Residential sorting by education level enhances the return on investments in transport infrastructure

Coen Teulings, Ioulia Ossokina 23 April 2018



There is a strong pattern of spatial sorting of various levels of education. In a study of 1,500 regions in 100 countries, Gennaioli et al. (2013) show that within all countries, the highly educated tend to cluster in particular regions. Interregional differences in GDP per capita per year increase in the mean level of education and are far above any reasonable estimate of the return to human capital. Glaeser and Saiz (2003) have argued that higher educated are crucial for urban revival. Figure 1 documents a similar pattern for the Netherlands.The left panel shows the spatial segregation between high and low educated workers. Roughly speaking, highly educated workers live (and work) in cities, while the low-educated live in the countryside. In cities in the western part of the country, there are neighbourhoods where up to 70% of the workers are highly educated, while in regions along the northern border with Germany, this is less than 15%. The strong spatial segregation has led many people to worry about society falling in different education classes that hardly interact (see, for example, the most recent book by Florida 2017). This worry has received more attention since the rise of populism.

These differences in mean educational attainment are closely mirrored by the local land rents (see the right panel of Figure 1). Land prices vary by a factor 300 or more, from 3,800 euro/m2 in Amsterdam Canal Zone (high share highly educated) to some 10 euro/m2 along the northeastern border with Germany (low share highly educated). See De Groot et al. (2015) for a broad overview of land rents and their determinants in the Netherlands. Local public goods are an important factor in explaining this wide variety in land rents.

Figure 1 Percentage highly educated per zip code (left) and land prices, in euro/m2 (right)

A recent study by Teulings et al. (2018) uses microdata to quantify the differences in the willingness to pay for particular locations between the high and low educated (omitting the medium education level) (Figure 2). It shows willingness to pay for the job availability (based on the locally available transport infrastructure to commute to these jobs) and urban amenities such as parks and historic scenery at the location. The highly educated (right panel) are very sensitive to the quality of a location. Their willingness to pay for downtown Amsterdam is twice as high as that of lower educated, while their willingness to pay for locations near the German border is six times smaller than that of the low-educated. The low-educated, on the other hand, are much more sensitive to the level of land prices when choosing their home location. Teulings et al. estimate their sensitivity to land prices to be twice as large as for the highly educated, suggesting the land is a normal good on consumption. Albouy et al. (2016) reported a similar result for the US.

Figure 2 Value of the location in euro/m2, low-educated (left) and highly educated (right)

Sorting leads to extra welfare benefits from local public goods

Figure 2 suggests that an investment in local public goods will increase residential demand for a location, especially of the highly educated, and thus induce residential sorting. By embedding these results in a general equilibrium model, one can calculate welfare effects of investments in local public goods. Teulings et al. (2018) use such a model to analyse the welfare benefits of the railway connections between Amsterdam and the region north of the city. The two areas are divided by a canal; two railway tunnels and five car tunnels connect them. The authors run a counterfactual and compare the equilibrium with and without the railway tunnels. Figure 3 illustrates the results.

The railway connection between Amsterdam and its periphery in the north increases the number of people willing to commute to the centre. The number of firms and jobs at the centre increases, at the expense of the periphery (Figure 3, left panel). However, the better accessibility combined to the jobs in Amsterdam makes the periphery a more attractive location for living (middle panel). These effects are most marked in locations close to stations along the railway connections to Amsterdam (the dark grey lines in Figure 3). While economic activity in terms of the number of jobs goes down in the North, land prices go up as the region becomes an attractive residential location (right panel). Higher land prices imply an increase in the population density in this area in the long run, as the demand for land by a worker is price elastic. This higher population density increases the return to the investment, as more people use the transport infrastructure.

Figure 3 Location of jobs and population, situation with tunnels compared to situation without tunnels

However, there is a marked difference in costs and benefits between levels of education (see Table 1). The table calculates the net benefits for three levels of education and for landowners in the north, in Amsterdam, and in the rest of the country. (We classify everybody as a renter; to the extent that somebody is a home-owner, he or she also captures part of the landowners benefit). The benefits of the reduced travel time predominantly accrue to the higher educated, since a much higher share of them travels by train and since they commute on longer distances. Similarly, the modal shift from the car to train also favours the higher educated since they are more viable to travelling by train in the first place. Job relocation is also much more valuable for highly educated workers, simply because the gain of working in Amsterdam is much higher to them. The most surprising conclusion is that the lower educated actually loose, because land close to stations becomes more expensive. The higher land rent drives many lower educated who don’t use the train out of their homes, forcing them to seek residence at less attractive locations. The share of the higher educated in the north increases. The total effect can be shown to be a reasonable estimate of the social benefit of the new transport infrastructure. Using this metric, Table 1 shows that the welfare benefits due to relocation are large – they amount to 30% of the total effect from the tunnels. The other 70% are direct travel time savings of consumers who do not relocate. The highly skilled relocate more and they get the major part of the total benefits. 

Table 1 Welfare benefits of the railway tunnels, in millions of euros net present value

Land redevelopment is optimal but can pose political economy challenges

Relocation of the highly skilled to locations near the railway stations in the north is efficient, because they value the train connection to Amsterdam the most. This relocation generates wider economic benefits equal to 30% of the total effect from the transportation investment. However, a necessary condition for the realisation of these benefits is the adjustment of the housing stock to the preferences of the highly skilled. This may mean a complete redevelopment of land near the railway stations, leading to welfare losses to other population groups who need to move to other neighbourhoods. This poses challenges to the political economy of the investment in transport infrastructure.

The insights obtained for the railway tunnels can be extrapolated to other local amenities. In various European countries local governments invest in amenities and public spaces to affect the population composition of deprived neighbourhoods and contribute to their gentrification (e.g. Cheshire 2009, Gonzalez-Pampillón et al. 2016 and the references therein). Our results suggest that investing in local public goods alone may be insufficient. In order to change the composition of the population in a neighbourhood, complementing adjustment in the housing stock and the land use intensity should be made. Recent fast revival of a heavily deprived neighbourhood, Katendrecht, in the city of Rotterdam offers some crude evidence (see Table 2). In 1999, Katendrecht was seriously lagging behind the city average in terms of the share of highly skilled and the housing stock, and was unlikely to be gentrified. Coordinated investments in the housing stock and local amenities brought a change. In 15 years, 30% of the dwellings was replaced with new construction focussed on middle income families. As a result, by 2014, the neighbourhood’s population and housing stock has become comparable to the city’s average.

Table 2 Revival of the neighbourhood of Katendrecht in Rotterdam, the Netherlands


Albouy, D, G Ehrlich, and Y Liu (2016), “Housing demand, cost-of-living inequality, and the affordability crisis”, NBER, Working Paper 22816. 

Cheshire, P (2009), “Policies for mixed communities: faith-based displacement activity?”, International Regional Science Review 32: 343-375.

Florida, R (2017), The new urban Crisis. Gentrification, Housing bubbles, Growing inequality, and what we can do about it, One world publications, Edinburgh.

Gennaioli, N, R La Porta, Rafael, F Lopez-de-Silanes, Florencio and A Shleifer (2013), "Human capital and regional development", The Quarterly Journal of Economics 128: 105-164.

Glaeser, E and A Saiz (2003), "The rise of the skilled city", NBER,Working paper 10191.

González-Pampillón, N, J Jofre-Monseny, and E Viladecans-Marsal (2016), “Can urban renewal policies reverse neighborhood etnic dynamics?”,CEPR Discussion Paper DP11676. 

De Groot, H, G Marlet, C Teulings and W Vermeulen (2015), Cities and the urban land premium, Edward Elgar Publishing.

Teulings, C,  I Ossokina and H de Groot (2018), "Land use, worker heterogeneity and welfare benefits of public goods", Journal of Urban Economics 103: 67-82.



Topics:  Labour markets Migration Politics and economics

Tags:  segregation, residential sorting, Public goods, transport infrastructure

Distinguished Professor of Economics, Utrecht University; CEPR Research Fellow

Assistant Professor, Department of the Built Environment, Eindhoven University of Technology.


CEPR Policy Research