Tuesday 13 November 2018

Measuring Urban Economic Density

[By Dzhamilya Nigmatulina]

What makes cities productive? For a long time, researchers have known that city size matters a lot – people who live in larger cities tend to earn more and be more productive. People (and firms) benefit from proximity to each other and these benefits are manifested in cheaper trade, improved hiring and learning, and a wider choice of inputs and products.

But does city size always generate proximity gains? A city that occupies a large area may also have very low density and low proximity, compared to a more compact city with the same population. In the same way, even a city’s high average density may not mean high proximity. Can we test if the shape of city density has economic implications? If so, how can we capture the meaningful dimensions of the city density?

In our recent CEP discussion paper we try to answer these questions using data from cities of sub-Saharan Africa.

A real world example

Cities with poor planning and land institutions may face issues of low proximity. In 1994 the urban scholar Lusugga Kironde living in Dar-es-Salaam noted about his city in his thesis:

"[…] Dar-es-Salaam is not systematically structured either in form of functional zones, or of income levels, or of types of developments. It does not reflect a structure responding to either a city well ordered by government, or, to the niceties of the market theory. It is this kind of unpatterned, irregular, and hotch potchy development, inconsistent with good land use planning or with the economic theories of land use structure, which engendered interest leading to this study."

Figure 1. Dar-es-Salaam, according to 1992 aerial photograph
Source: Kithakye 2009

Twenty years later and not much changed in Dar-es-Salaam. Even though the residential population distribution on Figure 3 tells us nothing about the location of firms and workers, it is evident that people reside far from each other: the highest density settlements are informal, and large residential areas are far away from the coastal centre of the city. The story of Dar-es-Salaam is not a unique one in Sub-Saharan Africa. Colonial history, imperfect transport systems, and lack of formal land has impelled cities to grow haphazardly, leaving many crammed in informal sector areas with poor access to the economic and social opportunities of urban life.

Figure 2. Dar-es-Salaam population distribution according to the 2012 population census.
Source: Tanzania National Bureau of Statistics, Population census 2012 and author’s calculations

Does shape matter?

In our study we use Landscan - a dataset the captures the distribution of ambient population, i.e. the average number of people in each square kilometre of the city over 24 hours. This measure captures not only residents, but also indirectly the location of firms. Understanding the actual proximity of people or workers within a city also requires a measure of the city extent, which defines a self-contained labor market. We choose population density to define city extents and use 1,500 people per km^2 as a simple cutoff.

Living Standard Measurement Study (LSMS) surveys from the World Bank give us income and wages for cities in six sub-Saharan Africa countries.

African cities are surprisingly dense and clustered

Comparing 599 large cities in Africa to other continents reveals that the African cities are denser, and more clustered, than those in the developed world. This is surprising, since the example of Dar-es-Salaam is often seen as representative of a more general problem of weaker land institutions in Africa (which would lead to lower clustering). If we look at differences of the cities in sub-Saharan Africa and the cities in other developing countries, we find that sub-Saharan Africa, Asia and North Africa are similar in terms of density and all clustering measures. It is Latin America that is less clustered.

Looking across African cities, density correlates with individual and household wages in that city, but clustering does not help explain differences in wages. This is puzzling. Especially when we see that for detailed firm-level data in Kampala higher worker density does matter for firm value added.


What explains the higher than expected clustering of African cities? One possible explanation is that our measure of “ambient” population captures the residential population more than the worker population. If higher density, or “cramming”, compensates for difficult commutes then the pattern of residential locations swamps the distribution of employment that matters in explaining productivity.

This could also explain why our clustering measures fail to explain wage premiums even though the Marshallian theories (and some additional findings for worker density in Kampala) highlight the benefits of proximity. Landscan only provides a proxy (e.g. built cover, building heights and other undisclosed data) measure for the things we care about. Comparing and contrasting novel spatial datasets, such as Landscan, with traditional survey and economic census data would help improve our understanding.

Adding transport costs can also paint a better picture on the true proximity between workers. See Akbar et al. (2017) for steps in this direction.

These questions and data limitations provide lots of opportunities for future research. As better data become within reach, we will be able to calculate our clustering indices and better test our proximity theories.