Cities, growth and poverty. Evidence paper 2: Data analysis. February Paul Sissons, Neil Lee & Ceri Hughes

Transcription

1 February 2014 Cities, growth and poverty Evidence paper 2: Data analysis Paul Sissons, Neil Lee & Ceri Hughes Cities, growth and poverty: data analysis 1

2 Acknowledgements This paper is part of a wider project focusing on Cities, Growth and Poverty. The research was commissioned by the Joseph Rowntree Foundation and conducted by a team consisting of Neil Lee, Paul Sissons and Ceri Hughes from The Work Foundation, Anne Green, Duncan Adam and Gaby Atfield from the Institute for Employment Research at the University of Warwick and Professor Andrés Rodríguez-Pose from the London School of Economics. The research was undertaken in early We would like to thank the Joseph Rowntree Foundation for supporting this research, and in particular the project manager, Josh Stott. The project has also benefited from the insights of an advisory group, including Alex Jones and Andrew Carter (Centre for Cities), Ed Cox (IPPR North), Michael Parkinson (Liverpool John Moores University), Tom Aldred (The Cabinet Office), Tom McInnes (New Policy Institute), Ruth Lupton (London School of Economics), and Tom Bridges (Leeds City Region). We are grateful for their support and would like to thank them for contributing to this project. We would like to thank Alex Fenton, Ruth Lupton and Amanda Fitzgerald for providing access to the Unadjusted Means-tested Household Benefit Rate (UMBR). We have also benefited from discussions with them with regard to the data and the wider implications of the project. Lloyd Martin provided very good research assistance in the early phases. Katie Schmuecker, Mike Campbell and Christine Appleton also provided valuable comments and support. In addition we would like to thank Tony Champion (Newcastle University) for reviewing an earlier draft of this paper. This study presents analysis which draws on a number of different data sources. Both Labour Force Survey and Annual Survey of Hours and Earnings data (Office for National Statistics) were accessed through the Secure Data Service (Data Archive). Administrative data on benefits as well as VAT statistics and Mid-Year Population Estimates (ONS) were accessed through NOMIS. GVA data were accessed via the ONS, and Land Registry data was accessed via The data we have used is Crown Copyright and has been made available by the Office for National Statistics (ONS) through the Data Archive and through NOMIS. It is reproduced with permission. Neither the ONS, NOMIS, nor the Data Archive bears any responsibility for the analysis and interpretation of the data presented in this study. The authors would also like to acknowledge the help of the Secure Data Service in accessing data. Cities, growth and poverty: data analysis 2

3 About the authors Paul Sissons contributed to this analysis as a Senior Researcher at The Work Foundation. He is a Senior Research Fellow at Coventry University. Paul s research interests include labour markets, welfare reform and local and regional economic development. Neil Lee led on this research as Head of the Socio-economic research centre at The Work Foundation. Neil is an Assistant Professor in Economic Geography at the London School of Economics. His work focuses on cities and the social dimensions of economic change. Ceri Hughes is a Research Assistant at The Work Foundation. Her research has focussed on aspects of labour market disadvantage. Cities, growth and poverty: data analysis 3

4 Contents Acknowledgements... 2 About the authors... 3 Executive summary Introduction Methods and data sources Economic Growth in British cities in the 2000s Poverty in British cities in the 2000s The relationship between economic growth and poverty Growth, wages and costs Conclusions References Appendix I - Regression results Appendix II - Rankings and values of poverty proxies Cities, growth and poverty: data analysis 4

5 Executive summary The Joseph Rowntree Foundation commissioned The Work Foundation, the Institute for Employment Research and the London School of Economics to undertake a project on the links between cities, growth and poverty. The main project report Cities, growth and poverty: a review of evidence is available from the Joseph Rowntree Foundation 1. This report outlines the results from the data analysis phase of the research. The aim of this phase was to examine empirically the link between growth and poverty in British cities. To do this we have assembled a dataset to assess the economic and poverty-related performance of the Travel-To-Work-Areas of the 60 largest British cities over the period 2001 to 2010 (with most of the analysis focusing on the period from 2001 to 2008). We consider both employment and output (Gross Value Added, or GVA) growth and two poverty proxies. These proxies utilise administrative data on benefits which correlate strongly with income based poverty definitions. The aims of this paper are to: Outline broad national trends in growth and poverty, providing background to the city-level analysis which is presented; Examine patterns of growth in British cities between 2001 and 2010 and consider some of the drivers of this growth; Analyse how different forms of growth are linked to poverty reduction; Explore how growth affects other factors, such as wages and housing costs, which potentially impact on poverty rates Key Findings There was continuing divergence in GVA performance during the 2000s. The fastest growing cities between 2000 and 2008 were the ones with the highest GVA per head at the beginning of the decade. The drivers of growth over this period included high skills and entrepreneurship. Our analysis shows the importance of the rate of new firm creation for growth as well as having a highly skilled workforce. There was some convergence in poverty rates during this period. A number of the cities which had some of the highest levels of poverty in 2001 also saw the largest comparative reductions over the period to Post-recession, the data suggest some divergence in the period , with the weakest cities most affected by the recession 1 Link: Cities, growth and poverty: data analysis 5

6 Convergence in poverty rates narrowed the gap somewhat between different cities but did not close it the geography of poverty changed very little during the 2000s. Despite a slight convergence in poverty levels, the rank order of places by poverty level was largely unchanged over the decade. Cities like Liverpool and Glasgow remained the poorest on both our measures. Places with high GVA per head tend to have lower poverty. Poverty tends to be lowest in the cities around London and considerably higher in many of the former industrial towns and cities in parts of northern England. There is a strong relationship between a higher employment rate and lower poverty. Demographic and social characteristics are also important in influencing rates of poverty. Employment growth leads to reductions in poverty. Panel data show a positive relationship between total employment growth and poverty reduction (although the exact strength of this relationship is difficult to judge from our data). In lagging cities the effect of employment rate growth on poverty reduction is stronger. Lower skilled workers are also shown to benefit most from local employment growth. GVA growth is positively associated with increasing wages for some as well as with house prices. Stronger GVA growth during the 2000s is associated with increasing wages for high earners but not those towards the bottom of the wage distribution. It is also associated with faster growth in house prices thereby raising living costs, with implications for those in poverty. Cities, growth and poverty: data analysis 6

7 1. Introduction The Joseph Rowntree Foundation commissioned The Work Foundation, the Institute for Employment Research and the London School of Economics to review the evidence on the links between cities, growth and poverty. The research project followed a four-stage methodology, including the development of a framework, a review of the evidence, data analysis and case study work in ten UK cities. In addition to this evidence paper, which describes the data analysis phase, evidence papers relating to the evidence review (Evidence Paper 1) and case studies (Evidence Paper 3) are available at Background The aim of this paper is to provide new data on growth in British cities and on the links between growth and poverty. It is important to understand these links in order to assess the extent to which growth and, specifically, different types of growth can lead to poverty reduction. It is also important to understand how and why this might work differently in different types of cities. The analysis can help inform the policy debates around two important areas: The role of cities as drivers of growth. Cities are increasingly being seen as important actors in the growth agenda, and new City Deals are devolving selected powers and funding to cities to facilitate and drive growth locally. Who gains from growth? Rising income inequality and the prevalence of low-wages in UK employment has lead many commentators to question whether the proceeds of growth are distributed fairly. The drivers of growth Cities are now commonly seen as the drivers of the national economy. They possess a number of structural features which are regarded as increasingly important in enabling innovation and, as a consequence, greater productivity, employment, and economic growth more generally. Cities bring together a number of important factors which influence development. The drivers of growth encompass four main areas: Enterprise and innovation such as the number and nature of new firms and the extent to which the local economy is innovative. Human capital the skills of the workforce and their ability to enter work, including both high skill levels and vocational qualifications which may help people into well paid employment. The physical environment the nature of the built environment, local amenities, the availability of new housing and commercial land, and transport links within and between cities. Leadership and governance the extent to which local leadership can help drive growth in a city, and the financial resources available to them for doing so. Linking growth and poverty Cities, growth and poverty: data analysis 7

8 Cities can also influence the relationships between growth and poverty at a local level. Predominately this local relationship is shaped by two factors 1) the scale of employment growth; and 2) the distribution of gains from employment growth (in other words who gets the jobs?). It is worth noting here that local areas do not posses all, or even most, of the important levers which influence the relationship between growth and poverty. Of particular importance is the national tax and benefit system, which may be independent of employment change, and which impacts strongly on poverty in one direction or the other. However, there are a number of policy areas where local actors can play an important role, including: Infrastructure development which can involve provision of local public transport services that link residents to jobs The planning system can be used to influence the location of jobs. It can also be used to shape the numbers, types and location of housing developments to help reduce housing costs. Local job brokering and matching services - for example pre-employment training and employer-led training focused on specific opportunities. Local skills programmes can also be important in enabling more of the population to benefit from employment growth. Community benefit clauses which can provide employment and training opportunities, linked to specific local development projects. Promotion of a local Living Wage to improve wages for those in low-wage work and ensure that residents benefit from growth. A more detailed review of the evidence on the drivers of growth and the links between growth and poverty is provided in Evidence Paper 1. Meanwhile, the case studies presented in Evidence Paper 3 highlight how some individual cities have attempted to link people with opportunities in the local area. 1.2 Report outline In this paper we present the results from new empirical work which examines the drivers of economic growth and the links between growth and poverty in British cities over the period In the following sections we: Examine patterns of growth and the drivers of this in British cities between 2001 and 2010 Analyse how different forms of growth are linked to poverty reduction Explore how growth affects other factors, such as wages and housing costs, which also potentially influence poverty rates While the aim of examining the links between growth and poverty might sound relatively straightforward, issues of data availability make it challenging in practice. In particular, there is no local income-based measure of poverty which is available over time. In this paper we analyse the links between growth and poverty using a number of poverty proxies - measures which correlate strongly with income-based definitions of poverty. Of course, these measures have limitations; in particular they only cover a specific subset of the population who are likely to be in poverty. However, they represent the most effective means currently available of attempting to estimate the approximate scale and direction of poverty change at the city-level during the period covered by our analysis The main poverty proxies we use are: Cities, growth and poverty: data analysis 8

9 The Unadjusted Means-tested household Benefit Rate (UMBR) - a measure collated from benefits data by researchers at the Centre for the Analysis of Social Exclusion (CASE) at the LSE A measure of the proportion of households claiming Housing Benefit, a means-tested benefit for low income households who are renting their home. Data have also been drawn together from a range of sources to create a panel dataset of indicators and influences on growth and poverty. We also present measures of costs that are associated with growth in the form of house prices and look at changes in these relative to wages. Our focus is on the 60 largest cities in Great Britain, as identified in the State of the Cities report. The focus of the analysis is largely on labour markets which we define here as Travel-To-Work-Areas (TTWAs). The data presented covers the period from 2001 to 2010, with the analysis of poverty separated into two periods. The first a period largely characterised by economic growth ( ), and the second a period of recession and stagnation ( ). The report is structured as follows: Section 2: provides a description of our methods, data sources and the measures and indicators that we use Section 3: presents a descriptive analysis of growth in cities over this period and examines the drivers of this growth Section 4: provides a descriptive analysis of spatial trends in poverty during the 2000s Section 5: examines the links between growth and poverty at the city level Section 6: analyses the impact of growth on wages and on housing costs Section 7: provides a summary of the research findings and discusses some of the implications for policy Cities, growth and poverty: data analysis 9

10 2. Methods and data sources In this section we outline our methods for investigating the links between growth and poverty. We also provide details of the data sources which are used to inform the analysis. We have constructed a panel dataset, with observations for each year between 2001 and 2010, drawing together data on cities from a number of different sources. The dataset brings together information about city-level growth, poverty, demographics, labour markets, human capital and housing costs. 2.1 Travel-to-Work Areas The data are analysed for the 60 largest British cities as defined in the State of the Cities report plus Swansea, Cardiff, Edinburgh, Glasgow and Aberdeen 2. As much of the focus is on labour markets we use city Travel-To-Work-Areas (TTWAs). These are defined to approximate to functional labour markets, and were originally constructed using data from the 2001 Census. They describe areas in which a minimum of 75 per cent of the economically active population work and live. As a result, Travel-To-Work-Areas do not map directly onto administrative boundaries. The mechanics of the construction of TTWAs used here is slightly different from that which can be produced using small geographies data which is primarily available from the Census and from benefits data. As we rely on a number of data sources where local authorities are the smallest spatial scale for which data are available, our TTWAs are necessarily built up from groups of these. There are different ways in which this aggregation can be approached. Our approach was to use a simple majority 0.50 point to define inclusion/exclusion of local authorities in a Travel-To-Work-Area. This is an iteration of the approach used by Kaplanis (2010).This means that the figures presented here are not directly comparable to those TTWA analyses which are built up from the most detailed geographies. 2.2 Measures of economic growth For the analysis of economic growth we primarily use specially constructed estimates of Gross Value Added (GVA) for cities. GVA is a measure of economically productive activity which is estimated and compiled by the Office for National Statistics. The sub-national estimates are compiled using an income approach which sums the income generated by resident individuals or corporations in the production of goods and services (ONS, 2011). The lowest level of spatial disaggregation for which GVA estimates are published is the NUTS 3 2 Northern Ireland is not included in the analysis because data for a number of key variables is missing Cities, growth and poverty: data analysis 10

11 level 3. Although this scale is smaller than many of our cities it does not map consistently onto the TTWA geography. We have therefore estimated local authority shares of GVA from the published NUTS 3 level estimates using a ratio of GVA by sector and employment by sector in each local authority. These were then aggregated up to TTWAs. In the analysis we provide data for total GVA as well as GVA per worker, GVA per head and GVA per adult, a measure of GVA divided by the population aged (to allow for comparison across cities of different size). All local GVA data is subject to error and this needs to be considered when interpreting results. Similarly, there may be errors in the sectoral comparisons made and the use of the share allocation we adopt may increase these, particularly with highly productive but low-employment sectors such as extractive industries. While this is the best and only current possible measure of GVA at city level, these limitations need to be considered when interpreting the results. In particular the GVA figures should be viewed as indicative rather than actual values. In addition to GVA growth we also look at employment growth and population growth in cities. To do this we use employment estimates from the Annual Business Inquiry and The Labour Force Survey, and population estimates from the Office for National Statistics mid-year population estimates. 2.3 Measures of poverty There are a number of ways in which poverty can be measured (for a discussion of recent trends across a number of measures see Cribb et al. 2013). Poverty may be measured in terms of material deprivation, the ability to meet basic needs, but it is more commonly expressed (at least in Europe) in relative terms and as some fraction of national median income and with adjustments made for differences in household sizes. In the UK the relative poverty measure is set at 60 per cent of national median household income (adjusted for household size). There are some significant constraints in estimating the level and rate of poverty for sub-national areas. This is particularly the case when change over time is of interest. There is no perfect single measure of sub-national trends in poverty on which to draw. With no income-based measures available which capture change over time. The Office for National Statistics (ONS) have previously estimated model-based small area statistics for income poverty, but these estimates only cover the year 2007/2008 so are unsuited for our purposes 4. In this research we therefore use a number of sources to provide proxy measures for poverty at the local level. These are measures that tend to correlate strongly with income-based measures of poverty. The main measures that we report are: The Unadjusted Means-tested Benefits Rate (UMBR), a poverty proxy developed by researchers at the Centre for Social Exclusion at the London School of Economics The proportion of households claiming Housing Benefit 3 NUTs is an abbreviation of the official term - Nomenclature of Units for Territorial Statistics. NUTS 3 is a level of country subdivision defined by the European Union and used for statistical purposes 4 Tests comparing the measures we use to the ONS model-based estimates find them highly correlated. In fact the majority of our cities on the UMBR measure lie within +/- 3 percentage points of the mean of the central estimate of poverty rates of MSOAs contained in the 60 British cities. However UMBR estimates are lower than the model-based estimates in the lowest poverty cities including Guildford and Aldershot, Southampton, Worthing and Oxford and Cambridge. Cities, growth and poverty: data analysis 11

12 Further details of these measures are provided below. In addition to these two measures, which are available over time, we also present estimates of child poverty from HMRC data. These are again drawn from benefits data but are only available for the latter part of the period under consideration. The possibility of using data from the Family Resources Survey (FRS) and Households Below Average Income (HBAI) was also investigated but these were found to be unsuitable for analysis at the TTWA scale because of the method of survey sampling and the sample sizes. Unadjusted Means-tested Benefits Rate (UMBR) The first measure we use is a poverty proxy which had been developed previously by researchers at the Centre for Analysis of Social Exclusion at the London School for Economics. The measure the Unadjusted Means-tested Benefits Rate (UMBR) represents the proportion of households which claim a number of major means-tested benefits in our group of cities. The benefits included in the measure are Jobseeker s Allowance 5, Employment Support Allowance, Income Support, and Pension Credit. These are benefits that tend to correlate strongly with wider income-based measures of poverty. This approach is similar (albeit with a number of exclusions) to that used in calculating the Income Domains of the Indices of Multiple Deprivation (IMD). The UMBR measure was designed to investigate change at the very local level using Super Output Area geographies. For our purposes we have aggregated these up to city TTWAs. A full description of the methods used to calculate the proxy and a discussion of its strengths and weaknesses is provided by Fenton (2013). The UMBR provides a reasonable proxy for local poverty but is an imperfect measure for a number of reasons. The relationship between UMBR and income poverty is linear but it is not constant - twice the UMBR does not mean twice the poverty rate (Fenton, 2013). The UMBR can also reduce in some circumstances without there being a related reduction in income poverty. For example, an individual moving into a low-paid job would exit the UMBR measure but may remain in poverty. The measure also does not account for the different costs of housing in different parts of the country. Finally, the measurement of change takes a static base rather than a moving one which forms the basis of relative poverty figures. There are a couple of discontinuities in UMBR over time. One of these changes is the introduction of a new Pension Credit in Pension Credit extended the availability of benefits to greater numbers of people over or approaching the state pension age. One implication of this is that the data are affected by differences in demographics, take-up rates of these benefits and the income distribution of the older population between cities. We tested for the effect of this by removing Pension Credit from the figures and found that the distribution of cities and their position relative to each other changed little. We also examined underlying trends in Income Support and Pension Credit which suggest that the change adds a relatively small amount to the total poverty proxy count in all cities. In general, somewhat smaller increases are observed in high poverty cities and larger increases in low poverty cities. The magnitude of these differences are however relatively small. The average increase for the ten highest poverty cities was 0.37 percentage points, for the ten lowest poverty cities it was The UMBR measure also includes JSA claims made on the basis of National Insurance contributions Cities, growth and poverty: data analysis 12

13 This change does not alter the substantive patterns observed, however it may be associated with some small changes in the rank order of cities. A second change is the introduction of Employment Support Allowance (ESA) to replace Incapacity Benefit (IB) which affects the most recent years of data. IB was not included in the UMBR estimate but ESA is because of the somewhat different financial profile of IB and ESA claimants and the relative benefit rates associated with the benefits (IB was typically more generous, particularly for those who had been on it for a long-time). This change has a very limited impact on the 2008 figures where ESA claims account for less than 0.40 percent of the UMBR measure in every city. They have a greater effect from 2009 onwards, in particular as former IB claimants are reassessed and placed onto ESA. Housing Benefits We also run the analysis with a second measure, using Housing Benefit data, which helps capture those who are in-work and on low incomes. Housing Benefit is available to those who are in rental accommodation and who are on certain benefits or in-work and on a low income. As access to Housing Benefits is strictly means-tested but is not solely provided to those outside the labour market it is likely to be the most suitable single poverty proxy (Fenton, 2013). An obvious shortcoming of the measure however is that it only covers people in rented accommodation; it therefore excludes owner occupiers in poverty and means that differences between areas will in part reflect differences in tenure as well as poverty. As with UMBR, we use an estimate of the proportion of households in each city who are claiming Housing Benefit and how this has changed. Housing Benefits data for local authorities are only available from 2003 onwards so our analysis here is restricted to the period Between 2007 and 2008 there was a change in the way in which Housing Benefit data were compiled, moving from a system of clerical returns from Local Authorities to one which is drawn from the Single Housing Benefit Extract (SHBE) data source. While we cannot isolate the impact of this change from wider changes in the labour market and population in the cities we study, it appears that this change was on average associated with a small increase in Housing Benefit counts, but with no consistent pattern of difference between high and low poverty areas. This change does not alter the substantive patterns observed in the paper which are consistent with patterns of change across earlier years of the data ( ), although it may be associated with some small changes in the rank order of cities and the level of percentage points change estimated. Child Poverty We also use a third poverty measure using HMRC estimates of child poverty to check for correlation with our other measures. The child poverty estimates are compiled using information on receipt of out-of-work means-tested benefits and tax credits. Rather than being expressed as a proportion of households, child poverty data is expressed as a proportion of all children aged 16 and under. This data is only currently available from 2008 onwards for England and 2009 onwards for Scotland and Wales. Cities, growth and poverty: data analysis 13

14 2.4 Assessing the proxies Both the proxy measures we use to estimate change in poverty in different cities over time (UMBR and Housing Benefits) have obvious limitations. They do not correlate precisely to income poverty. They only partially cover the population who are in relative poverty, and also include some households who are not in poverty. Measures of the extent to which the proxies accurately capture the population we want them to (those in households in poverty) are known as the validity and coverage associated with the proxies. Table 1 provides estimates for the validity and coverage for both measures in These are calculated by Fenton (2013) using national aggregates of the data compared to estimates from the Family Resources Survey (FRS) data on income and benefits. The coverage is the proportion of all income-poor households identified by each proxy. The validity is the proportion of households captured by each proxy who are actually in relative income poverty. The benefits data from the FRS is likely to underestimate the coverage of the measures; this is because, for a number of reasons, people are less likely to report details of benefits claims in surveys (Fenton, 2013). Previous research has suggested that under-reporting of Housing Benefit by private tenants in official surveys can lead to estimates that are half the size of that recorded in administrative records (Fenton, 2010). It is therefore highly likely that the true coverage figures are higher than those presented below. For the UMBR, the estimates of validity are 47 and 62 per cent respectively for poverty measures using 60 and 70 per cent of median income. The coverage is lower, at 30 per cent, for both levels. For Housing Benefits the validity is higher, at 54 and 70 per cent respectively. The coverage is similar to that for UMBR, at 29 and 28 per cent. For the population of working-age only, the validity is higher for both these measures at 62 and 70 per cent. Table 2.1: Estimates of validity and coverage of proxy measures for whole population (working-age in parenthesis) <60% median AHC <70% median AHC Validity Coverage Validity Coverage UMBR 47 (62) 30 (32) Housing Benefits 54 (70) 29 (31) (Source: Fenton, 2013) It is important to stress that the measures we use are proxies; they indicate rather than measure (relative) poverty and change in poverty. They only partially cover the population who are in poverty and also include households who are not in poverty. Tests on the UMBR data for different years at the national and regional level do however show that a lower UMBR rate is strongly associated with a lower incidence of income poverty. Fenton (2013; 52) provides an approximation for this relationship, calculated using a range of correlation relationships for regions and area types, and estimating that on average a 1 per cent increase in UMBR implies approximately a 0.4 per cent increase in the income poverty rate. In summary, while the poverty proxies which are used do have limitations, they remain the best estimate of the direction and likely scale of poverty change which is currently available at the city-level. Cities, growth and poverty: data analysis 14

15 An additional and important challenge in relation to understanding changes in poverty through the proxy measures is to isolate the drivers of change at city-level. Changes in the measures can be driven both by expansion in opportunities but also by changes in the nature of the benefits safety net, which may have a differential impact across cities. Reductions can be driven by changes in the labour market, for example more local people moving into employment and ceasing to claim benefits. They can also be influenced by changes within the benefits system, including to the means-test rate and other eligibility criteria. It should be noted however that until 2008 there were no major changes affecting eligibility for the benefits included in our measure (apart from those already discussed). From 2008 onwards changes to benefit rules for lone parents and reforms to sickness benefits may have a greater influence on the figures. The proxies can also be affected by population and demographic changes in cities in different ways. A wider point is that it is difficult to isolate the result of local change from wider policy impacts - for example, the impact of national welfare to work provision, which is likely to vary across cities. 2.5 Variables which influence the relationship between growth and poverty variables We have also used a number of other data sources to give a fuller picture of the relationship between economic growth and poverty. These include data from the Labour Force Survey, Annual Business Inquiry, VAT registry data, the Annual Survey of Hours and Earnings, Mid-year population estimates, and Land Registry data. The variables combined in the dataset provide measures of: Labour markets for example change in the levels and rates of employment in different cities, the sectors in which employment is growing, and the relationship between public and private sector employment growth. Wages we examine the extent to which GVA growth influences wage growth in different parts of the wage distribution. Human capital the level of skills in the local population is an important driver of growth, it also influences the extent to which the local population is able to capitalise on, and benefit from, employment growth. Population characteristics and change population growth has important implications for growth within cities. The characteristics of the local population are also important in influencing growth and the relationship between growth and poverty. We also look at what impact migration has on growth and poverty. Costs growth may also create costs for the local population. We look at the relationship between growth and house prices. Enterprise we look at measures of enterprise and innovation (including the number of patents registered locally, new business registration, and the level of self-employment) to analyse their influence on growth. 2.6 Methods We provide descriptive and multivariate analysis of our indicators. Cross-sectional regression models are used to examine the static relationship between GVA and poverty in 2008 and the characteristics of places that appear to influence levels of poverty in cities. Panel regression models, which look at change in variables over time, are then used to analyse the links between growth and poverty, growth and wages, and growth and housing costs. Where panel models are used we use the longest run of data covering the period for the variables included. Further details on the specifications of Cities, growth and poverty: data analysis 15

16 these are provided in the relevant sections and full output is provided in Appendix I. Cities, growth and poverty: data analysis 16

17 3. Economic Growth in British cities in the 2000s The UK economy experienced a period of relatively strong and uninterrupted growth from the early 1990s up to the onset of recession in Between 1997 and 2010, GDP per capita growth was faster in the UK than that experienced in France, Germany, Italy, Japan and the US; while productivity growth was second only to the US over the same period (Corry et al, 2011). The UK also experienced relatively strong growth in the labour market, with both total employment and the employment rate increasing. Since the onset of the recession there has been much debate over the extent to which this period of growth was balanced and sustainable (see BIS, 2010; Corry et al, 2011). In this section we examine the indicators of economic growth in British cities during the 2000s, and the drivers of this growth. Our analysis mainly focuses on GVA growth which has been calculated for cities using the method specified in Section 2. Growth figures are provided as totals and per person. The per head figure allows us to equivalise the GVA figures to reflect the very different size of cities in the analysis. We also provide evidence on the link between output and employment growth in British cities. 3.1 GVA growth in cities Levels of economic output are highly uneven between cities. Differences in total Gross Value Added (GVA) a measure of economic output reflect both city size and city performance. London is by far the largest city; it contains around 23 per cent of the total population of the 60 cities. Its dominance is even more pronounced on the GVA measures. GVA in London is equivalent to around one-third of total GVA generated by the 60 cities. Figures on GVA per head also vary substantially across the 60 cities. Our estimates for 2008 are provided in Table 3.1, it should be noted that because of the potentially large confidence intervals associated with sub-national GVA figures, our city level estimates should be viewed as indicative rather than exact. In London, GVA per adult is estimated at more than 48,000. Other highly ranked cities on this indicator are those which are close and/or well connected to London (including Luton & Watford, Reading and Bracknell, Guildford and Aldershot, Milton Keynes and Aylesbury), as well as some larger cities (Edinburgh, Newcastle). Aberdeen also does well primarily because of its offshore sector. Cities with weaker economies tend to be in the North of England (including Barnsley, the Wirral and Ellesmere Port, Rochdale and Oldham, Hastings, Burnley, Nelson and Colne). GVA per head in these cities is less than half that for the top ranked cities. These large differences in GVA matter because GVA is a measure of the extent to which wealth is created locally. It is an important indicator of the overall health of a local economy and over time rising output growth is related to rising standards of living. These large differences therefore represent sizeable gaps in the relative economic performance of British cities. Cities, growth and poverty: data analysis 17

18 Table 3.1 GVA figures for selected cities, 2008 Total GVA ( ) GVA per adult ( ) 2008 Ranked by GVA per adult ( ) 2008 Luton & Watford 17,634 49,436 1 London 292,205 48,459 2 Reading & Bracknell 13,314 46,102 3 Aberdeen 13,068 43,272 4 Edinburgh 18,611 41,804 5 Guildford & Aldershot 18,432 40,778 6 Newcastle & Durham 21,072 39,138 7 Milton Keynes & Aylesbury 10,541 39,114 8 Gloucester 2,940 38,784 9 Crawley 14,016 38, Blackpool 4,578 22, Doncaster 4,153 22, Hull 8,693 22, Huddersfield 5,841 22, Stoke-on-Trent 6,394 21, Burnley, Nelson & Colne 2,397 21, Hastings 2,207 21, Rochdale & Oldham 5,605 20, Wirral & Ellesmere Port 3,482 18, Barnsley 2,438 16, Source: Authors' estimates based on ONS GVA statistics, ONS mid-year population estimates. Note: these figures should be seen as indicative rather than exact 3.2 Change in GVA Looking beyond the static picture of GVA in British cities provided in Table 3.1, this section examines patterns of GVA growth during the 2000s. London accounted for around 37 per cent of total nominal GVA growth that occurred across the 60 cities over the period. This is more than 11 times the contribution to total city growth made by Manchester, the city with the second highest level of total GVA growth. Figure 3.1 presents change in GVA per head between 2001 and In general GVA per head Cities, growth and poverty: data analysis 18

19 Change in GVA per adult ( nominal) growth was greatest in those cities which already had high levels of GVA per head in 2001, with continuing divergence between high and low GVA cities over the period. The cities that have seen the greatest increases in GVA per head include London, Reading and Bracknell, Milton Keynes and Aylesbury, Edinburgh and Aberdeen. Those that have seen the weakest growth include Barnsley, Burnley, Nelson and Colne and Blackpool. The patterns for percentage change in GVA per head over the period are also similar, with little evidence of catch-up by lower GVA cities. Figure 3.1: Change in GVA per head, Milton Keynes & Aylesbury London Aberdeen Luton & Watford Edinburgh Reading & Bracknell Blackpool 2000 Barnsley Burnley, Nelson & Colne GVA per adult ( ) 2001 Source: Authors' estimates based on ONS GVA statistics and ONS mid-year population data 3.3 Drivers of growth The Evidence Review (Evidence Paper 1) outlined a number of factors which tend to drive or be supportive of economic growth in cities. The importance of a number of these has been assessed for British cities for the period The basic method of estimation used is fixed effects panel regression models. In a number of models we have also used lagged variables to investigate the extent to which a change in an initial year might influence the output variables one or two years later. In most cases the models are also presented both with and without controls. The variables included in the models are measures of factors which the literature suggests are likely to be important drivers of growth. It is worth stressing that we cannot model all the possible drivers of growth because of data availability. In the models we test measures of the following: Cities, growth and poverty: data analysis 19

20 Enterprise and innovation, using measures of: registrations of patents a measure of firm innovation; self-employment a measure of entrepreneurship; and, VAT registrations a measure of new firm creation. Human capital, using measures of: the proportion of the population with a degree or equivalent qualification or above a measure of the highly qualified population; the proportion of the population born abroad a measure of diversity. An additional driver of growth which is included in the models is population growth. The output variables of the models are various measures of economic growth including total GVA, GVA per head, and GVA per worker. The full model results are provided in Appendix I. In summary, the models show that new firm creation is positively associated with all measures of economic growth included in the analysis (Table A). However, self-employment has no association with growth on any of the measures used (Table B). This is likely due to the fact that the self-employed are such a heterogeneous group. Many selfemployed workers do not enter self-employment to create new, growth-oriented companies, but instead for lifestyle reasons or because they cannot find work elsewhere in the labour market. We also find some evidence that innovation leads to output growth, but no evidence it is associated with employment growth (Table C). The models also show that, as would be expected, having a highly qualified workforce is important for urban economic performance. However, the results are not as clear as we would expect. Table D gives results of the basic regressions, in which there is no effect from an increase in the share of the workforce with degree level qualifications and measures of output growth. During the period we consider, the public sector was expanding relatively rapidly increasing the number of graduate jobs in places which were experiencing weaker economies (Wright, 2010). As this would counter the growth effect, it would bias down the coefficient. To assess this, we use an alternative method of estimation Arellano-Bond in columns 5 8. This method better controls for the endogeneity of increases in human capital and shows a positive effect. 3.4 Output growth and employment growth The extent to which output and employment growth in cities are linked has important implications for poverty. Employment tends to protect individuals from poverty relative to being out of work. As part of the empirical work for this paper, we examined the relationship between output and employment growth (Table E). The findings suggest, as would be expected, that GVA growth (on all measures) and total employment growth are positively associated, but that increases in the employment rate lag GVA growth by around a year. This relationship holds whether GVA is measured on a total, per capita or per worker basis. The link between output and employment growth is intuitive, but needs some empirical caveats. Increases in output will not be the only determinant of the employment rate, which will also be determined by factors such as the sectoral composition of the economy. Moreover, increases in output will not inevitably lead to employment and this employment will not always be compatible with poverty reduction. For example people in poverty may not able to access new employment opportunities due to skills mismatch. The size of the effect is also unclear, and we do not know how Cities, growth and poverty: data analysis 20

21 much more output would be necessary to create a certain number of new jobs. These are all important areas for future research. Cities, growth and poverty: data analysis 21

22 Percentage in poverty (%) 4. Poverty in British cities in the 2000s Between 2001 and 2008, relative poverty rates 6 for the UK were largely static at around 18 per cent of households. However, from 2008 onwards these rates began to decline. The decline in poverty since the recession appears counterintuitive. But rather than indicating an improvement in living standards for those at the lower end of the income distribution, incomes at the lower end fell, but by less than median incomes. Absolute poverty, which measures the change in poverty from a base year (in this case 2010/2011); fell more significantly over the period, from 22 per cent in 2000/2001 to 15 per cent in 2009/2010 before beginning to rise again. Figure 4.1 Percentage of individuals in relative and absolute low-income, Before Housing Costs (BHC) in the UK / / / / / / / / / / / /12 Relative low income (BHC, 60%) Absolute low income (BHC, 60% 2010/11 median) Source: FRS HBAI methodology, DWP (2013) The poverty rate varies considerably depending on the economic status of households. The risk of poverty is greatest for those who are out of work. Working-age adults in workless households have a much higher risk of poverty than those in households where at least one adult is in work (at 47 per cent compared to 21 per cent in 2011/12). 6 Which capture the proportion of households whose income, equivalised for household size, falls below 60 per cent of the median Cities, growth and poverty: data analysis 22

23 Yet being in employment does not mean that an individual will escape poverty. In 2011/12 60 per cent of working-age adults in poverty lived in households where at least one member of the household was in employment 7, compared to 40 per cent living in workless households 8. This is an important point which reinforces the fact that tackling poverty is not just about moving people into employment, it is also about policies which reduce poverty among those who are working (for example minimum wages and in work credits). Within this broad national picture of poverty in the UK there are pronounced differences between individual regions and cities. It is these city-level differences which are the subject of the remainder of this section. 4.1 Poverty in British cities We now consider how our poverty proxy measures differ across cities and how they changed during the 2000s. Table 4.1 shows each of our poverty measures for high and low ranked cities in Both UMBR and the Housing Benefit measure are expressed as a proportion of all households, the child poverty data expresses the estimated proportion of children aged 16 and under living in poverty. The measures are presented together firstly to set-out the scale of poverty on these measures, and secondly to see whether there is significant variation in the distribution of the cities on the different measures (i.e., do they seem to be measuring the same thing consistently). In general the measures are very highly correlated with each other 9. Looking at the rankings, the first obvious observation is that poverty, across all measures, varies very significantly across cities. In general cities with lower proxy measures tend to be the more affluent cities around London, while former industrial towns and cities generally have much higher rates. The UMBR measure ranges from 9.7 per cent in Guildford and Aldershot up to 34 per cent in Liverpool; the Housing Benefit measure ranges from 9.0 per cent of households in Guildford and Aldershot to 24.4 per cent in Glasgow. There is also a significant range for the child poverty measure, from 9.1 per cent in Guildford and Aldershot up to 23 per cent in Birmingham. Table 4.1: Measures of poverty in cities, 2008 UMBR rank UMBR Housing Benefit Child Poverty (Scotland and Wales data are for 2009) measure Guildford & Aldershot Crawley HBAI methodology includes working pensioners in the data, so the economic status of a household is determined on the basis of the employment status of all working-age adults and only includes those pensioners who are still working 8 Linked to the fact that the majority of working-age adults live in households where someone is in work 9 The correlation between the UMBR and the Housing Benefit measures is.944, between UMBR and child poverty it is.929. All are statistically significant at the 0.01 level Cities, growth and poverty: data analysis 23