Geographical inequalities in labour market outcomes
Evidence on attitudes to inequality in the UK published by the IFS Deaton Review last year (Benson, Duffy, Hesketh and Hewlett, 2021) has shown that geographical inequalities have become perhaps the area of most concern to the general public. In recent years, geographical inequalities have also become important issues in driving political debate, and the ‘levelling-up’ agenda – the subject of a much-anticipated recent White Paper – has risen to the top of the government’s stated policy aspirations. So geographical inequalities, and what should or might be done about them, will naturally form an important component of the deliberations of the IFS Deaton Review panel.
A panellist’s introduction, by James Banks
As part of the Review’s background evidence on inequalities and the COVID-19 pandemic, we have already published work that gives some big-picture evidence on key geographical economic inequalities in the UK and how they changed in recent years leading up to the pandemic (Agrawal and Phillips, 2020). The evidence revealed a complex story of changes since the early 2000s, with different stories at different levels of geographical granularity and for different outcomes such as productivity, earnings, total incomes before or after adjusting for housing costs, and wealth.
To this we are now adding some more in-depth analysis of spatial differences in labour market outcomes, the drivers of such differences, and implications of this for the design of economic place-based policies. An evidence chapter by Henry Overman (LSE) and Xiaowei Xu (IFS) provides a detailed study of trends in, and drivers of, spatial inequalities in wages and employment rates over the period 1998–2019 in the UK. Maarten van Ham (Delft), David Manley (Bristol) and Tiit Tammaru (Tartu) contribute an accompanying piece giving a geographer’s perspective on the drivers of patterns in socio-economic spatial inequality. Finally, Enrico Moretti (Berkeley) contributes a piece briefly setting some of the Overman and Xu findings in broader international context and then going into some detail about the economic consequences of such findings for the debate around so-called place-based policies.
The extent of spatial inequalities depends inversely on the size of the areas being analysed. Since they focus on labour market outcomes and want to be able to link datasets over time, Overman and Xu use 136 areas to cover the UK, based on local authority approximations to travel to work areas (TTWAs, sometimes called commuting zones in other international contexts). To the extent that local economic conditions drive local labour market outcomes, these are most naturally expected to operate at the TTWA level, which makes TTWAs a natural unit of analysis for wages and employment. But many other geographies may matter (as van Ham and co-authors point out) when considering broader outcomes or thinking of the population’s experiences of, or attitudes to, spatial inequality.
Overman and Xu document and quantify several key patterns and trends for the UK:
- Geographical inequalities in wages are larger for smaller geographical units. At the individual level, the top 10% of earners earn, on average, 4.8 times higher wages than the bottom 10%. Using the Lower Layer Super Output Area (small local areas of around 1,000–3,000 people) level as a measure of geography, the top 10% of the population live in areas that have around 3.4 times higher average wages than the bottom 10%. This ratio falls to 2.2 for wages by Middle Layer Super Output Areas (which are areas of 5,000–15,000 people) and to less than 1.4 for the large TTWAs (of between 100,000 and 1 million people, plus London at 9 million) that correspond most closely to labour markets.
- Differences between labour market areas in nominal wages are large and persistent, with average wages in the highest-paid area (London) being £20 in 2019 and those in the lowest-paid area (Scarborough) being £13. These differences across areas are driven mainly by where the high-skilled high-wage individuals are located – the wage of the worker at the 10th percentile (with wages higher than 10% of their peers and lower than the other 90%) is remarkably similar between poor and rich areas, whereas the 90th percentile of the wage distribution is 80% higher in the richest areas. Within-area inequalities are much greater than between-area inequalities, and are also greater in the richest areas than in the poorest areas. This means that some of the poorest people live in the richest places, which limits the ability of place-based redistribution to help the worst-off.
- While the differences in average wages and employment rates between the richest and poorest areas are still large, they have not been rising but have instead fallen slightly over the last 20 years. However, deindustrialisation and the fall in manufacturing output over the preceding 30 years had previously led to a significant widening of regional disparities, on which these small reductions in the past two decades have had little effect.
- Differences in education and skills across areas are striking. There are large differences between the fraction of the working-age population with degrees across rich to poor areas (varying from over 50% in Brighton or London to around 15% in an area such as Doncaster) and these differences are even bigger if one considers just workers aged 16–34 since young adults are increasingly located in richer city areas, and are also more likely to have degrees because of increases in education in younger cohorts. These differences arise from both differences in attainment of children growing up in different areas and selective patterns of mobility for those graduating from higher education. 19% of 27-year-olds who grew up in Grimsby, for example, are graduates, but half had left by age 27, and this was only partially offset by inward migration of graduates such that only 12% of those living in Grimsby at age 27 are graduates. This contrasts with large high-skill cities which have higher graduation rates amongst those born there (e.g. 35% in London, 26% in Leeds) and then further attract graduates through migration such that the fraction of 27-year-old residents who have degrees is considerably higher still (44% in London, 34% in Leeds). These educational and skill differences are key since an area’s wages and employment rates are highly correlated with the proportion of graduates in the area.
Given the strength of the correlation between an area’s labour market outcomes and the characteristics of the people living there, a large part of the Overman and Xu evidence is devoted to assessing how much of the observed differences between areas in the UK can be thought of as due to the individuals who live in those areas, as opposed to a ‘pure’ area effect. This requires a decomposition of the variance into different components, along the lines of previous work in the UK (Gibbons, Overman and Pelkonen, 2014) and more recent work in other countries, particularly by Card, Rothstein and Yi (2021) for the US. Even the simplest decomposition, which just uses the observable characteristics of individuals that are available in the wage data (age, gender and an occupation-based measure of skill), suggests that around 40% of area disparities can be attributed to the characteristics of residents in those areas. This approach takes the geographical distribution of skills and occupation as an individual characteristic, a point acknowledged by the authors since they are interested in geographical labour market processes. But it is important to note that individual differences in educational attainment may themselves reflect area effects if they are caused by factors such as different levels of school resources, aspirations and/or educational opportunities across areas.
There are many other characteristics of individuals that may affect wages that are unobserved and not controlled for in the simple analysis. To go beyond this, Overman and Xu use longitudinal data and exploit information on individuals who move between areas whilst being observed in the sample. Using such an analysis, the authors estimate that between 64% and 90% of area disparities can be attributed to the characteristics of the people living in those places. In comparison, Card, Rothstein and Yi (2021) estimate around two-thirds of disparities between commuter zones in the US to be attributable to individual characteristics and one-third to be an area effect. Their results are similar to estimates from France and Germany, and the paper includes a number of robustness checks for the methodology. When following an identical specification to Card et al., Overman and Xu estimate 72% and 28% respectively in the UK.
Hence, while the size and nature of place effects will be context-dependent, for the case of wages the story across a number of developed economies looks quite similar. That said, technological change and industrial shifts may well mean that these effects are quite different from those present in decades past (e.g. when the service economy was smaller and more employment was tied fairly directly to places, such as in mining or steel) or those we will see in future (e.g. as a rise in remote working weakens the relevance of location for job opportunities).
The implications of this decomposition of place and area effects for our understanding of geographical inequalities and what to do about them are discussed in Overman and Xu and, in a more general sense, are also taken up in the separate discussion pieces by van Ham, Manley and Tammaru and by Moretti. Van Ham et al. note that similar decompositions or attempts to control for spatial sorting for other outcomes often result in individual effects explaining a large amount but not all of area effects. They go on to discuss why this does not mean that place is unimportant, arguing that the geographical sorting process, potentially over many generations of social mobility and immobility when links between parents and children are taken into account, may itself be a function of neighbourhood or area effects. The authors also return to the theme that different levels of geographical sorting will be relevant for different types of outcomes. Moretti gives a comprehensive exposition of the pros and cons of place-based policies, covering both equity and efficiency rationales, in the light of evidence on the estimated size and nature of area effects and geographical sorting patterns such as those observed by Overman and Xu and others. All three pieces note the important role for place-based policies but comment on the need for them to address specific mechanisms and processes governing skill acquisition, social mobility and the location of high-skilled jobs, as opposed to just attempting to make areas more equal.
The labour market analysis and evidence released here will form only part of the IFS Deaton Review’s work on geographical inequalities. Forthcoming chapters on health inequalities and on race and ethnic inequalities will also be providing evidence on inequalities that have a systematic spatial dimension. Additionally, the evidence chapters on education and social mobility will, as is clear from the Overman and Xu analysis and the two accompanying commentary pieces published here, relate directly to underlying processes that are key in generating the spatial patterns observed in the wage and employment data. As such, there is still more that needs to be documented and learnt before a fuller body of evidence on geographical inequalities in the UK can be assembled. Thus the Deaton Review panel will be deliberating on these issues extensively over the next stages of its activities.
In the meantime, additional detail on the government’s ambitions for ‘levelling up’ and the role of place-based policies in addressing this challenge will hopefully add some much-needed substance and clarification to the UK policy debate. As Moretti points out, any place-based policies need to be informed by an understanding of the underlying sources of geographical disparities and the various trade-offs involved, rather than just an aspiration to make areas more equal. The evidence in Overman and Xu suggests that addressing the skill mix in deprived areas of the UK is one crucial example, and that policies aiming to equalise not just individual-level outcomes but also outcomes across places need to address the demand for and the supply of skills simultaneously if they are not to be undone by worker mobility.
