three general approaches to doing this. The first approach is to basically add covariates representing them in the single-equation migration model. For example, one can add the relative house prices variable, or the proportion of homeowners in a region, in the migration equation. A second and more elaborate approach is to use a multi-equation system or simultaneous equation models. This approach recognises that many variables are changing simultaneously, and thus one needs to reflect the simultaneous determination of a range of variables and the feedback and interaction amongst them. Earlier simultaneous equation models were developed during the 1970s to explain the determinants and consequences of migration (see for example Muth (1971) and Greenwood (1975b), as cited from Greenwood (1997)). Another example is the study by Gordon and Lamont (1982, as cited from Crampton (1999)) who used an 11- equation simultaneous model, which has as endogenous variables the following: inmigration and out-migration under three separate streams, net commuting, employment growth, unemployment, house building and owner-occupied house prices. An overview of these approaches is provided in Mueller (1982, as cited from Greenwood (1997)). However, there has been relatively little further development in this area during the past one-and-a-half decades (Greenwood (1997)). A study worth mentioning here is by Fry, Fry et al. (1999), who used data from 1982 to 1996 to estimate a structural econometric model of inter-regional migration in Australia. The net migration equation itself is quite straightforward. The explanatory variables include lagged migration, time trends, real wage differentials, unemployment rate differentials, and housing price level differentials. The migration equation is, however, embedded in a fuller model of the economy, which allows the authors to compare model predictions allowing for a migration response with predictions excluding that response. A third approach is to use a case study approach on known shocks, such as the closure of the Patea Freezings works in 1982 and the state-sector restructuring 64 Reefton in the late 1980s 68 . For example, Patea, which is a small town situated in Taranaki, lost its only substantial industry when the Patea Freezing works closed in September 1982. A total of 720 people were employed by the works, of which 348 were residents of Patea. When a small town loses its only substantial industry, the effects of that loss can be far-reaching. A few reports were produced to monitor and examine the impact of the closure upon the Patea community, in terms of numbers likely to leave, their characteristics, and the effect on business, sports, community, cultural and other organisations, and on health and educational, and religious services in the town. These reports include a review report by John Martin dated 15 December 1984, and two interim monitoring reports dated 4 October 1982 and 17 January 1983, which are attached as appendices to the 1984 report (see Martin (1984)). Although these reports have slightly different estimates (due to the timing difference), the main conclusions are essentially the same. For those cases, the researcher can focus on how the shock affects not only the labour market, but also the housing market, the education sector (e.g. teachers leaving), and the transport market, just to name a few. One may well find that for particular subgroups, there are certain barriers to migration occurring, such as the inability to sell their property, or being tied to their “homeland”. There are tradeoffs between the different approaches. While a multi-equation approach allows the researcher to take into account the various interdependencies and feedbacks amongst a set of variables, it often requires that the individual equations in the model to have relatively simpler specifications and fewer regressor variables, compared to when we use a single equation model only. For example, in Johnes and Hyclak (1994)’s four-equation model (wage inflation, unemployment rate, net migration and house price inflation), the net migration equation only included the wage level and house prices of both transmitting and receiving regions. This is relatively parsimonious compared to Poot's (1995) migration model, which had up to 18 regressor variables. Meanwhile, a case study approach allows one to look at cases with a known (or easily identified) shock and trace through the impact of that shock. However, the results from that case may not be generalisable. In summary, because regions and individuals exhibit different characteristics, studies have to control for a whole lot of things before they are able to confidently say something about the labour market story. However, most people would agree that it is not possible to perfectly control for everything. Another approach to uncover the labour market story, which arguably has fewer things to control for (i.e. fewer “nuisance variation”), is to ask migrants directly what motivates their migration decisions. This is the topic of our next subsection. 4.2.2 As revealed by people when asked about the connection Studies that have tried to detect the labour market story by asking people directly usually ask about people’s intentions. Some of them focus on people who moved and ask for factors that have led them to make their migration decision. Meanwhile, other studies include in their sample also those who had not moved. The researcher might also pose some “speculative” questions to these people who had not moved. The questions posed basically attempt to reveal people’s judgements and decision-making process. For example, the respondent may be asked whether he or she would have moved if he or she were given an additional supplement (e.g. $1000)? 68 See Le Heron and Pawson (1996) and Peck (1985) for other ideas of possible case studies. 65 An example of a questionnaire that allows such a direct approach is the output from the internal migration component of the US Current Population Survey. This type of data allows researchers to analyse the determinants of internal migration in a way previously impossible. Examples of studies that have examined the reasons for geographical mobility based on survey-based questionnaires include Lansing and Mueller (1967) and Bartel (1979) (as cited from Greenwood, Mueser et al. (1991)). Bartel (1979) examined the role that job mobility plays in the migration decision. She used three data sets that encompass different age groups to demonstrate the importance of the relationship between job mobility and migration at different points in the life cycle. The data sets are the National Longitudinal Survey (NLS) of Young Men, NLS of Mature Men, and the Coleman-Rossi Retrospective Life History Study. The NLS of Mature Men, for example, provides information on whether a move was undertaken for economic reasons or personal (for instance, family, health) reasons. A more recent example is Böheim and Taylor (2000) who used data from 1991 to 1997 from the British Household Panel Survey (BHPS) to investigate the reasons for moving house and the extent and determinants of house moves. The BHPS is a nationally representative sample of some 5500 households recruited in 1991, containing approximately 10000 persons. These same individuals are interviewed each successive year (i.e. longitudinal data). The questionnaire collects information about income, labour market status, household composition and consumption, education and health at each interview, as well as information on employment changes that have occurred within the period between interviews. Respondents are also asked about their intentions to move house and reasons for that 69 . Panel data such as these are really suitable for the study of migration, because they provide detailed information on individuals and households before and after any move. In the New Zealand context, such survey-based data are scarce. An early study that attempted to assess the motivations for moving at the individual level is Barrington and Davey (1980). However, this study focused on international out-migrants only. It is worth noting that Statistics New Zealand is currently considering the idea of having a survey explicitly for investigating the reasons for moving, both internally within New Zealand and overseas (as discussed in section 1.2.2.3). One possibility is to append a set of questions to an existing survey to collect the necessary data 70 . If this eventuates in the future, then researchers can take advantage of the methodological and theoretical work already done in the US context. Such data would enable researchers to examine what motivates migration decisions (i.e. this key question), as well as the characteristics of movers and the non-mobile population (i.e. the third key question). However, even if this project eventuates, the data would not be available in the near future. Therefore, it is likely that the first approach (i.e. inferring motivations from actual mobility patterns) is going to prevail for a while. Another paper in New Zealand that attempts to uncover individuals’ motivations to move is by Heenan (1999). This paper uses autobiographical details of residential mobility for a small sample of New Zealand male career professionals collected in 1981. Classified by cohort and employment sector, the information is used to examine 69 The residential mobility questions asked at each annual interview are “If you could choose, would you stay here in your present home or would you prefer to move somewhere else?”, “What is the main reason why you would prefer to move?”, and “Can I just check, have you yourself lived in this (house/flat) for more than a year, that is before September 1 st , 199{0,1,2,3,4,5,6}?” (as cited from Böheim and Taylor (2000, p. 12)) 70 The contact person at Statistics New Zealand for this proposed project is Mansoor Khawaja (email: [email protected]). 66 migration in the context of personal career development. Aspects explored include age of entry to the workforce, frequency of movement, migration and career promotion, and the spatial structure of migration. Apart from date of birth and career related details, the questionnaire did not explicitly request information on life-cycle events and related matters. Therefore, this paper focuses on a narrow segment of longitudinal or life history mobility in New Zealand. However, as is true for everything else, there are pros and cons. One criticism against this approach is that people may not have conceptualised their motives or find the questions difficult to articulate. Also, researchers have to rely on people’s reported reasons for migration. Another criticism is that the sample size is often much smaller than census data (for example, there were only 33 usable responses in the autobiographical study by Heenan (1999)). Nevertheless, it does give us directly a potentially useful perspective on migration motives. 4.3 Findings This subsection summarises some of the evidence in relation to the first key question under two headings. The first looks at the evidence from studies employing the indirect way of inferring the labour market story. The second covers the findings from studies using the direct approach to determine specific motivations of individual migrants. 4.3.1 As revealed by actual mobility patterns We begin by providing a picture of whether the labour market story is borne out using descriptive techniques, for OECD countries and New Zealand. We then discuss some of the findings when researchers control for the “nuisance variation” as discussed in the previous subsections. The question is whether the basic relationship (i.e. labour market story) changes when these additional controls are made. Table 4.1 below provides an international comparison of the relationship (correlation) between net in-migration flows and unemployment rates. In almost all the countries71, the sign of the relationship is of the expected sign (negative), and of these, the correlation coefficient is statistically significant. The evidence generally implies that low-unemployment regions will experience a net inflow of people from regions with higher unemployment, hence supporting the existence of a labour market story. Table 4.1 Correlation between net migration flows and unemployment by region, for OECD countries 71 The two exceptions are Czech Republic and Portugal, but the correlation coefficients for these two countries are not statistically significant. Source: OECD Employment Outlook (2000, p.56) 67 The stylised facts of the labour market story above are in relation to unemployment rates. A logical question to ask is whether this is true for other labour market indicators. The answer is generally yes. For example, Blanchard and Katz (1992) point out that the correlation of state employment growth and net migration rates in the United States is 0.84 for the 1950-1987 period, and 0.91 for the 1970-1987 period. Again, the evidence seems to support the presence of a labour market story. Now that we know how the labour market story holds up in other countries, one might ask whether this evidence can also be found in the New Zealand context. We have performed some preliminary tests to investigate the presence and significance of a labour market story for migration in New Zealand. A good starting point would be to examine simple correlations between selected labour market indicators and each of three migration flow measures (inflows, outflows and net flows) at different levels of spatial disaggregation, as summarised in Figure 4.2. Figure 4.2 Correlations of migration flows with labour market indicators Labour Market Indicators (86-91) Against Net Migration (91-96) 0 0.1 0.2 0.3 0.4 0.5 AU TLA RC Correlation Coefficients Emp Rate Pop Share Partic. Rate Income A Labour Market Indicators (86-91) Against In Migration (91-96) -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 AU TLA RC Correlation Coefficients Emp Rate Pop Share Partic. Rate Income B Labour Market Indicators (86-91) Against Out Migration (91-96) -0.2 -0.1 0 0.1 0.2 0.3 AU TLA RC Correlation Coefficients Emp Rate Pop Share Partic. Rate Income C Notes: The data points enclosed in boxes are significant at 5% AU = Area Unit, TLA = Territorial Local Authority and RC = Regional Council. The indicators shown in Figure 4.2 are the employment rate, the participation rate, the share of national population, and per capita income72. For each of these indicators, we would expect a positive correlation with net migration and inflows, and a negative correlation with outflows. There are significant positive correlations with net migration and inflow rates at the area unit and TLA levels. The correlation at the regional council level, while positive, is not significant. The expected negative correlation with outflow rates holds only at the area unit level. 72 The patterns of per capita income differ from the patterns of median income, as noted earlier. 68 The observed patterns suggest that the link between labour market conditions and net migration is due more to a link with inflow rates than with outflow rates. We speculate that people may leave locations for non-labour market reasons, but their choice of destination is more closely linked to labour market prospects. The exception is that people appear less likely to leave area unit that are doing well. The relationships at regional council level are not significant, suggesting that internal migration may be aiding labour market adjustment more at the sub-regional level. The descriptive evidence thus seems to confirm that there is a connection between migration and relative labour market conditions, although it is clearly not the only factor driving migration patterns. We turn now to findings from studies that attempt to model the relationship more formally. Modelling the relationship between migration and labour market conditions We noted above that the relationship between migration and labour market conditions could be modelled either with aggregate (area level) data or with data on individuals. When we look at the findings in the literature, we find that this difference matters. By and large, studies using aggregate data show less consistent results, especially when testing the link between unemployment and migration. Studies using data on individuals (unit record data) are more likely to confirm the importance of labour market considerations in influencing migration flows. It seems that the relatively better performance of unit record studies is due to the fact that it is possible to incorporate much better controls for factors that can bias aggregate studies. Most of the studies using aggregate data focus on the characteristics of places, and thus fail to detect the traits of individuals (such as educational attainment and age), and their influence on the decision to migrate. This often results in insignificant coefficient estimates for their explanatory variables including labour market variables (see Greenwood (1985), as cited in Goetz (1999)). It is worth summarising first the general results from gravity models, in terms of which variables are usually significant with the correct sign, and which have mixed results. Generally, distance and size come out as important. In many studies, the coefficient of distance is consistently negative, which suggests that on average, people prefer to move short distances if at all (see for example, Day (1992) and Bukenya, Schaeffer and Gebremedhin (1999)). This is found to be true regardless of the proxy used – the forgone wage cost of moving from region i to j, the cost of transporting oneself and one’s belongings from i to j, or the actual distance variable between i and j. Evidence of the deterrence effect of distance on migration is also borne out in New Zealand studies (e.g. Poot (1986b) and Maré and Timmins (2000)). Therefore, it is important to include distance when trying to uncover the labour market story. Focusing on the labour market story, Poot (1986b) finds that labour market variables are indeed important. These variables have the right sign and are significant. In summary, out-migration is a function of the labour force size, real income, employment growth, age composition, average temperature and an opportunity cost index. Meanwhile, in-migration is a function of nominal income, employment growth, population size, and a relative attractiveness index. Poot (1986b) finds that the labour force participation rate and the percentage of individuals over forty years at the origin area act as a “push” force. Meanwhile, the “pull” forces include the real income and employment growth in the destination area. People tend to migrate from low real income to high real income and high job growth regions. 69 However, as pointed out by Vaithianathan (1995), the direction of causality is unclear for these economic factors. Employment growth may as easily be caused by migration. The other variables such as the average rainfall, average temperature, and the proportion of non-Europeans in the region were not found to be significant (see Vaithianathan (1995)). An earlier study by Giles and Hampton (1978) found that there is evidence of a clear interdependence between in-migration and out-migration. Regional income is found to have an influence on in-migration, but not on out-migration, thus acting as a “pull” force only. The authors also found that the closer an area is to Auckland, the greater the rate of in-migration and the lower the rate of out-migration. The population in the South Island also appears to be less mobile (see Vaithianathan (1995)). In contrast, a more recent paper by Maré and Timmins (2000) finds that the evidence for a labour market influence on migration flows at the regional council level is weak at best. Few of the coefficients are significant, and the pattern of signs is not stable. The labour market variables included are the unemployment rate, the employment rate, and median income. The paper estimates different forms of gravity models applied to data on regional councils in New Zealand. The authors carry out a multivariate analysis of migration-labour market links over the period 1986 to 1996. The next question is whether inflows are more responsive to labour market variables than outflows, or are they equally responsive? A common finding amongst earlier gross migration models in the US and Canada is that income (and job) opportunities can better explain in-migration than they do of out-migration (Shaw, 1985, as cited from Greenwood, 1997). In other words, there is a paradox that origin economic conditions do not matter (for out-migration) but destination economic conditions do matter (for inmigration). Similarly, a recent study by Shen (1999) also finds a strong “pulling” effect of rapidly growing regions to migrants, but the “pushing” mechanism is not operating effectively in inter-regional migration. However, this paradox has been challenged by several researchers (see Poot (1986b)). One explanation put forth for this apparent paradox is that the use of population migration as a proxy for labour migration underestimates the sensitivity of workers to economic conditions in the origin (that is, it dilutes the labour market story). Consistent with this view, by focusing on male workers only, Poot (1986b) finds that employment growth is significant in both the inmigration (a “pull” factor) and out-migration (a “push” factor) models. However, not all labour market variables perform as well. For example, we noted earlier that the relationship between unemployment and migration is the most fragile. Examining estimates of this relationship paints the least flattering picture of what aggregate studies can tell us, but provides a useful illustration of some of the differences between area level and unit record studies. Empirical studies that have examined the influence of area unemployment rates on migration have found mixed evidence. The hypothesis is generally that areas with high unemployment rates should have more out-migration and less in-migration, all else equals. However, some obtained significant coefficients with the anticipated signs, while others found insignificant coefficients and/or unanticipated signs. Several possible explanations have been put forth for this mixed evidence. However, until microdata were available to test these explanations, they remained as speculative hypotheses (Greenwood, 1997). 70 The importance of microdata is highlighted in the literature review by Herzog, Schlottmann et al. (1993) dealing with the relationship between personal employment status, migration and the likelihood of re-employment. Without microdata, personal characteristics have to be proxied by mean characteristics. In general, results from migration studies using aggregate data may not be all that clear. These studies find that labour market variables (commonly entered in levels) end up with insignificant or perverse coefficients, particularly for unemployment. Such findings are not surprising since the aggregate variables may be virtually uncorrelated with the individuals’ traits of concern. For example, in a net-migration model, the results showing a negative sign on the regional unemployment differential variable could mean at least two things. First, the regional unemployment differential could be proxying for differences in employment probabilities in the regions. Therefore, people are moving out of the region because there are fewer employment prospects there. Second, unemployed workers are more likely to move out than employed ones. If a region’s unemployment rate rises relative to the unemployment rate of other regions, the region’s net migration falls. Not many studies have estimated the relationship at both the individual and area levels, although there are exceptions - Navratil and Doyle (1977) provide a good case for discussion because they estimated both aggregated and disaggregated models. These authors find a significant regional unemployment variable in the aggregate model, but not in model using disaggregated data. Vaithianathan (1995) opines that the reason for observing such results is because in the disaggregated model, where the personal and regional unemployment is separated, the regional unemployment variable becomes insignificant while the personal unemployment status is significant. It is noted that without individual data, it is not possible to separate out these aggregate and individual effects. Similarly, other studies subsequent to Navratil and Doyle (1977) that employ microdata to look at the impact of personal unemployment on the likelihood of migration mostly find that personal unemployment significantly augments migration likelihood (see literature review in Herzog, Schlottmann et al. (1993)). In other words, personal unemployment encourages migration to another labour market. However, this effect varies in magnitude across countries and subgroups of the population by race, gender, occupation, ethnicity, and prior geographic mobility. Besides, the stimulative effect of joblessness on migration falls with search duration (Herzog, Schlottmann et al. (1993)). The evidence generally indicates that labour migration is micro-efficient, by responding to personal unemployment. The evidence also generally suggests that labour migration is macro-efficient, since higher area unemployment rates induce labour outmigration. Some researchers such as Antolin and Bover (1997) have gone further by distinguishing between persons that are registered as unemployed at the Official Employment Office in Spain, and the unregistered unemployed. The reason for this distinction is that registration not only has a direct negative effect on the probability of migration, but it also alters the effect of regional unemployment. This implies that a higher than average unemployment in the individual’s region will only have a positive effect on the probability of migration if the person is a non-registered unemployed, but will have an important negative effect if the person is registered. In short, migration from a high unemployment area to a low unemployment area is valid only for the nonregistered unemployed, at least in Spain. Meanwhile, Juarez (2000) who used gross migration models, concludes that a high unemployment rate in a region tends to increase out-migration because unemployed 71 people search more actively than the employed. However, this is qualified in a sense that for unemployment rates greater than a certain threshold, there is a relative reduction of this effect. A high unemployment rate in a region also tends to discourage in-migration, but without the same qualification. Gallin (1999) argues that some authors have found weak or contradictory evidence for the relationship between unemployment and migration, because they ignore the role of expected future labour market conditions (an omitted variable bias) (see for example, Greenwood, 1969; Fields, 1979; and Pissarides and McMaster, 1990; as cited from Gallin (1999)). By controlling for future conditions, Gallin (1999) finds significantly different results. For example, in a model without unemployment, the strength of the relationship between current wages and net migration is less than half as large. On the other hand, in a model with unemployment, the effect of current wages is larger and the effect of current unemployment is smaller. The key message from the evidence thus far is that there is a labour market story. But it is not the only story. As discussed earlier, there is “nuisance variation” that may distort the labour market story. One could end up finding no evidence of a labour market story when there is in fact one, or over-estimating the relationship between migration and labour market when there is none or a weak one only. That is why we need to control for the “nuisance variation”. The question of how much the “nuisance variation” actually biases the labour market story is an empirical one. The findings from studies that have tried to control for these sources of “nuisance variation” (that is, differences across regions and across individuals) are discussed next. Dealing with differences across regions This subsection summarises the evidence from studies that have controlled for non labour market differences across regions. We then ask whether these controls do in fact matter. However, this is not asking whether the controls (for instance, for amenities) influence migration, but whether the controls alter the estimates of the labour market story. Unfortunately, this question is difficult to answer since most studies that incorporate such controls do not have the labour market story as their primary focus. For example, while Day (1992) finds, amongst other things, that individuals are less likely to move to a province where the chance of being unemployed is higher, ceteris paribus, the primary focus of her paper is whether government policies affect provincial migration. Therefore, the robustness of the labour market story according to the model chosen (aggregate gravity model versus logit versus probit versus multinomial logit) cannot be directly commented upon. Controlling for amenities Studies that control for amenities are primarily interested in the impact of amenities on migration. However, the question we would really like to answer is whether controlling for amenities changes the labour market story. Ideally, we would like to have examples of studies that begin with labour market variables alone, and gradually expand the explanatory variables to include amenities. This way, we can work out how controlling for amenities affects the labour market story. Unfortunately, most of the studies do not look at the effect of labour market variables with and without amenities. Instead, their primary focus is on amenities. For example, an early study of the influence of locationspecific amenities on migration by Graves (1979) demonstrates that when income levels and unemployment rates are accounted for, climatological amenity variables are 72 important in explaining age- and race-specific net migration during the 1960s (as cited from Greenwood (1997)). Therefore, it is difficult to infer about the impact that the inclusion of amenities has on the labour market story. Nevertheless, it is still worth examining the stock of evidence with respect to the significance of amenities, as covered in the next few paragraphs. In general, studies find that amenities are important. Numerous studies have been published on the implicit valuation of amenities (see references in Goetz (1999) and Greenwood (1997)). As an illustration, Clark and Knapp (1995) find that professors at all ranks attach a value to the amenity of sunshine and the dis-amenity of violent crime. Meanwhile, Day (1992) finds that other things being equal, most people prefer warm climates to cold ones. However, many studies find less support for the amenities hypothesis. Greenwood and Hunt (1989) re-examined Graves’ amenity hypotheses and findings, and found that employment opportunities are far more important in explaining metropolitan migration compared to location-specific amenities (as cited from Greenwood (1997)). For a more recent study, see Clark and Knapp (1995) (as cited in Goetz (1999)). A relationship has been found between amenities and labour market variables (e.g. Elhorst (2000)) and separately between amenities and migration flows (Goetz, Ready and Stone (1996) as cited from Goetz (1999)). Although we cannot infer from these facts whether and how the exclusion of amenities bias our estimates of the relationship between labour market variables and migration (i.e. the labour market story), it would be prudent at least to test for any such bias by including amenities measures in migration models. In summary, amenities are generally important in influencing migration. Although we do not know for sure whether, and the extent to which, they change the labour market story, the evidence nevertheless serves to prompt researchers to control for them. Cost of living differences The evidence of the significance of the cost of living component has been quite mixed. For example, Juarez (2000) finds that the coefficient on relative prices (origin relative to destination) in the migration equation is statistically significant and has a positive sign. This suggests that other things equal, migration flows are from regions with high prices to those with lower prices. In contrast, Antolin and Bover (1997) finds that cost-of-living differentials as measured from the Consumer Price Index are not significant, but house price differentials are significant with the expected sign. They conclude that people in regions with higher-than-average house prices have a higher probability of moving. Meanwhile, when Bukenya, Schaeffer et al. (1999) included housing rent in the migration equation to proxy for the role of living expenses, they found that the variable exhibits a statistically significant positive effect at origin, but not the expected negative effect at destination. They conclude that these results do not show any effect of living expenses on migration. Public policy effects Based on the literature survey by Charney (1993), the evidence in general suggests that some public policies have a positive impact on migration (as well as income and employment), some have a negative impact, while some others have shown mixed results. For example, education tends to attract migrants and in some studies diminish 79 other a priori information, are generally called structural models 77 . We will begin with the approaches that use a short time series. 5.2.1 Short-time series approaches The minimum time variation in order to be able to say anything about an adjustment, is one time window (i.e. a snapshot of the world at two points in time). The basic idea is to estimate transition (migration) probabilities as a function of labour market conditions and then possibly simulate. We then compare the base-line results with the simulation outcomes to infer the importance of migration. At least two methods have been used – simulations of gravity models and the Markov modeling approach. An example of the use of simulation methods applied to a gravity model is provided by Nijkamp and Poot (1987). However, it is noted that simply taking the gravity model and simulating it may not give meaningful results. More care is needed in specifying the form of gravity model if it is going to be used for simulation purposes (see Nijkamp and Poot (1987)). A Markov or transition matrix contains the probability that an individual will remain in the region for another period and the probabilities that the individual will migrate to each of the other regions in the system. These probabilities are usually based on past rates. According to Rogers (1968, as cited in Isserman, Taylor et al. (1986)), the empirical results of using Markov chains to project future population totals have been disappointing because of “the restrictive assumption of unchanging movement probabilities”. Attempts to adjust the transition probabilities (i.e. to use endogenous probabilities) have been shown to lead to greater accuracy (see references cited in Isserman, Taylor et al. (1986)) 78 . Another approach that uses a short time span is the labour market accounts technique. The labour market accounts method is basically a decomposition technique, to analyse inter-relationships between changes in labour supply and labour demand. In particular, this technique addresses the question of what the relative contribution of natural change, participation change, and net in-migration are to labour supply changes, and whether job shortfalls are translated into increases in unemployment and/or net outmigration. The analysis can be performed both through time and across space. In some circumstances, it may be worthwhile to focus on certain local areas with similar experiences of change, to allow the main types of experience of labour market change to be identified. An illustration of this technique is provided by Green (1994) and Beatty, Fothergill et al. (1997). Green (1994) examines the role of migration in bringing labour supply and labour demand into balance (or in mitigating balance) in Britain in the 1980s. Meanwhile, Beatty, Fothergill et al. (1997) explores how different coalfields and individual districts adjust to external pressures in the UK coal industry during the 1980s and early 1990s. Although the approaches above can shed some light on the regional labour market adjustment process, a short time span limits the extent of analysis one can perform. More importantly, these approaches do not explicitly model (i.e. have a simplistic treatment of) the other forms of adjustment, such as changes in the labour force 77 However, nowadays, there is a lot of emphasis on the “middle way”, namely structural VAR models (where theory dictates the restrictions imposed). Two recent examples include Balakrishnan and Michelacci (2001) and Leeves (1999). 78 The advantages and disadvantages of the Markov modelling approach are discussed in Nijkamp and Poot (1987). 80 participation rate, and wage levels. Therefore, one cannot say much about the contribution of migration to the full adjustment mechanism. This brings us to the next set of approaches that have the luxury of longer time series. 5.2.2 Longer time series approaches With a longer time series, there is much more information that we can infer from the data about the role of migration. In fact, one is more likely to be able to assess the contribution of migration to the full adjustment mechanism. There are generally two ways that we can exploit the longer time series. The first one does not impose any prior structure on the data set, that is, it allows the data to speak for themselves. Conversely, the second is to take advantage of any a priori information we have, to impose some structure on the relationships amongst the set of variables of interest. This is usually done by incorporating insights from theory. Both have been used in examining the second key question. Under the first approach, which imposes little prior structure or none at all on the data set, one can focus on the correlations amongst a set of variables (the “system”), introduce a shock to the “system” and then trace the impact of this shock. Using such an approach, the researcher needs to select the appropriate variables to include in the “system”. Some researchers have looked at labour market variables alone, and some have used a broader range of variables. We shall cover both in turn. Using time series correlations (VAR) As it turns out, the largest strand of the literature examines the second key question by analyzing the correlation amongst a set of labour market variables. A convenient modelling technique in this context is the vector autoregressive or VAR model. The VAR technique models the systematic co-variation amongst the selected set of variables, and uses this to get predicted paths for all of the variables. In other words, this technique examines the joint fluctuations of the selected set of variables over time. This then forms the baseline results. Box 2 provides an illustration of the VAR model, as adopted by Blanchard and Katz (1992), Decressin and Fatas (1995) and Debelle and Vickery (1998)). Box 2 Illustrative VAR equations Blanchard and Katz (1992) use three equations involving labour market variables, with lagged values on the right-hand side (except for the second and third equations, where the first variable is allowed to feed through simultaneously into them). The first variable (em) is the natural logarithm of employment in the region as a proportion of total employment. The second variable (ur) compares the region’s unemployment rate to the national unemployment rate. It is defined as the logarithm of the ratio of the region’s “employment rate” to the national “employment rate” (where the “employment rate” is one minus the unemployment rate). The third variable is the logarithm of the ratio of the region’s participation rate to the national participation rate. In each case, the variable is a ratio between the region’s and national levels, hence a relative performance of the region (the rationale is discussed in the text). The exact specification of the VAR would depend on the time series properties of the variables of interest. Without going into too much detail, it is sufficient to say that if a variable is found to be integrated of order one (I(1) or non-stationary), it has to be modelled in first difference, whereas a variable that is integrated of order zero (I(0) or stationary), is modelled in levels. 81 jt n ssts n ssts n sstsjjt jt n ssts n ssts n sstsjjt jt n ssts n ssts n sstsjjt prurempr pruremur pruremem 3 1,3 1,3 0,3,3 2 1,2 1,2 0,2,2 1 1,1 1,1 1,1,1 εδδδα εδδδα εδδδα ∑∑∑ ∑∑∑ ∑∑∑ =− =− =− =− =− =− = − = − = − +∆+∆+∆+= +∆+∆+∆+= +∆+∆+∆+=∆ The variables above are typically defined as follows:         =         =         =        − =         = population lforce/totalabour total region in population/region in forcelabour ln rateion participat national region in rateion participat ln forcelabour /totalemployment total region in forcelabour /region in employment ln ratent unemployme national-1 region in ratent unemployme1 ln employment total region in employment ln jjj j pr jjj j ur j j em Some studies also include a wage equation in the VAR model (see for example Debelle and Vickery (1999)). For a closer look at the application of this technique, see Blanchard and Katz (1992), Debelle and Vickery (1998), and Debelle and Vickery (1999). In order to see how the set of variables in the VAR respond to some “shock” (allowing for lagged effects), the researcher can obtain the impulse response function. The impulse response function (IRF) traces the impact of a shock on all the variables in the VAR over time. Using the IRF, the researcher basically introduces a shock to, say, the level of employment in one period and uses the same model of systematic variation to get an alternative predicted path. When compared to the base-line results, one can then infer about the relative importance of each variable and the time taken for adjustment. In short, the VAR technique can be used to address the following questions. When hit by an adverse shock to one of the regional variables, how has the typical region adjusted (i.e. what changes and by how much?)? Specification issues for VAR In using the VAR technique, there are some specification issues that need to be addressed. Some of these specification issues are similarly treated across different studies (e.g. type of variable included, as discussed below), whereas other issues have been handled in different ways by various researchers. These issues are discussed below. The first specification issue relates to the choice of variables. In particular, the researcher needs to determine the type and number of variables one should include in the VAR model. Given that our objective is to uncover the importance of a labour market story, the natural set of variables would be labour market variables, such as the unemployment rate, wage level, level of employment and the labour force participation rate. However, there is no universal agreement as to which variables are the most appropriate – it will depend on the circumstance. For example, there are problems with using the unemployment rate in aggregate migration models, as noted by Greenwood (1985) and Fields (1979) (as cited from Poot (1995)). Instead, these authors note that layoff rates, quit rates and unemployment duration, where available, may be better 82 measures of labour market conditions. Furthermore, what variables enter the VAR can be constrained by what data is available. For example, wages are excluded from the VAR in Jimeno and Bentolila (1998) due to the lack of quarterly data. Similarly, some studies (e.g. Blanchard and Katz (1992)) may choose to leave out one or two variables from their VAR because including them may reduce the size of the sample (i.e. at least one of the series has a smaller sample size) and/or would introduce additional righthand-side variables, resulting in too few degrees of freedom. Ideally, it would be great to have a migration variable in the VAR model. However, data on migration flows is either not available on a quarterly basis, or the sample period for which such data is available is not sufficient for formal estimation work. Therefore, migration is not directly estimated in studies that have used the VAR model. In order to overcome this problem, most researchers calculate migration as a function of the other variables that are modelled. In other words, an estimate of migration flows is backed out of the model based on the time paths of the other variables in the VAR. Different studies have done it (slightly) differently. The following is the approach taken by Blanchard and Katz (1992), Decressin and Fatas (1995) and Debelle and Vickery (1998)). Based on the three equations shown in Box 2 above, migration (as a percentage of the working-age population) can be derived from the identity equation below: Migration = d(em) – d(ur) – d(pr) This method essentially equates the movement in the labour force to the migration of workers. This method is based on the assumption that most of the changes in the labour force are due to migration, rather than to natural population increases (which arguably are relatively stable, particularly if one uses quarterly data). This assumption is usually tested with some diagnostic test to check if the assumption is reasonably valid (see for example, Blanchard and Katz (1992)). Maryanne Aynsley’s research essay uses a slightly different way to infer the role of migration, but based on similar principles as above (see Aynsley (2001)). She treats the movement in the working age population as characterizing worker migration flows. The relative importance of migration as a labour market adjustment mechanism is then assessed by estimating two models. The first model assumes that the working age population is exogenous and hence there is no mobility of labour (given her treatment above). Therefore, adjustments in response to a shock have to be in the form of changes in employment, unemployment, labour force participation and/or wages. The second model allows for labour mobility (by making the working age population endogenous) and thus comparing the two models would indicate the role of labour mobility as a labour market adjustment. Although one can choose to extend the VAR model to include non-labour market variables, most studies that use the VAR technique in the literature on regional labour market adjustment do in fact use labour market variables only. These studies often use information on how different labour market series such as unemployment, wages, employment and labour force participation change together over time in response to some shock, allowing for lagged effects. In other words, the technique examines the joint fluctuations of the selected set of labour market variables over time. In terms of the use and interpretation of the impulse response function, the VAR technique addresses the following questions. When hit by an adverse shock to employment (for example), how has the typical region adjusted? Did the shock mainly produce a decline in relative wages, an increase in joblessness, a decrease in the labour force participation rate, and/or an increase in labour migration? 83 As mentioned above, there are not many examples of VAR studies that have extended the Blanchard and Katz (1992) work to incorporate a broader range of variables (i.e. non-labour market variables). One such example is the study by Fredriksson (1999) that takes into account active labour market programs in Sweden. Using the same scenario in the previous paragraph, where we asked how the typical region adjusts when hit by an adverse shock to employment, there is a contention that the presence of some active labour market programs may have adverse effects on the regional adjustment process. Active labour market programs could have lock-in effects, that is, people who would have migrated in the absence of programs choose to stay on in a depressed region (following an adverse shock to employment). The reason for this is that active programs decrease the unattractiveness of joblessness (either by offering higher compensation or improving subsequent re-employment probabilities), thus reducing the incentive to migrate to other regions with more favourable employment opportunities. Furthermore, active labour market programs may also prevent discouraged workers from exiting the labour force. In other words, a labour market program could be a substitute for out-migration and labour force participation changes. This is only one example of the incorporation of a broader range of variables in the VAR. Conceptually, one can include other appropriate variables provided that there is a justifiable reason to do so and there are sufficient data available. As mentioned earlier, in applying the VAR technique, there is a range of specification issues that researchers have to consider. Up to here, we have discussed the choice of variables to include in the model. The other specification issues include the stationarity of the series, the definition of shocks, the distinction between temporary and permanent shocks, aggregate and region-specific shocks, and the way to address permanent differences across areas and population heterogeneity, as well as the appropriate level of disaggregation (e.g. regions or sub-regions) 79 . Each of these is discussed in turn. Given the set of variables selected, an issue arises as to whether the variables should be modelled in levels or in the first difference form. As a precursor to answering this question, and thus using the VAR technique, one needs to determine the time series properties of the variables of interest. In particular, one needs to test whether the underlying stochastic process generating the series is invariant with respect to time, that is, stationary 80 . This is because variables entering the VAR have to be stationary. If the variables in the VAR are nonstationary, but not cointegrated (i.e. it is not possible to construct a stationary linear combination of the variables), then the correct procedure is to estimate the VAR in the differenced form 81 . Meanwhile, if the variables are nonstationary but are cointegrated, a vector error correction model (VECM) can be constructed. The VECM is a VAR of the differences variables with, in addition, a vector of residuals of the cointegration regression 82 . For example, if relative unemployment 79 There are of course other specification issues not discussed here, for example, the choice of lags. 80 There are two ways to test for stationarity. The first is to use a descriptive approach. For example, one can calculate correlation coefficients (between current and past levels) and plot persistence diagrams to assess whether regions have experiences sustained differences in the labour market variables of interest. Some of these persistence measures have already been illustrated in Section 3. A second and more formal method to test whether a series is stationary or otherwise, is to use unit root tests, such as the Augmented Dickey-Fuller (ADF) test (see for example Blanchard and Katz (1992) for an application). For a review of the concept of stationarity and other important concepts in time series modelling, see Enders (1995) and Gujarati (1995). 81 Another rationale for first differencing the variables in the VAR is to account for fixed effects, as discussed further below. 82 For an introduction to VAR, VECM and the concepts of stationarity and cointegration, see Enders (1995) and other standard (time series) econometric textbooks. 84 rates are found to be stationary, they enter the VAR in levels. Conversely, if there is evidence that relative employment contains a unit root (i.e. non-stationary), then one should model the series in first difference form (i.e. employment growth) (see for example Blanchard and Katz (1992)). In some circumstances, it may be worthwhile to try different specifications, particularly if the unit root tests provide mixed or inconclusive evidence. Jimeno and Bentolila (1998) provides such an example. Their first specification models the unemployment rate and labour force participation rate in levels, and employment in first difference. Meanwhile, their second specification models all three variables in first differences. Another specification issue facing researchers employing the VAR technique is the choice of identifying assumptions 83 . Although this has been discussed previously in Section 4.2, it is worth noting again the typical approach used by researchers to define shocks in VAR-type studies. What most researchers (e.g. Blanchard and Katz (1992), Decressin and Fatas (1995), Jimeno and Bentolila (1998), Fredriksson (1999), and Debelle and Vickery (1999)) have done is to basically associate unforeseen movements in employment with innovations in labour demand (i.e. equating a shock to employment to a labour demand shock). Blanchard and Katz (1992) argue that this assumption is approximately correct because they consistently find that positive shocks to employment increase wages and reduce unemployment. An issue related to the above is the distinction between a once-and-for-all (or temporary) shock and a shock with subsequent adjustments (permanent shock). As pointed out in Section 4.2, the type of shock analysed has quite different implications for the evolution of the variables of interest over time (see Bartik (1993)). The distinction between aggregate and region-specific shocks has been discussed in Sections 2.4 and 4.2. Recall that we are ultimately interested in shocks that can change the relative attractiveness of locations. A region-specific shock certainly qualifies for this. On the other hand, aggregate shocks that impact proportionately on all regions do not change relative attractiveness of locations. Meanwhile, aggregate shocks, to the extent that they impact disproportionately or unevenly on some regions, need to be accounted for 84 . The issue of controlling for such aggregate shocks is particularly important when shocks are predominantly aggregate in nature (i.e. a large proportion of the variance in regional labour market variables could be explained by aggregate shocks). Researchers usually test whether this is indeed the case (see for example, Blanchard and Katz (1992)). If aggregate shocks are found to be significant, then researchers control for them by specifying the regional variables as deviations from the national means. Most studies employing the VAR technique in this area in fact tend to define their variables as being relative to their national counterparts (see for example Blanchard and Katz (1992), Decressin and Fatas (1995), Jimeno and Bentolila (1998), Fredriksson (1999) and Debelle and Vickery (1999)). Most studies employing the VAR model attempt to control for permanent differences across regions that do not necessarily induce migration flows. The permanent difference arises because some variables (e.g. local participation rate and employment rate) reflect unobserved “fixed effects” of local areas. For example, unobserved aspects of an area’s demographics or amenities could affect the participation rate, such that some areas will have permanently higher relative participation rates while others 83 Bartik (1993) and Blanchard and Katz (1992) provide a good summary of this issue. 84 As discussed in Sill (1997), regions may be affected by the same economic shock in different ways and with some variation in the timing of a response to shocks (i.e. leading and lagging regions). One likely reason for such differences is industry mix. 85 have permanently lower relative participation rates. If these unobserved aspects are correlated with an area’s job growth, then ignoring them will lead to biased estimates of local job growth effects on local participation or employment rates. Similar permanent differences could arise for other reasons. Researchers can use at least two ways to avoid the bias. One is to include fixed effects controls (i.e. region-specific constants) in the estimation (see for example Blanchard and Katz (1992) and Jimeno and Bentolila (1998)). The second way is to eliminate fixed effects by first differencing all level variables. Under this second method, the estimation then considers how changes in local employment cause changes in local labour markets (e.g. changes in local participation or employment rates, as above). Besides, the VAR studies above have examined regional labour market adjustment in relation to the labour force as a whole. It may be worthwhile taking a closer look at specific labour force groups (for example, high-skilled versus low-skilled, or females versus males). For instance, one may expect that in response to an adverse shock, the highly skilled are more likely to migrate than remaining unemployed or dropping out of the labour force, compared to the low skilled. An example is provided in the Spanish context by Mauro and Spilimbergo (1998) and Mauro, Prasad et al. (1999). These studies look at five educational groups (ranging from illiterate to the college-educated) and estimate how rapidly different groups of workers respond to regional labour demand shocks. Another issue that is important to note is that some researchers have used a different level of disaggregation. For example, Maryanne Aynsley’s research essay (as mentioned above) looks at labour market adjustment mechanisms in Australasia (Australia and New Zealand as two regions). In other words, her work focuses on adjustments at the national level, rather than the regional or state level. It is thus important that we eventually highlight how her work complements ours given the different units of analysis (national vs regional). Up to here, we have discussed what the VAR technique does, some specification issues, and cited some relevant references. We shall next briefly talk about the advantages and disadvantages of using this technique. One of the advantages of the VAR technique is that it does not require much theoretical structure or a priori information 85 . This is particularly so in cases where there is much uncertainty over the structure of any behavioural relationship governing migration patterns. Another advantage of this method is that it is possible to trace the impact of a "shock" to any of the variables over time. For instance, if there were an employment shock, the initial effect would be represented as an increase in ε E . In the following period, E -1 would be higher, which would lead through to wages and unemployment, which in turn would feedback to other variables over time. The figure below shows an example of an impulse response function, as extracted from Blanchard and Katz (1992). 85 This is arguably a strength as well as a weakness of the VAR technique. 86 Figure 5.1 Impulse Response to a 1 percent Negative Employment Shock (in the US) It is worth giving an illustration of how the impulse response function above is interpreted. The impulse response results selected for illustration are obtained from Blanchard and Katz (1992). According to Figure 5.1 above, when a state experiences a one percent adverse shock to its employment, in the first year, there is an increase in the unemployment rate of 0.32 percentage points and a decrease in the labour force participation rate of 0.17 percentage points. These estimates are the initial impact. Over time, the effect on employment builds up, and reaches a peak of –2.0 percent after four years and a plateau of about –1.3 percent. The effects of the negative employment shock on unemployment and participation steadily fall and disappear after five to seven years. However, it is important to note that the VAR approach, as for other approaches, have a few limitations. Firstly, VAR is only a partial modeling approach, rather than a system modeling approach (examples of the latter to be discussed further below). Secondly, it usually assumes a symmetric response to shocks – a positive shock in one variable leads to the same response in the set of variables as a negative shock, except in opposite direction. There is a debate about the validity of this assumption. Thirdly, VAR-type studies are of an aggregate nature, in the sense that it picks up the relationship for a typical region. However, different regions may well adjust in different ways to the same shock. The latter limitation can be addressed by using other approaches such as case studies, as discussed below. 87 Case studies approach The VAR technique analyses correlations from the data (labour market variables alone, or both labour market and non-labour market variables) imposing little prior structure on them. Another approach that does not impose much theoretical structure is to use case studies. Case studies are particularly suitable when there are some aspects that cannot be measured easily, but nevertheless are important. Under such circumstances, instead of restricting the analysis so that the VAR technique can be used, some researchers have chosen to using case studies to capture the qualitative aspects. Some of these unique qualitative aspects are region-specific, and thus regional case studies might be appropriate. Furthermore, case studies allow researchers to focus on places with big and/or identifiable shocks. In the New Zealand case, Reefton is a clear example. Reefton is a small town of 1100 people in the West Coast. During the long boom, its main industries were farming, coal mining and forestry. Reefton received support from the state for these activities, and the provision of infrastructure to develop the economy there. There were a series of state-sector restructuring in that small town in the late 1980s. David Conradson briefly summarises the impact of this restructuring, focusing on the employment paths of people rendered redundant (see Le Heron and Pawson (1996)). Conradson examined the question of what adjustments were undertaken by people made unemployed within that town. Another case study that is worth highlighting is the closure of the Patea Freezing works in 1982. As discussed previously in Section 4.2, there was a rapid rise in the number of registered unemployed immediately following the closure in September 1982 (i.e. the initial impact). This figure went up to a peak of about 300 people in late October 1982 86 . However, the situation stabilised from November 1982. Again, this case study allows one to ask what the significant adjustment mechanisms were in response to the shock identified. Up to here, we have discussed approaches that basically address the key question without imposing a prior structure on the data (i.e. let the data speak for themselves). We shall next discuss approaches that impose some structure on the data, usually obtained from theories and other a priori information. Structural modelling approach Just to refresh the reader’s memory, as discussed in Section 2.2, when a region experiences a shock, the region can adjust in various ways, for instance via the labour, housing, capital, goods and services markets. The previous discussion of the VAR technique has summarized how some researchers have attempted to uncover the regional labour market adjustment story. These can be broadened to account for the other forms of regional adjustment that are ignored but are expected to be important. Two examples are given below. One example would be to take into account regional adjustments in the housing market. There are at least two reasons for considering the housing market when we look at the role of migration. First and more importantly, relative house prices are important for migration decisions. Housing cost forms a significant proportion, if not the largest, of households’ expenditures. A shock to a region can often affect the house prices simultaneously with the labour market. For example, a positive shock may lead 86 See Le Heron and Pawson (1996) and Peck (1985) for other ideas of possible case studies. 88 to a rise in relative wages and employment opportunities (which encourages migration into a region), and a corresponding increase in house prices (which discourages inmigration). Therefore, some authors (see below) have argued for endogenising the determination of house prices. Secondly, numerous studies have highlighted the impact of housing tenure upon migration. Some researchers have argued and presented evidence that the lack of flexibility in the housing market can hamper migration of workers. This can take the form of higher transaction costs on the purchase and sale of residential real estate (e.g. via taxation), long minimum duration of rental contracts (for whatever reason), public allocation of rental housing, etc. Studies that have argued and found evidence supporting the importance of the housing market include Johnes and Hyclak (1994), Cameron and Muellbauer (1998), Böheim and Taylor (2000) and Oswald (1999). The other example would be to recognize that commuting 87 is often an alternative to migration. As explained in the previous paragraph, the housing market can be an impediment to the migration process. It is often argued that commuting is a way of overcoming such housing market impediments (see for example Cameron and Muellbauer (1998)). This is because if house prices increase concurrently with employment opportunities and wage levels as a result of a positive shock in a region, individuals from other regions can still work in that region while commuting back and forth to their original location (instead of migrating). Therefore, it may be appropriate to model regional commuting and migration choices in a common framework, as done by Cameron and Muellbauer (1998). One approach that captures the interactions between labour market and non-labour market variables that are expected to be simultaneously determined in the system (e.g. housing prices and/or commuting, as discussed above) is the structural model. A structural model gets its name from the fact that its form is given by the underlying theory or behavioural relationship governing migration patterns. This approach contrasts with the VAR technique, which is a reduced form approach 88 . A structural model is usually more complicated, and often involves multiple equations in the model. One could have a small structural model with a few equations only. An illustration of such a small structural model that incorporates a migration equation is provided by Johnes and Hyclak (1994). These authors used a four-equation model which has as endogenous variables the following – wage inflation, unemployment rate, net migration and house price inflation. The endogenous housing and labour market variables are therefore simultaneously determined, hence emphasising the linkages between the two markets. Alternatively, one can go further and allow for a large set of feedbacks, resulting in a structural model with many equations, and sometimes many regions (i.e. multi-regional structural model) 89 . However, this approach has a cost. It would have to be done at the expense of each equation in the model becoming simpler. In other words, although we are taking into account the interactions between different equations, we will have to settle for more straightforward individual equations. 87 Green, Hogarth et al. (1999) provides a review of the trends, issues and some implications of longer distance commuting as a substitute for migration in the UK. 88 The reduced form approach is preferred by some because of the uncertainty over the structure of the behavioural relationship. 89 See Bolton (1985) for an early comprehensive survey of structural econometric models. 95 important a role as predicted, for reasons such as housing, the lack of mobility of unskilled workers and Maori links to their homeland, as discussed in Section 4.3. The findings from the case studies can be summarized as follows. In Reefton, there has been a great diversity of personal experiences in response to the restructuring in Reefton. Similarly for Patea, the effects of the closure were not only felt on labour market, but also on the housing market, education sector (teachers leaving, etc.), transport sector, etc. Such diversity of adjustment mechanisms in a region may not be picked up by standard models, which inevitably have structures imposed on them. Furthermore, the case studies offer researchers an opportunity to analyse the impact of unique shocks occurring in unique localities. 5.3.3 Structural modelling approach The VAR-type studies and the case studies approaches discussed thus far impose little prior structure, if any, on the data. In contrast, structural models are more complicated, often involving many equations in the model. The question is whether we gain anything further (beyond the VAR or case studies approach) in terms of understanding the role of migration as a regional labour market adjustment mechanism. Johnes and Hyclak (1994) used a four-equation model of the labour market – wage inflation, unemployment rate, net migration and house price inflation. They looked at the Southeast region of the UK, and find that house prices does indeed play an important role in the short run regional labour market adjustment, by exerting a strong (negative) influence on the movement of labour. A number of other studies have also looked at the link between the housing market (housing tenure and house prices) and migration (e.g. Cameron and Muellbauer (1998), Böheim and Taylor (2000) and Oswald (1999)). These studies tend to find that homeowners have relatively lower mobility rates, and that high relative house prices discourage net migration to a region. A more extensive model by Fry, Fry et al. (1999) find that when the results from their structural econometric model of inter-regional migration are used to re-specify and calibrate their CGE model, it provides a more detailed picture of labour market responses when they simulate the response of net inter-state migration to changes in State Government spending 91 . However, the authors suggest that more work is needed to help understand whether the observed differences are significant in practice. Another quite extensive model is by Groenewold (1997) which provides an illustration of the use of a 24-equation structural model to examine the importance of migration as a form of labour market adjustment in Australia. The author finds that inter-state migration and relative wage movements play a role in labour market adjustment, but they do not eliminate inter-state unemployment differentials. The adjustment following a shock is slow, with only half of the adjustment complete after 20 quarters. 5.3.4 Other approaches The discussion thus far is on the findings from approaches that use longer time series. It is worth noting also the findings from approaches that use shorter-time-span data sets, such as simulations of gravity models and Markov switching techniques (see Nijkamp and Poot (1987)), as well as the labour market accounts technique. 91 This is a realistic shock designed to disturb unemployment rates and house prices, which turn out to be significant and have the correct signs in the migration equation (this evidence of a labour market story has been highlighted in Section 4.3). 96 The labour market accounts technique has been used by Green (1994) and Beatty, Fothergill et al. (1997). Green (1994) finds (among other things) that although in the short term, net out-migration may not be an immediate response in some areas to a deterioration in labour demand, the effect of long term economic decline in stimulating out-migration (and depressing in-migration) is evident. The study also finds that local areas respond over different time periods, and in different ways. Similarly, Beatty, Fothergill et al. (1997) find that there is diverse range of labour market adjustment processes, and that there is no single model of labour market adjustment that is applicable across a range of localities. Unemployment can arise in markedly different ways in terms of changes in migration, commuting, labour force participation, job loss and job creation. 5.4 Concluding Remarks The methods used to examine the question of whether migration is a significant form of regional labour market adjustment include approaches that use short time series and longer time series. Approaches of the former type include simulations of gravity model, the Markov transition method and the labour market accounts technique. Meanwhile, approaches that utilise longer time series can be divided into those that do not impose a prior structure on the data, and those that do. Those that impose little structure or none at all on the data include VAR-type studies and case studies. In contrast, those that do impose some structure, whether it is based on theory or some other a priori information, are generally called structural models. The crucial question is whether there is a difference in the results of these various approaches. How much do we gain (in terms of a better understanding of our key question) by adding further complexities to the model? If there is not much of a difference, then we would of course prefer the less complex models. The answer is that different approaches tell us about different parts of the story. No one method is always better than another; it depends on the circumstance. For example, results from case studies and labour market accounts technique suggest that external pressures and shocks impact upon different local labour markets quite differently. This might suggest that VAR-type studies, although having advantages, will not pick up idiosyncratic patterns of labour market adjustment in different regions. In summary, however, it is clear that internal migration is a significant regional labour market adjustment mechanism in some countries (e.g. the US and Australia), but not in others (most of the European countries). The relevant question is where New Zealand fits in. Up to here, we have been looking at the direction (Chapter 4) and size (Chapter 5) of internal migration as a regional labour market adjustment mechanism. In the next chapter, we address the question of who moves and whether it matters for regional labour market adjustment. 97 6 KEY QUESTION 3: WHO MOVES AND DOES IT MATTER FOR REGIONAL LABOUR MARKET ADJUSTMENT? The first and second key questions looked at the direction and size of migration as a regional labour market adjustment mechanism. This chapter focuses on the composition of migration flows and its implications on regional labour market adjustment. A few of the issues and points raised in this chapter have already been discussed in previous chapters, particularly in Chapter 4. Relevant references will be made at the appropriate places. As for the previous chapters, the discussion here is divided into three subsections – “what?”, “how?”, and “findings”. For each of these subsections, there are two parts. Firstly, we ask whether migration behaviour is homogeneous across individuals, and if not, what are the characteristics that make some people more mobile than others. The second part considers the question of whether who moves matters for regional labour market adjustment, and if so, how it matters. 6.1 What? 6.1.1 Identifying heterogeneity Some groups within a society may be more mobile than others. For example, highskilled workers are expected to be more mobile than low-skilled ones. Besides, one might expect that younger individuals are more mobile than older individuals for reasons linked to life-cycle events, such as the departure from the parental home, the start of tertiary education, entry into the labour force and the establishment of independent living arrangements. We could also examine whether this difference in mobility exists for other group classifications, such as gender, ethnicity, family structures (e.g. husband-wife families versus unattached individuals), etc. There is also a strand of the literature that argues that workers in different occupations have varying propensities to migrate. Although this is essentially an occupational mobility issue, it is not unrelated to geographical labour mobility. For example, certain regions may have greater concentrations of workers in particular occupations and thus spatial and occupational mobility are related (see for example, Padoa-Schioppa (1991)). As will be pointed out later in the findings subsection, it is clear that there is heterogeneity in migration behaviour. However, migration propensities are different across a range of dimensions, not only one or two. In addressing our third key question, we will have to ask whether there is anything we can say conclusively about who is more likely to move and who is more likely to stay, when the region experiences a shock. Are there any similarities between people who choose to move? What makes them more mobile? When a particular region is hit by an adverse shock (for example), there are generally three types of people. There are those who have a lower cost of moving, be it in the form of psychological cost, physical cost, etc. This group will be the first to move out of the region for better prospects elsewhere (i.e. the most mobile people). At the other extreme, there is another group of people that finds it costly to move (be it for financial reasons, family ties, etc.), and thus is willing and/or able to bear the consequences of the adverse shock instead of moving out of the region straight away (unable, and/or not wanting, to adjust by migrating out). This group is the least mobile group. Finally, 98 there are those who are somewhere in between, for whom the cost-benefit difference is not sufficient to induce a move yet. The important feature that distinguishes these three groups of people would be the cost of adjustment that they have to bear, in order to move. For the second and third groups, the form that the costs take will depend on factors such as labour supply responses, housing markets, etc. 6.1.2 Implications of heterogeneity on regional labour market adjustment Given that migration behaviour is in fact far from homogeneous, there are at least four implications for us. The first two implications relate to policy and arise because the heterogeneous migration behaviour results in different individuals bearing a different cost of adjustment. The first implication relates to the share of the cost of adjustment. When a region experiences an adverse shock, for example, there needs to be some adjustment. This cost is spread across the heterogeneous population. More mobile people (i.e. those with a lower cost of moving) can capture the gains 92 (avoid the costs 93 ) of regional shocks. The second implication is that the size of the cost of adjustment (due to a negative shock) may be magnified if there is “cumulative causation”. “Cumulative causation” implies that once a region becomes uncompetitive, it is very difficult to return to its original condition, in level terms. In other words, there is a downward spiral. This could happen for various reasons. For example, it could be due to the interaction between economic performance and the creation of human capital (see for example, Taylor (2000)). If this is the case, one has to consider the impact of the migration flow of particular subgroups on the origin and destination communities. For example, a continuous outward migration of young and highly skilled from declining regions may have negative effects of deskilling the regional population and further weakening the regions’ growth potential. The theoretical approach proposed by Roy (1951) argues that the size and direction of migration flows depend on regional differences in the returns to skills (as well as regional differences in mean income). Regions that pay higher returns to skills would attract more skilled workers than regions that pay lower returns. If this is the case, the economic impact of migration depends on which people move as well as on how many people move. Therefore, one has to consider the spillover and neighbourhood effects of migration patterns. These first two implications can be thought of as the consequences of migration (see for example Greenwood (1997)). The consequences of migration can be addressed at two levels. One refers to the performance of migrants in their new locations relative to a benchmark, such as their presumed performance in their previous place of residence had they not moved. The second deals with the impacts that migrants have on others in origin and destination areas. The third and fourth implications relate to further empirical work. The third one is that any modelling of migration needs to control for differences in population compositions. The fourth and final implication is that one has to take into account the selectivity bias. The migration decision in fact separates the population into those who expect to gain 92 In the case of a positive regional shock, the more mobile people are the ones benefiting. For example, if the government introduces assistance programs in particular regions, in-migrants will benefit from them. 93 On the other hand, when there is an adverse shock to a region, the more mobile people are able to leave while those remaining are the ones who have to bear the cost of lack of adjustment. For example, if a factory closes down, the more mobile people are able to move to other regions to look for jobs, but those unable to leave will remain unemployed in their region. 99 by moving, and those who expect to gain by staying. If we ignore this selection process when we estimate the earnings equation (or any other labour market variables) of the two sub-groups, then we may wrongly find more gains to migration than there actually are. Therefore, we have to consider the selectivity bias problem when we conduct our empirical work. 6.2 How? This subsection discusses how researchers identify the heterogeneous migration behaviour, as well as the implications of this heterogeneity on regional labour market adjustment. Before we proceed, it is worth highlighting three main ways in which the approaches used are different. The first distinction is between univariate (using a single characteristic to explain migration behaviour) and multivariate studies (using more than one characteristic). The second distinction lies in the unit of analysis (essentially the type of data used). Some studies use area (or group) characteristics to explain migration rates of the area (or group). These are generally called aggregate studies. On the other hand, there are studies that use not only area characteristics, but also individual characteristics. These studies have access to micro-level data (be it from the census, or survey data, along the lines proposed by Statistics New Zealand as discussed in Section 4.2). Finally, the third distinction is between studies that use a short time series and those that use a long time series. The minimum time variation in order to be able to say anything about the implications of who moves on regional labour markets, is one time window (i.e. a snapshot of the world at two points in time). The New Zealand census is one example of such a data set 94 . With a longer time series, there is much more information that we can infer from the data about the implications on regional labour market, because the adjustment process would have had enough time to work itself out. An example of a data set that provides such long time series is the National Longitudinal Survey of the Labor Market Experience of Youth (in short, National Longitudinal Survey of Youth), in the United States. McCarthy (1992) used this data set to assess the unemployment and wage motivations for migratory behaviour among young males. Given these distinctions, we shall discuss in more detail the range of approaches that researchers have used to examine the patterns of migration behaviour. 6.2.1 Identifying heterogeneity When identifying personal characteristics that influence mobility patterns, one can focus on a single dimension (univariate studies) or look at the interaction of multiple dimensions (multivariate studies). We shall begin with a discussion on univariate studies, which is the easiest to carry out. Univariate and simple descriptive studies Studies of this type have typically used descriptive techniques, such as cross tabulations, simple descriptive statistics and bar charts, to infer about the links between personal characteristics and migration. A few examples are worth mentioning here (as cited from Herzog, Schlottmann et al. (1993)). Saben (1964) inferred from cross tabulations that unemployed workers in the United States are more likely to migrate than are the employed. Meanwhile, Lansing 94 A good discussion of the scope, limits and reliability of using the New Zealand census data on internal migration is provided by Poot (1986a). 100 and Mueller (1967) document the important link between personal unemployment and migration, to the extent that such causation can be revealed within summary tables. However, this approach has a few limitations. Firstly, such an approach does not serve to formally quantify the relationships between migrant and area characteristics, and migration propensities. Secondly, it (wrongly) assumes that there is causality running from one variable to another (e.g. unemployment driving migration decisions). Thirdly, such descriptive techniques do not adequately control for other personal and place characteristics that may affect migration. In other words, the univariate approach focuses on one characteristic alone, in isolation of, or ignoring other characteristics (i.e. only a partial analysis). Without such controls, one may well reach the wrong conclusions. For example, the stereotypical image in Australia and some (naïve) stylised facts may suggest that indigenous Australians are more mobile than the rest of the population (as cited from Taylor and Bell (1996)). However, these observations do not consider variations in age structure across different ethnicity groups. The higher mobility rate among indigenous Australians could be due to the greater concentration of indigenous people in the more mobile, youthful age groups. By standardising the data for the effects of variable age structure, Taylor and Bell (1996) find that the mobility rates for the two groups are essentially the same. Therefore, it is important to consider the interaction between and among different characteristics that are expected to influence migration behaviour. One can of course extend the descriptive analysis to two or more personal characteristics. However, descriptive studies still face other limitations, some of which were discussed above. This leads us to seek multivariate approaches, which (as the name implies) permit the researcher to examine the relative contribution of a range of characteristics on migration behaviour. Multivariate studies A simple multivariate approach is the shift-share technique. As discussed in Section 4.2, the shift-share method is basically a decomposition technique. This is one way for decomposing the change in a particular variable in a region (e.g. change in migration flow) to changes in the variable of interest (change in migration behaviour or mobility) and other influencing variables such as demographic effects (e.g. age). In principle, one can of course control for other dimensions of interest, for example skill, provided that there are sufficient data available. However, the shift-share technique is still very much a descriptive technique. It does not explain why certain regions have locational advantages, or why mobility patterns have changed as they did. An example of the use of the shift-share technique is provided by Green (1994). Green (1994) provides an illustration of the use of the shift-share approach to examine the role of migration in bringing labour supply and demand into balance (or in mitigating imbalance) in Britain in the 1980s. The author decomposes the change in the number of migrants between a base period and a terminal period into three additive components: the population base component (the ‘age composition’ effects), the mobility component (the age-specific rates of inter-area mobility), and the geographicdistribution component (the change in attractiveness of specific destinations for outmigrants from specific origins). The identity equation is as below: 101 ijkijkijkijk CBAM ++=∆ where ijk M∆ is the change in the number of migrants from origin i to destination j in age group k during the intervening period. ijk A is the population base component, which is a purely demographic supply measure, encompassing ‘age composition’ effects. ijk B is the mobility component. ijk C is the geographic-distribution component, that is, the portion of the change in migration by persons of age k from i to j that is attributable to increases or decreases in the destination-specific flow after the effects of the population base ( ijk A ) and mobility (ijk B ) components are removed. Although the shift-share technique can help control for some characteristics of interest, while analysing people’s propensities to move, it can only do so to a certain extent. The number of dimensions one can control for are limited to usually three or four; the analysis gets rapidly more complex with more variables. A more sophisticated and elaborate multivariate approach is the regression technique. The regression technique allows one to assess the impact of a particular characteristic on migration behaviour, controlling for other characteristics (i.e. under true ceteris paribus conditions). The regression technique usually uses the ordinary least squares (OLS) estimator and variants of it. Since these have been covered earlier in Section 4.2, we shall only highlight issues that are particularly relevant to this key question, for example the use of unit record level data. Aggregate studies The regression analysis could be based on average area characteristics (or composition of the area). Such aggregate studies formed the bulk of early research. These studies basically modelled the number of individuals moving across areas as a function of distance, and area characteristics (e.g. the unemployment rate and the average wage level at the origin and destination). Later studies extended the explanatory variables to include demographic characteristics of the migration streams such as the average age and education levels. However, these studies are still of an aggregate nature. Studies of this category have already been discussed in Chapter 4. However, it is worth noting a recent example of such a study in the New Zealand context. Kerr, Maré et al. (2001) attempted to identify the characteristics that push people to move and/or pull them to a new location. Their paper investigates the effect that the characteristics of a community have on the likelihood of people leaving and/or travelling to the community. The data are aggregated to an Area Unit (equivalent to a suburb); therefore, the smallest definable community is an Area Unit. They use geographical information system (GIS) tools to define variables based on aggregations of meshblocks around the area units of interest. Such aggregate studies have been criticised on several grounds, three of which are particularly noted here. Firstly, the theoretical foundation on which most of these studies were based is Sjaastad’s (1962) human capital approach and variants of it, which focuses on the behaviour of individuals. In order to analyse the individual’s migration decision, strictly speaking one needs to have individual characteristics as well as area characteristics, rather than area characteristics alone. A related criticism 102 of aggregate studies is that a high qualification area may have a high mobility but this does not necessarily mean that it is the high-qualification people who are moving. Secondly, aggregate studies fail to adequately capture the heterogeneity of the stream of individuals moving from one area to another. Aggregate data is likely to conceal differences in the underlying determinants of migration of various population subgroups. Thirdly, there is still the selectivity bias problem, as discussed in Section 4.2. While it is noted that one necessarily requires unit record data to control for selectivity bias, using microdata in itself does not necessarily absolve one from the selectivity bias problem. Individual (microdata) studies There is quite an extensive list of studies employing individual data in countries such as the United States and Britain (see for example, Böheim and Taylor (2000), Gardner, Pierre et al. (2000), Enchautegui (1997), and Day (1992)). In the New Zealand context, there have been few such studies. However, one such study is by Vaithianathan (1995) who used micro level data (1991 census data) to model the Maori and non-Maori males regional migration decision. The author employed a binary logit model (move/stay) to assess the effect of observed regional and personal characteristics on the probability that an individual leaves an area. Longitudinal data Furthermore, there is a strand of literature that takes into account a number of life-cycle considerations that are potentially important in an individual’s or a family’s decision to migrate. For example, a migration decision is likely to be made in conjunction with other events such as marriage or divorce, birth and ageing of children, graduation from school, episodes of unemployment and finding new jobs, entering the workforce and retirement. These studies employ event history data or longitudinal data (which tracks the same individuals or families over time). Alternatively, where no “pure” longitudinal data set is available, one can use pooled cross-sectional microdata (i.e. “synthetic” longitudinal data). The advantages of using longitudinal data and some key references (e.g. Odland (1997)) have already been discussed in Chapter 4. As far as we are aware, little work has been done in New Zealand to capture such dynamic effects (e.g. migration history). The only example we have come across is Heenan (1999). This paper is based on autobiographical details of residential mobility for a small sample of New Zealand male career professionals collected in 1981. Classified by cohort and employment sector, the information is used to examine migration in the context of personal career development. Aspects explored include age of entry to the workforce, frequency of movement, migration and career promotion, and the spatial structure of migration. However, this study has a small sample size. Despite this caveat, this New Zealand study is nevertheless a good attempt to account for the context and individual decision-making processes involved. One reason for the scarcity of longitudinal studies in New Zealand is because we do not have a suitable longitudinal data set. We only have cross-sectional data (based on the national census), which are available over several five-year periods. Based on such data, one can only use either aggregate studies (impact of “area” characteristics, or “average among individuals in the area” characteristics, on migration flows) or individual studies (if unit-record data is available). 103 Case studies Another approach that has been used to identify sub-groups within the population who are bearing the cost of adjustment or the lack of adjustment (i.e. identifying heterogeneity in migration flows) is using case studies. This approach contrasts with the other approaches discussed earlier because it focuses on a particular location and has advantages particularly when mobility patterns and their dynamics differ across locations. Le Heron and Pawson (1996) and Peck (1985) provide a good starting point for thinking about possible case studies in New Zealand to address our key questions. 6.2.2 Implications of heterogeneity on regional labour market adjustment As discussed earlier, the consequences of migration can be addressed at two levels – one deals with the migrants themselves, and the other deals with migration’s impact on others in the origin and destination areas. In general, although there has been much research on the determinants of migration, little research has focused on the consequences of migration (see Greenwood, Mueser et al. (1991) and Greenwood (1997)). However, the first type of consequence (i.e. impact on migrants themselves) has been studied relatively more than the second (i.e. impact on origin and destination). As Greenwood (1997, p.650) puts it, “… the second [consequence], which although potentially important, has not been studied in sufficient depth regarding internal migration”. Greenwood, Mueser et al. (1991) argues that this is because micro and panel data, while valuable for studying the behaviour of individuals (rather than streams of aggregate movement) and on decision-making processes, “…do not lend themselves to an analysis of the consequences of migration, except the personal consequences”. Many studies, however, put forward speculative and often untested statements that imply that the question of who moves have some impact on origin and/or destination areas. As an illustration, the quote below is taken from Green (1994): “The concerns of many policymakers with retaining more highly qualified young adults in their origin areas to contribute to local economic development would appear to be well placed.” (p. 1576) Of those studies that have in fact looked at assessing the effects of shocks and migration flows on the community, case studies are probably the most common approach used. One could focus, for example, on the impact of industrial closedowns and mass redundancies. As mentioned earlier, Le Heron and Pawson (1996) and Peck (1985) provide a good starting point for thinking about possible case studies in New Zealand to address this second part of the third key question. 6.3 Findings This subsection summarises the findings from research that seeks to identify the heterogeneous migration behaviour, and the impacts of this heterogeneity. For the latter, the discussion will be divided into two parts; one looks at the impact on individuals themselves, and the other focuses on the link between who moves (i.e. the heterogeneous pattern) and the effects on the origin and destination communities. The first part effectively covers the share of the cost of adjustment while the second looks at the size of the adjustment cost. 104 6.3.1 Identifying heterogeneity The conclusion from this subsection is that migrants’ behaviour is far from homogeneous. It depends on factors such as qualifications, family structure, gender, home ownership, ethnicity, etc. And this is true for New Zealand and for other countries. Generally, there is evidence that younger, highly skilled, single males, who do not own homes, are relatively more mobile. As for ethnicity, there is no strong evidence of such a systematic mobility relationship. The relationships are more complex, as evident below. We shall now examine the various pieces of evidence that point to these conclusions. Univariate studies First, let’s look at some stylised facts of mobility differences, according to descriptivetype studies. Table 6.1 Proportions of migrants by age, for the 1996 total population 1 Note: 1. These percentages are calculated from a total excluding the “not specified” category. Data source: Unpublished data files prepared by the Customer Services Division of Statistics New Zealand from the 1991 and 1996 Censuses of Population and Dwellings. This table was extracted from Goodwin and Bedford (1997), p.8. Table 6.1 shows the proportions of each five-year age group that are local and nonmigrants (i.e. those that have not moved out of their previous region of residence), inter-regional migrants, and overseas migrants in 1996. It is clear that the most mobile age group are those aged 15-39 years, particularly those between 20 and 34 years of age. For example, approximately 22 percent of all 20-24 year olds had moved out of their usual region of residence into another region in New Zealand, whereas only 7 percent of the 50-54 years age group have done so. This is also true for international migrants. This pattern (i.e. more mobility amongst younger people) is apparent in other countries as well. Figure 6.1 illustrates the incidence of internal migration by age in selected OECD countries. 111 “The effects of population loss on those who remain could be severe. Some of the people now contemplating leaving have played significant roles in the community and their departure will leave roles which are difficult to fill” (p. 1, TUC Energy Monitor Progress Report No. 8, 4 October 1982). For example, due to the freezing works closure, some sports clubs and community organisations in Patea were experiencing a decline in membership. More importantly, the departure of key members from these organisations raised questions about their future viability. Meanwhile, in the education sector, the Patea High School student roll fell by 14% over the year following the closure. Similarly, Patea Primary School experienced a 13% decline in student numbers over the same period. Therefore, the question of who actually moves out of a declining area (in this case, Patea), has substantial impacts on that area. Other case studies for further exploration can be found in Le Heron and Pawson (1996) and Peck (1985). 6.4 Concluding remarks In summary, we have sufficient evidence that migration propensities differ across the population. Some characteristics have been found to be affecting migration in a systematic way (e.g. age and education) more than others (e.g. ethnicity). While there has been much research on the determinants of migration (area and personal characteristics), relatively less research has focused on the consequences of migration. The consequences of migration can be divided into two – the impact on migrants themselves (personal consequences), and the impact on origin and destination areas (broad consequences). There is relatively little research on the broad consequences of migration. However, the composition of migration flows could have significant implications on origin and destination areas, and this in turn have some policy implications (e.g. debate on whether or not to have area-based economic development policies, or people-based strategies). A challenge for researchers is to ensure that their work catches up with the existing debate on issues. 112 7 RESEARCH OPTIONS 7.1 Introduction This section of the paper outlines selected research projects that could shed light on the key questions that have been identified. The list of projects is not exhaustive but reflects instead our view of what is feasible, and is most likely to provide reliable findings of relevance for public policy. In judging relevance, we have drawn on the analysis and discussion that has been set out in the preceding chapters. We have focused on research that can help us answer the three key questions discussed above. The links of these questions to policy issues have already been articulated, so will not be repeated here. We have also kept in mind the research gaps that were identified in Chapter 1 and have chosen research projects that do not replicate existing or current research. We have focused our attention on feasible research options - those for which appropriate data are available, and for which there are plausible ways of addressing the 'identification problem' discussed in Chapter 2. We have considered cost and scope as part of feasibility, and have chosen projects that can be undertaken at moderate cost. We have therefore excluded complex projects that require significant investment, such as the building of complex multi-regional models. This is in part a pragmatic choice, and in part a recognition that the knowledge gaps that exist in the New Zealand literature are large. Faced with fundamental gaps in knowledge, uncovering broad patterns can be of great use, and will provide a good foundation for more complex analyses that may follow. Finally, in selecting research options, we have considered the expected quality of findings. In our reviews of the methods that have been used to answer the key questions, we have noted the potential sources of bias associated with different empirical methods. Our research options incorporate suggestions about how to account for these biases, so as to generate findings that are as robust as possible. 7.2 Choosing which of the options to proceed with As noted in Chapter 1, no single research project is able to provide answers to all three of the key questions that we have focused this paper on. The methods and types of data that researchers generally use to focus on each question differ by question. The questions, and the research options that are presented, have in fact been chosen to be largely complementary. The choice of which research option is the most important or should be done first therefore centres on which of the three questions is of greatest interest. Having said that, choosing any one option may shed some light on all three questions. It is just that what will revealed about the questions that the option was not designed to answer will be less robust. For instance, option one is designed to detect the existence and size of migration responses to labour market changes. One of the suggested sets of controls to be included in the study are controls for demographic composition of areas. We will therefore learn something about "who moves" - one of the parts of key question three. 113 Similarly, option two is designed to learn about the relative size of the migration response to regional labour market changes, compared with other forms of labour market change, and about the speed of regional labour market adjustment. It will therefore reveal something about the presence and strength of the migration response to labour market change (key question one). It does not, however, allow as much attention to be paid to biases that may arise from differences in demographic composition or non-labour market differences across regions. The case study option, option three, focuses on the way that labour market adjustment in the form of migration affects particular communities. Given the breadth of information that can be collected as part of a case study, it will be possible to say something about the existence and relative importance of migration adjustment for the communities or incidents studied. However, being case studies, they will not be able to inform us about how significant the patterns that are identified are for New Zealand as a whole. Options one and two are fairly clearly defined, although some work is still needed to confirm that the necessary data are available for option two. Option three is less well defined, but we suggest that a first step would be to develop the proposal in more detail along the lines that we set out below. The choice of whether to proceed with one or more of the options comes down to a question of available resources, and a judgement about the importance of answering each key question, relative to other possible demands on research resources. It is hoped that the extensive discussion in the body of this paper about the strengths and weaknesses of various methods, and about the range of findings found in the existing literature, will greatly reduce the inevitable uncertainty surrounding this judgement. 114 7.3 Option 1. Identifying the link between migration flows and regional labour market adjustment 7.3.1 Introduction This first research option relates to key question one, which considers the evidence for a link between migration flows and labour market conditions. We see the estimation of a spatial interaction model, based around a generalised gravity relationship, as the most appropriate way forward. There is a well-developed empirical literature using these models which provide guidance and will allow for comparison of results (including comparisons with existing New Zealand studies (Poot (1986b); Maré and Timmins (2000)). Our proposal is for a study using New Zealand area-level census data, examining the relationship between inter-censal (5-year) migration flows and labour market conditions at various levels of aggregation (area units, territorial local authorities, and regional councils). The discussion of this research option describes the analysis that we envisage, and outlines a range of specification and sensitivity issues that the study should address. 7.3.2 Methods The foundation for the study is a spatial interaction model as outlined by Alonso (1978). This model formulates migration flows as a function of relative attractiveness of origin and destination areas, and of frictions between areas. The flow from area i to area j is modelled as: M ij (t) = c(t) O i (t) D j (t) F ij (t) where M ij (t) is the migration flow from i to j c(t) is a constant O i (t) is the total forces 'pushing' people out of region i D j (t) is the total force 'pulling' people into region j F ij (t) is the friction affecting flows between i and j In the simplest gravity model, O i and D j are replaced by the populations of areas i and j respectively (each with an estimated exponent), and the friction term is replaced by a measure of the distance between i and j (also with an estimated exponent). See Nijkamp and Poot (1987) for a fuller discussion of the components of the spatial interaction model. One of the main criticisms of such models is the lack of an explicit behavioural foundation that would justify the functional form used. 96 While accepting that criticism, for our purposes, we suggest relying on a simple gravity relationship to capture the strong empirical patterns that are known to exist between population size, distance, and migration flows. Our main interest is to estimate the relationship between migration and labour market conditions. We therefore suggest augmenting the basic model by adding appropriate measures of the relative attractiveness of local labour markets that are more closely linked to behavioural factors. Other measures could be added capturing factors that 96 Some progress has been made in developing microfoundations for gravity models in the context of international trade (Anderson and van Wincoop (2001) and Deardorff (1998)), although Deardorff (1998) has pointed out that a gravity equation is derivable from any plausible model of trade. 115 are suggested by theory and that could bias our estimates of labour market impacts if they were omitted. The general form of the estimating equationwould be 97 : ( ) 321321 β ββ α αα ijjiijjiij FXXDPcPM = where P i and P j are populations of areas i and j respectively D ij is the distance (or travel time) between i and j X i and X j are other measures of the attractiveness of areas i and j F ij contains other measures of friction of flows between i and j. Labour market measures Measures of relative labour market attractiveness could include: ! relative employment rates: employment/working age population, or employment/labour force (depending on whether the employment chances of discouraged workers and non-participants are relevant for the migration decision) ! relative employment growth rate; ! relative wage rate (measured as average or median) All relative measures are to be measured as deviations from national average so as to capture relative attractiveness 98 Other measures of area relative attractiveness In our discussion of key question one, we emphasised the potential importance of permanent differences in the attractiveness of an area. These could arise due to the presence of amenities, sustained differences in cost of living, etc. Where direct measures of environment, climate, infrastructure, etc. are available these can be included as covariates. Otherwise, location-specific attractiveness can be accommodated by methods such as fixed effects modelling. 99 Where available, a range of other measures of attractiveness can be included, such as ! fiscal measures (net public provision of goods and services to an area); ! diversity or concentration measures (e.g. a broad based industrial structure may be attractive as it allows some 'insurance' by pooling industry-specific risks (see Maré and Timmins (2000)). ! neighbourhood measures: If using spatially coded data, it is possible to derive measures of characteristics of areas surrounding a particular location, which may be relevant for migration decisions (see Kerr, Maré et al. (2001)). Composition differences Another issue emphasised in the discussion of key question one was the importance of allowing for differences in demographic or employment composition. One of the costs of using area-level data rather than individual data is that the ability to control for intergroup differences in migration rates is limited. To the extent that these are also 97 For larger geographical areas, the equation can be estimated in log form by linear regression. For smaller geographical areas, such as area units, some areas will have zero inflows or outflows, so the log transformation is not appropriate, and some form of nonlinear estimation (eg: maximum likelihood) is needed. 98 This would not be appropriate if we wished to understand international migration flows as well, as it ignores the relative attractiveness of New Zealand compared with other countries. See below for a brief discussion of the treatment of international migration 99 See Mátyás (1997), Mátyás (1998), and Eggers (2000) for a discussion of estimating gravity models with panel data. See also Isard (1998) and Sen and Smith (1995) for a more complete discussion of gravity model specification issues. 116 correlated with labour market conditions, estimates of the relationship between migration and labour markets could be biased. At a minimum, measures of demographic composition along key dimensions such as age, gender, marital status, ethnicity, home ownership, industry mix should be included as covariates. As with the attractiveness measures, these should be measured as deviations from national means. Measures of Friction As noted above, the most common measure of migration friction is distance between locations. We consider that travel time is superior to linear (crow-flight) distance. When using area-level data, each area must be assigned a specific location. We consider population centroids as the most appropriate measure for migration modelling. Indicators of physical barriers between any pair of locations, such as Cook Strait or the Southern Alps can be included in order to test whether they represent any additional friction. When looking at links between migration and labour markets, it is plausible to consider more general forms of 'distance' or dissimilarity between locations. For instance, migration may be stronger between areas that have similar occupational mixes, age structure, ethnic mix, or industrial composition. Conley and Topa (2000) argue that such measures capture aspects of social and information networks. Some may also reflect the importance of industry or occupation-specific human capital. A related concept that is common in the internal migration literature is that of the "wellworn-path". Migration flows will be stronger between two areas where there has been significant past migration. Sometimes, lagged migration flows are included to reflect this factor. An alternative that we suggest is to express the sum of inter-area flows for a given pair of areas as a proportion of the sum of their populations. This approximates the proportion of the two areas' populations that have (recently) lived in both regions. Administrative boundaries have also been included as possible sources friction. In the context of a New Zealand internal migration study, regional council or territorial local authority boundaries can be used where the unit of observation is an area unit. 7.3.3 Specification issues to be resolved As with most empirical studies, the devil of the project is in the detail. We list in this section a range of technical, conceptual, or modelling issues that will need to be dealt with in some way by a credible applied study. International Migration Many existing studies of migration and regional adjustment have focused solely on internal migration, ignoring the adjustment that takes place via external migration. As we have noted earlier, this (lack of) treatment of border flows is more of a problem in the New Zealand context than for other larger countries. Two main issues arise if we are to incorporate external flows into an analysis of migration and regional labour markets in New Zealand: • Labour Market Attributes: Ideally, we would like to treat foreign countries in the same way as we do any other location – using information on relative labour market conditions, attractiveness, and frictions. In practice, it is not possible to get as complete information on all potential source and destination countries as 117 we have available for New Zealand locations. Similarly, it is not clear what distance or population measures should be chosen to associate with flows to the rest of the world. • Missing Emigration rates: If we are to use Census data, which contains retrospective 'prior location' questions, there is no information on people who had left New Zealand, as they were not present on Census night. 100 The way that these two issues are resolved depends on how misleading the omission of external flows would be for our inferences on the role of migration as a regional adjustment mechanism. Figure 3.2 shows that for the 1991-1996 period, international inflows occur at a relatively low rate that does not vary greatly across regions (with the exception of non-UK non-Australian inflows into Auckland). There is therefore limited scope for external inflows to serve as a general regional adjustment mechanism. Figure 3.3 show that the pattern of external inflows across regions is similar to the pattern of (estimated) external outflows, again with the exception of Auckland. Although external migration may play a significant role in facilitating to changes in relative labour market conditions between countries, it appears to play a more limited role in facilitating labour market adjustment between locations within New Zealand. While this suggests that an analysis of internal flows alone is likely to be informative for the question at hand, we consider that it is still prudent to check the sensitivity of any inferences to the inclusion of external flows. Estimates based on gross internal migration flows alone could be used as a base specification. Including the data that is available on inflows from outside New Zealand entails adding one observation for each area in New Zealand, capturing the flow from the rest of the world to each area. Area attributes could be entered as sample means, and then the relative attractiveness of New Zealand compared with the rest of the world absorbed by a dummy variable/ fixed factor. If estimated external outflow rates are available, these could be treated symmetrically 101 . Commuting Short term migration (essentially residential moves) and commuting are substitutes. A credible study would need to either model commuting patterns explicitly, or find a way to focus attention on non-residential moves. A plausible way to reduce the influence of residential moves on estimates is to exclude short-distance moves. For instance, moves of less than 10km or less than 20km could be excluded. Figure 3.1 shows that a 10km limit excludes around 50 percent of moves, and a 20km limit excludes around 60 percent of moves. Endogeneity of labour market conditions A well known issue associated with estimating flows from stock measures, as is done when migration flows are modelled as a function of labour market conditions and population sizes, is that choices need to be made about when to measure the stocks, and that these choices can matter. For instance if we wish to understand migration flows between 1991 and 1996, we could use relative wage levels from 1991, or from 100 It is possible to estimate outflows by using estimated mortality rates to isolate the unexplained population loss, as James Newell has done at the Regional Council level for flows between 1991 and 1996. 101 Given the inevitable additional noise that is introduced by the modelling of outflows, it would be informative to report estimates of key labour market parameters based on internal flows and external inflows only. 118 1996. Measures from 1991 may miss subsequent wage changes that induced migration flows. However, to the extent that migration influences wages, estimates based on 1996 wages would confound this effect with the influence in the other direction (wages to migration), which is what we are trying to estimate. One way of steering through this dilemma is to instrument for 1996 wages (or 1991-96 wage changes) using information available in 1991. Adding-up constraints Estimates from a spatial interaction model will not necessarily 'add-up' unless additional constraints are imposed. When modelling internal migration flows, total outflows are equal to total inflows, since everyone has to be somewhere! There is a range of constraints that have been applied in the literature. Singly constrained models require that the sum of estimated inflows equals actual total inflows, or that the sum of estimated outflows equals actual total outflows. A 'doubly-constrained' model requires both. If the estimates are going to be used for simulation or prediction, it is important that these adding up conditions are taken seriously (see Nijkamp and Poot (1987)). Symmetry of inflows and outflows Modelling gross migration flows can allow for relative attractiveness to affect inflows and outflows differently. Modelling net flows imposes symmetry – the effect on inflows is assumed to be exactly the opposite of its effect on outflows. Gross flow modelling also reveals more about the underlying influences on flows. In New Zealand, where migration effectiveness is relatively low, it is particularly important to use information on two-way flows, and thus to model gross flows. Level of aggregation The relationship between labour market conditions and migration flows may differ at different spatial scales. Table 3.5 shows that the estimated patterns of persistence in labour market conditions is different for area units compared with regional councils. No level of aggregation is of greater importance or interest a priori if the objective is to understand the links between migration and labour markets. Care is however needed when drawing inferences, given that there is likely to be more idiosyncratic noise in data on smaller spatial units. It would be useful to produce estimates for area units, TLAs and Regional Councils. 7.3.4 Data The New Zealand quinquennial census is the most appropriate source of data for this study. The database that the Treasury and Motu have established contains information from the 1981, 1986, 1991 and 1996 censi. These data provide sufficient information on migration flows, labour market characteristics and conditions, and broad demographic characteristics of area populations at area unit level. 102 Area unit data can be aggregated to TLA or Regional Council boundaries. In the course of the project, it may be possible to supplement the dataset with additional measures of amenities. 102 The exceptions as noted earlier are the lack of data on external out-migration, and the lack of labour market characteristics for countries outside New Zealand. 119 The data are stored as geographic information, so it is relatively straightforward to derive spatially defined variables such as neighbourhood characteristics and travel times. Unit record analysis would require access to unit record data from Statistics New Zealand. At this stage, we consider that a study using spatial units as the unit of observation would enable us to advance our knowledge substantially. Once the broad patterns and relationships have been identified, consideration can be given to a more detailed analysis using individual data. 7.3.5 Resources & Timing The main resources needed for the project are in the form of the time of researchers. The data are already largely assembled for this project. Although we have not done detailed estimates or costings, we judge that this project would take a competent empirical researcher a minimum of 4-6 weeks full-time to complete, probably spread over several months. Dealing credibly with the range of specification issues and sensitivity testing could double this estimate. 120 7.4 Option 2. Identify relative strength of migration as a regional labour market adjustment mechanism 7.4.1 Introduction The second research option is most closely linked to the second key question, looking at the relative strength of migration adjustment, compared with other forms of labour market adjustment. The suggested method is to use a VAR modelling technique to capture the time series variation in labour market measures. Following the approach in the literature, the migration response will be inferred from estimates of variation in other labour market measures. 7.4.2 Methods The general form of Vector Autoregressive (VAR) models is discussed in Chapter 5, as is the literature that uses VAR models to estimate regional labour market adjustment patterns. Given the discussion in Chapter 5, the presentation here will be somewhat condensed, reiterating some of the main issues with which a credible empirical project will need to deal. Base specification Box 2 in Chapter 5 presents a stylised Vector Auto-Regressive (VAR) model using three labour market variables. While this gives a good idea of the form of the model, there is a range of issues that need to be resolved even before settling on a base specification. Specification issues to be resolved The discussion in Chapter 5 has already covered many of the specification issues. They are therefore discussed only briefly here. Choice of variables The first choice that a project in this area will need to settle is the choice of which labour market variables to include. There is a variety of choices in the literature, reflecting data availability, data quality, and the main interest and focus of particular authors. Adding more variables uses up degrees of freedom fairly quickly, so parsimony is essential. The insights from including non-labour market variables must similarly be weighed against the (statistical) costs of doing so. The form in which the variables enter the equations is also requires some judgement by the researcher. As noted in the text, the errors of the VAR system of equations should be time-series stationary. This can be achieved by testing and suitably transforming the labour market variables. Blanchard and Katz (1992) also transform their labour market variables to be deviations from national means, so as to exclude the (common) impact of aggregate shocks, and focus their findings on the impact of changes in relative labour market conditions. Choice of lag structure Choosing an appropriate lag structure for each variable also reflects a tension between capturing non-trivial dynamics and reducing the time period for which estimation is 127 Blanchard, O. J. and L. F. Katz (1992). “Regional Evolutions.” Brookings Papers on Economic Activity 1: 1-75. Böheim, R. and M. Taylor (2000). Residential mobility, housing tenure and the labour market in Britain. Essex, Institute for Social and Economic Research and Institute for Labour Research, University of Essex. Bolton, R. (1985). “Regional Econometric Models.” Journal of Regional Science 25(4): 495-520. Box, S. (2000). "Economic Geography - Key Concepts." Treasury Working Paper 00/12, Wellington NZ, New Zealand Treasury, www.treasury.govt.nz/workingpapers/ Brosnan, P. and J. Poot (1987a). “Modelling the Determinants of Trans-Tasman Migration after World War II.” Economic Record 63(183): 313-329. Brosnan, P. A. and J. Poot (1987b). “Economic and Demographic Aspects of Trans- Tasman Population Exchange.” New Zealand Population Review November: 4-16. Bukenya, J. O., P. V. Schaeffer, et al. (1999). The Effect of Wage Differentials and Regional Job Growth on Migration: A Case of West Virginia, Department of Agricultural and Resource Economics. Burnley, I. H. (1993). “Migration and Mobility Issues in Australia and New Zealand in the 1990s.” New Zealand Population Review 19(1&2): 15-31. Cameron, G. and J. Muellbauer (1998). The Housing Market and Regional Commuting and Migration Choices. Oxford, Nuffield College. Carlsen, F. (2000). “Testing Equilibrium Models of Regional Disparities.” Scottish Journal of Political Economy 47(1): 1-24. Cashin, P. and L. Strappazzon (1998). “Disparities in Australian Regional Incomes: Are They Widening or Narrowing?” Australian Economic Review 31(1): 3-26. Chapple, S. (2000). Spatial inequality and labour market adjustment in urban neighbourhoods: 5th draft. Wellington NZ, New Zealand Department of Labour. Charney, A. H. (1993). “Migration and the Public Sector: A Survey.” Regional Studies 27: 313-326. Clark, D. E. and T. A. Knapp (1995). "The Hedonic Price Structure of Faculty Compensation at U.S. Colleges and Universities." Review of Regional Studies 25(2): 117-142. Comín, D. (1999). Regional Mobility and Geography. Cambridge MA, Department of Economics, Harvard University. Conley, T. G. and G. Topa (2000). Socio-Economic Distance and Spatial Patterns in Unemployment. Illinois, Department of Economics, Northwestern University, and New York, Department of Economics, New York University. 128 Crampton, G. R. (1999). "Urban Labour Markets." In: E. S. Mills and P. Cheshire (eds.). Handbook of Regional and Urban Economics. Amsterdam, Elsevier Science, North- Holland. 3: Applied Urban Economics: 1501-1557. Davies, R. B. and A. R. Pickles (1985). "Longitudinal versus cross-sectional methods for behavioral research: A first round knock out." Environment and Planning A 17: 1315-1329. Day, K. M. (1992). “Interprovincial Migration and Local Public Goods.” Canadian Journal of Economics 25: 123-144. Deardorff, A. (1998). "Determinants of Bilateral Trade: Does Gravity Work in a NeoClassical World." In: J. Frankel (ed.). The Regionalization of the World Economy. University of Chicago for the NBER: 7-32. Debelle, G. and J. Vickery (1998). Labour Market Adjustment: Evidence on Interstate Labour Mobility. Sydney, Reserve Bank of Australia. Debelle, G. and J. Vickery (1999). “Labour Market Adjustment: Evidence on Interstate Labour Mobility.” Australian Economic Review 32(3): 249-63. Decressin, J. and A. Fatas (1995). “Regional Labour Market Dynamics in Europe.” European Economic Review 39: 1627-1655. Duke-Williams, O. and M. Blake (1999). Database Fusion for the Comparative Study of Migration Data. Adelaide, Key Centre for Social Applications of GIS (GISCA), University of Adelaide. Dumond, J. M., B. T. Hirsch, et al. (1999). “Wage Differentials across Labor Markets and Workers: Does Cost of Living Matter?” Economic Inquiry 37(4): 577-98. Elhorst, J. P. (2000). The mystery of regional unemployment differentials; a survey of the theoretical and empirical explanations. Groningen, Faculty of Economics, University of Groningen: 1-56. Enchautegui, M. E. (1997). “Welfare Payments and Other Economic Determinants of Female Migration.” Journal of Labor Economics 15(3): 529-554. Enders, W. (1995). Applied Econometric Time Series. New York, John Wiley & Sons, Inc. Fields, G. S. (1979). "Place to place migration: Some new evidence." Review of Economics and Statistics 61: 21-32. Fredriksson, P. (1999). “The Dynamics of Regional Labor Markets and Active Labor Market Policy: Swedish Evidence.” Oxford Economic Papers 51(4): 623-48. Fry, J. M., T. R. L. Fry, et al. (1999). “Inter-regional Migration in Australia: An Applied Economic Analysis.” Australasian Journal of Regional Studies 5(2). Gallin, J. H. (1999). Net Migration and State Labor Market Dynamics. Washington DC, Federal Reserve Board. 129 Gardner, J., G. Pierre, et al. (2000). Moving for Job Reasons. Coventry, UK, Department of Economics, University of Warwick. Gibson, J. (1993). “Regional Consequences of New Zealand Industry Assistance Policies.” New Zealand Journal of Geography April: 2-6. Giesecke, J. A. (2000). FEDERAL-F: A multi-regional multi-sectoral dynamic model of the Australian economy. PhD. dissertation, Adelaide, University of Adelaide. Giles, D. E. A. and P. Hampton (1978). “A Note on Urban Migration in New Zealand.” Journal of Urban Economics 5(3): 403-408. Goetz, S. J. (1999). Migration and Local Labor Markets, Regional Research Institute, West Virginia University. Goetz, S. J., R. C. Ready and B. Stone (1996). "U.S. Economic Growth vs. Environmental Conditions: A State-Level Analysis." Growth and Change 27: 97-110. Goodwin, J. E. and R. D. Bedford (1997). "Inter-regional Migration in New Zealand, 1986-1996: A Preliminary Analysis." Briefing Paper for the Population Conference: Communities, People and Growth, No. 4, Wellington NZ, 12 to 14 November. Gorbey, S., P. Briggs, et al. (1993). "Backdating the Household Labour Force Survey." NZIER Working Paper 93/5, Wellington NZ, New Zealand Institute of Economic Research. Gorbey, S., D. James, et al. (1999). “Population Forecasting with Endogenous Migration: An Application to Trans-Tasman Migration.” International Regional Science Review 22(1): 69-101. Gordon, I. (1985). "The cyclical interaction between regional migration, employment and unemployment: a time series analysis for Scotland." Scottish Journal of Political Economy 32: 135-158. Gordon, I. and D. Lamont (1982). "A model of labour market interdependencies in the London region." Environment and Planning A 14: 237-264. Gorringe, P. (2001). "Methodology of Policy Advice." In: A. Grimes, A. Jones, R. Procter and G. Scobie (eds.). Economics For Policy: Expanding the Boundaries - Essays by Peter Gorringe. Wellington NZ, Institute of Policy Studies, Victoria University of Wellington: 230-250. Graves, P. E. (1979). "A life-cycle empirical analysis of migration and climate, by race." Journal of Urban Economics 6: 135-147. Green, A. E. (1994). “The Role of Migration in Labour-Market Adjustment: The British Experience in the 1980s.” Environment and Planning A 26(10): 1563-77. Green, A. E., T. Hogarth, et al. (1999). “Longer Distance Commuting as a Substitute for Migration in Britain: A Review of Trends, Issues and Implications.” International Journal of Population Geography 5: 49-67. Greene, W. H. (1990). Econometric Analysis. New York, Macmillan. 130 Greenwood, M. J. (1969). "An Analysis of the Determinants of Geographic Labor Mobility in the United States." Review of Economics and Statistics 51: 189-194. Greenwood, M. J. (1975). "A simultaneous-equations model of urban growth and migration." Journal of the American Statistical Association 70: 797-810. Greenwood, M. J. (1985). "Human Migration: Theory, Models, and Empirical Studies." Journal of Regional Science 25: 521-544. Greenwood, M. J. (1997). "Internal Migration in Developed Countries." In: M. R. Rosenzweig and O. Stark (eds.). Handbook of Population and Family Economics. Amsterdam, Elsevier Science. 1B: 647-720. Greenwood, M. J. and G. L. Hunt (1989). "Jobs versus amenities in the analysis of metropolitan migration." Journal of Urban Economics 25: 1-16. Greenwood, M. J., P. R. Mueser, et al. (1991). “New directions in migration research.” Annals of Regional Science 25: 237-270. Groenewold, N. (1997). “Does Migration Equalise Regional Unemployment Rates? Evidence from Australia.” Papers in Regional Science 76(1): 1-20. Gujarati, D. N. (1995). Basic Econometrics. New York, McGraw-Hill. Hanson, G. H. and M. J. Slaughter (forthcoming, 2001). “Labor Market Adjustment in Open Economies: Evidence From U.S. States.” Journal of International Economics. Heenan, B. (1999). “Career Migration as Personal Biography: New Zealand Case Studies.” New Zealand Geographer 55(1): 12-24. Herzog, H. W., A. M. Schlottmann, et al. (1993). “Migration as Spatial Job-Search: A Survey of Empirical Findings.” Regional Studies 27(4): 327-340. Herzog, H. W., A. M. Schlottmann, et al. (1986). “High Technology Jobs and Worker Mobility.” Journal of Regional Science 26(3): 445-459. Hughes, G. A. and B. McCormick (1987). "Housing Markets, Unemployment and Labour Market Flexibility in the UK." European Economic Review 31: 615-645. Hunt, G. L. (1993). “Equilibrium and Disequilibrium in Migration Modelling.” Regional Studies 27(4): 341-349. Isard, W. (1998). "Gravity and spatial interaction models." In: W. Isard, I. J. Azis, M. P. Drennan et al. (eds.). Methods of Interregional and Regional Analysis. Aldershot UK, Ashgate: 243-279. Isard, W., I. J. Azis, et al. (eds.) (1998). Methods of Interregional and Regional Analysis. Aldershot UK, Ashgate. Isserman, A., C. Taylor, et al. (1986). "Regional Labor Market Analysis." In: P. Nijkamp (ed.). Handbook of Regional and Urban Economics. Amsterdam, Elsevier Science. 1: 543-580. 131 Jimeno, J. F. and S. Bentolila (1998). “Regional unemployment persistence (Spain, 1976-1994).” Labour Economics 5(1): 25-51. Johnes, G. and T. Hyclak (1994). “House Price, Migration, and Regional Labor Markets.” Journal of Housing Economics 3: 312-329. Juarez, J. P. (2000). “Analysis of Interregional Labor Migration in Spain Using Gross Flows.” Journal of Regional Science 40(2): 377-399. Karagedikli, Ö., D. C. Maré, et al. (2000). “Disparities and Despair: Changes in Regional Income Distributions in New Zealand 1981-1996.” Australasian Journal of Regional Studies 6(3): 323-347. Kelejian, H. H., D. P. Robinson, et al. (1997). Convergence and Divergence of US Cities: 1970-1990. Kennedy, S. and J. Borland (2000). “A Wage Curve for Australia?” Oxford Economic Papers 52(4): 774-803. Kerr, S., D. C. Maré, et al. (2001). "Internal Mobility in New Zealand." Treasury Working Paper 01/04, Wellington NZ, New Zealand Treasury, www.treasury.govt.nz/workingpapers/ Kerr, S. and J. Timmins (2000). "Economic Geography and Spatial Statistics: Theory and Empirics of New Zealand regions." Treasury Working Paper 00/11, Wellington NZ, New Zealand Treasury, www.treasury.govt.nz/workingpapers/ Lansing, J. B. and E. Mueller (1967). The geographic mobility of labor Ann Arbor, MI, Michigan Survey Research Center. Leeves, G. (1999). “Countercyclical Unemployment Outflows and Labour Reallocation.” Australian Economic Papers 38(3): 290-309. Le Heron, R. and E. Pawson, Eds. (1996). Changing Places: New Zealand in the Nineties. Auckland, Longman Paul. Lichter, D. T. and G. F. De Jong (1990). "The United States." In: C. B. Nam, W. J. Serow and D. F. Sly (eds.). International handbook on internal migration. Westport CO, Greenwood Press: 391-417. Lidgard, J., R. D. Bedford, et al. (1998). "International Migration from Northeast Asia and Population Change in New Zealand, 1986 to 1996." PSC Discussion Paper No. 26, Hamilton NZ, Population Studies Centre, Department of Geography, University of Waikato. Lloyd, R., A. Harding, et al. (2000). Regional Divide? A Study Of Incomes In Regional Australia. Canberra, National Centre for Social and Economic Modelling (NATSEM), University of Canberra. Long, L. (1992). “Changing Residence: Comparative Perspectives on Its Relationship to Age, Sex, and Marital Status.” Population Studies 46(1): 141-158. Lowe, R. J. (1988). "The migration factor in regional population change in New Zealand." New Zealand Population Review 14(1): 4-21. 132 Maddala, G. S. (1983). Limited-dependent and qualitative variables in econometrics Cambridge, Cambridge University Press. Maré, D. C. and J. Timmins (2000). Internal Migration and Regional Adjustment: Some Preliminary Issues. Paper presented at the Ninth Conference on Labour, Employment and Work, Institute of Geography, Victoria University of Wellington. Martin, J. (1984). "Patea Two Years After - A Review Report." Wellington NZ, Town and Country Planning, Ministry of Works and Development. Mátyás, L. (1997). “Proper Econometric Specification of the gravity model.” World Economy 20(3): 3963-3968. Mátyás, L. (1998). “The Gravity Model: Some econometric considerations.” World Economy 21(3): 397-401. Mauro, P., E. Prasad, et al. (1999). "Perspectives on Regional Unemployment in Europe." IMF Occasional Paper No. 177, Washington DC, International Monetary Fund. Mauro, P. and A. Spilimbergo (1998). "How do the skilled and the unskilled respond to regional shocks? The case of Spain." IMF Working Paper 98/77, Washington DC, International Monetary Fund. Mazumdar, D. (1987). "Rural-Urban Migration in Developing Countries." In: E. S. Mills (ed.). Handbook of Regional and Urban Economics. Amsterdam, Elsevier Science. 2, Urban Economics: 1097-1128. McCarthy, G. W. (1992). The Role of Unemployment in Triggering Internal Labor Migration, Levy Economics Institute. Milne, W. J. (1993). “Macroeconomic Influences on Migration.” Regional Studies 27: 365-373. Mincer, J. (1978). "Family migration decisions." Journal of Political Economy 86: 749- 773. Molho, I. (1984). "A dynamic model of interregional migration flows in Great Britain." Journal of Regional Science 24: 317-337. Molho, I. (1987). “The Migration Decision of Young Men in Great Britain.” Applied Economics 19: 221-243. Morrison, P. S. (1999). "Regional Labour Markets in New Zealand, 1991." Occasional Paper 1999/2, Wellington NZ, Labour Market Policy Group, New Zealand Department of Labour. Morrison, P. S., K. L. Papps, et al. (2000). "Local Labour Markets, Migration and Wage Determination: Theory and Evidence for the Wage Curve in New Zealand." GSBGM Working Paper 1/00, Graduate School of Business and Government Management, Victoria University of Wellington. Mueller, C. F. (1982). The economics of labor migration New York, Academic Press. 133 Mulder, C. H. and M. Wagner (1993). "Migration and marriage in the life course: A method for studying synchronized events." European Journal of Population 9: 55-76. Muth, R. F. (1971). "Migration: chicken or egg?" Southern Economic Journal 37: 295- 306. Nakosteen, R. A. and M. Zimmer (1980). "Migration and income: the question of selfselection." Southern Economic Journal 46: 840-851. Nakosteen, R. A. and M. Zimmer (1982). "The effects on earnings of interregional and interindustry migration." Journal of Regional Science 22: 325-341. Newell, J. (2001). "Scoping Regional Migration and its Interaction with Labour Markets in New Zealand." Occasional Paper 2001/2, Wellington NZ, Labour Market Policy Group, New Zealand Department of Labour. Nijkamp, P. and J. Poot (1987). “Dynamics of Generalised Spatial Interaction Models.” Regional Science and Urban Economics 17(3): 367-390. Nijkamp, P., P. Rietveld, et al. (1986). "Regional and Multiregional Economic Models: A Survey." In: P. Nijkamp (ed.). Handbook of Regional and Urban Economics. Amsterdam, Elsevier Science. 1: 257-294. NZPC (1989). Diversity and Change: Regional Populations in New Zealand, New Zealand Planning Council. Odland, J. (1996). "Longitudinal analysis of migration and mobility: Spatial behavior in explicitly temporal contexts." Forthcoming in: M. J. Egenhofer and R. G. Golledge (eds.). Spatial and Temporal Reasoning. Odland, J. (1997). "Longitudinal Approaches to Analysing Migration Behaviour in the Context of Personal Histories." In: M. M. Fischer and A. Getis (eds.). Recent Developments in Spatial Analysis: Spatial Statistics, Behavioural Modelling, and Computational Intelligence. Berlin, Springer-Verlag: 149-170. OECD (2000). "Disparities in Regional Labour Markets." OECD Employment Outlook. Paris, Organisation of Economic Cooperation and Development: 31-78. Oswald, A. J. (1999). The Housing Market and Europe's Unemployment: A Non- Technical Paper. UK, Department of Economics, University of Warwick. Padoa-Schioppa, F. (ed.) (1991). Mismatch and labour mobility. Cambridge, Cambridge University Press. Papps, K. L. (2000). Wage Determination in Local Labour Markets: Theory and Evidence for the Wage Curve in New Zealand. M.C.A. thesis, Department of Economics. Victoria University of Wellington. Peck, L. (1985). "A Review of New Zealand Literature Pertaining to Industrial Closedowns and Mass Redundancies: 1980-1984." Wellington NZ, Town & Country Directorate, Ministry of Works & Development. 134 Persson, J. (1994). Convergence in Per Capita Income and Migration Across the Swedish Counties 1906-1990. Sweden, Institute for International Economic Studies, Stockholm University. Pissarides, C. A. and I. McMaster (1990). “Regional Migration, Wages and Unemployment: Empirical Evidence and Implications for Policy.” Oxford Economic Papers, N. S. 42(4): 812-31. Poot, J. (1984). Models of New Zealand Internal Migration and Residential Mobility. Department of Economics. Victoria University of Wellington. Poot, J. (1986a). "The Scope, Limits and Reliability of New Zealand Census Data on Internal Migration." In: C. Crothers (ed.). The Uses and Limitations of Census Data. Wellington NZ, New Zealand Demographic Society: 91-106. Poot, J. (1986b). “A System Approach to Modelling the Inter-Urban Exchange of Workers in New Zealand.” Scottish Journal of Political Economy 33(3): 249-274. Poot, J. (1995). “Do Borders Matter? A Model of Interregional Migration in Australasia.” Australasian Journal of Regional Studies 1(2): 159-182. Rappaport, J. (2000a). Why Are Population Flows So Persistent?, Research Division, Federal Reserve Bank of Kansas City. Rappaport, J. (2000b). How does Labor Mobility Affect Income Convergence?, Research Division, Federal Reserve Bank of Kansas City. Rogers, A. (1968). Matrix analysis of interregional population growth and distribution. Berkeley, University of California Press. Roy, A. D. (1951). “Some thoughts on the distribution of earnings.” Oxford Economic Papers 3: 135-146. Saben, S. (1964). "Geographic mobility and employment status, March 1962-March 1963." Monthly Labor Review 87(8): 873-881. Schultz, T. P. (1982). "Lifetime migration within educational strata in Venezuela: estimates of a logistic model." Economic Development and Cultural Change 30: 559- 593. Sen, A. and T. E. Smith (1995). Gravity Models of Spatial Interaction Behavior. New York, Springer-Verlag. Shaw, R. P. (1985). Intermetropolitan migration in Canada: changing determinants over three decades Toronto, NC Press Ltd. Shen, J. (1999). “Modelling regional migration in China: estimation and decomposition.” Environment and Planning A 31: 1223-1238. Sill, K. (1997). Regional Employment Dynamics. Philadelphia, Federal Reserve Bank of Philadelphia. 135 Snaith, J. (1990). "Migration and Dual Career Households." In: J. H. Johnson and J. Salt (eds.). Labour Migration: The Internal Geographical Mobility of Labour in the Developed World. London, David Fulton Publishers: 155-171. Statistics New Zealand (1995). Directory of Regional Statistics. Wellington NZ, Statistics New Zealand. Statistics New Zealand (2000). "Summary of Regional Data Prepared for the Ministry of Economic Development." Wellington NZ, Statistics New Zealand. Stillwell, J., M. Bell, et al. (2000). “Net Migration and Migration Effectiveness: A Comparison between Australia and the United Kingdom, 1976-96.” Journal of Population Research 17(1): 17-38. Taylor, J. (2000). Regional Inequalities: A European Perspective. Paper presented at the 24 th Annual Conference of the Australia and New Zealand Regional Science Association International, Hobart, Tasmania. Taylor, J. and M. Bell (1996). “Population Mobility and Indigenous Peoples: the View from Australia.” International Journal of Population Geography 2: 153-169. Treyz, G. I., D. S. Rickman, et al. (1993). “The Dynamics of U.S. Internal Migration.” Review of Economics and Statistics 75(2): 209-214. Vaithianathan, R. (1995). The Impact of Regional Unemployment and Iwi (Tribal) Affiliation on Internal Migration. M. Com. thesis, School of Commerce and Economics. Auckland, University of Auckland: 149. Young, J. and R. D. Bedford (1996). "Inter-Regional Migration in New Zealand, 1986- 1991: Patterns and Profiles." PSC Discussion Paper No. 13, Hamilton NZ, Population Studies Centre, Department of Geography, University of Waikato.