VoxEU Column Development Migration

Rethinking urbanisation

This column says that the two central stylised facts of population growth – Gibrat's Law and Zipf’s Law – do not hold in a wider sample of rural and urban areas. Analysis using new data that describe population patterns in both urban and rural areas suggests that populations experience a divide, as the dense grow and the sparse depopulate.

Urbanisation – the concentration of population in cities and towns – is one of the most striking features of economic development. The share of the world’s population living in cities grew from less than one tenth in 1300 to around one sixth in 1900 to around one half today. While this transition from rural to urban is largely complete in developed countries such as the US, the urbanisation process continues apace in developing countries such as Brazil, China, and India. In China alone, 240 million people are expected to migrate from rural to urban areas by 2025, helping to raise the share of the world’s population living in cities to 60% by 2030.

Urbanisation matters

Understanding the determinants and pattern of urbanisation is central to a host of policy issues. Urbanisation raises major environmental and social concerns. The conversion of open countryside to built-up areas involves the loss of natural habitat, the elimination of recreational space, and increased demands for agricultural food surpluses to feed the urban population. Dense population concentrations also create major infrastructure demands: rapid transit systems for commuting, supplies of clean water and power, sanitation and waste management systems, and public health facilities.

Yet despite the high levels of mortality observed in cities in nineteenth century Europe and the developing world today and all these drawbacks, hundreds of millions have moved from rural to urban areas. There must be compensating advantages such as higher productivity and wages, a wider range of employment opportunities, goods, and services, and a greater diversity along ethnic, cultural, and other dimensions.

Paucity of research

In spite of the magnitude of this rural-urban migration, there has been relatively little study of the process which transforms rural areas into urban areas. Partly for reasons of data availability, most existing research on population growth focuses on urban areas. This research emphasises two key stylised facts.

•    Population growth appears largely uncorrelated with initial population size (Gibrat's Law). (See Simon 1955 and Gabaix 1999.)
•    There is a stable population distribution that can be approximated by Zipf's Law (the second largest location has half the population of the largest location, the third largest location has a third the population of the largest location, and so on). (See Rosen and Resnick 1980 and Soo 2005.)

This exclusion of rural areas from existing research is a concern because – historically in developed countries and contemporaneously in developing countries – these areas account for a large share of the overall population. As it turns out, this exclusion has warped our understanding of urbanisation.

Revising the stylised facts

Our research shows that these two stylised facts for population growth are strongly rejected if both rural and urban areas are considered. Using newly constructed data based on sub-county units in the US from 1880 to 2000 and based on municipalities in Brazil from 1970 to 2000, we find that population growth is decreasing in initial population density at low densities, increasing in initial population density at intermediate densities, and largely uncorrelated with initial population density at the high densities observed in urban areas.

The resulting U-shaped relationship for population growth is shown in Figure 1 and results in an unstable population distribution over time. Along the upward-sloping segment of the U-shape, locations with a high initial population density grow relatively rapidly and become urban areas, while other locations with a lower initial population density grow relatively slowly and remain rural areas. As a result, even though there has been a substantial increase in the US population between 1880 and 2000, we find an increase in the number of sparsely populated locations as well as an increase in the number of densely populated locations.

Figure 1. Population growth and density, 1880 - 2000

Note: The solid line shows mean population growth rate from 1880-2000 for US Minor Civil Divisions. The dashed lines show 95% confidence intervals based on robust standard errors clustered by county.

Our research explains these patterns of population growth in terms of the distinction between agriculture and non-agriculture (i.e. manufacturing and services). Two key ideas play an important role in our analysis.

•    The employment growth process differs between agriculture and non-agriculture.

In agriculture, employment growth is negatively related to initial employment size, because after positive shocks the relative employment of locations reverts towards a long-run mean determined by, for example, climate and soil. In contrast, in non-agriculture, employment growth is largely unrelated to initial employment size, because persistent factors such as climate and soil impose fewer constraints on employment growth. As a result, the share of non-agricultural employment in the population is increasing in initial population density, at least for sufficiently high initial population densities.

•    Over time, faster productivity growth in agriculture than in non-agriculture activities has resulted in a reallocation of employment away from the agricultural sector.

The combination of these two ideas can explain the observed U-shaped relationship between population growth and initial population density. At low population densities, agricultural employment dominates, and mean reversion in agriculture generates a decreasing relationship between population growth and initial population density. In contrast, at high population densities, non-agricultural employment dominates, and the absence of mean reversion implies that population growth is largely uncorrelated with initial population density. In between, the positive correlation between the share of non-agriculture in employment share and initial population density, combined with faster employment growth in non-agriculture than in agriculture, generates an increasing relationship between population growth and initial population density.

Policy implications

Our findings have a number of policy implications. As large developing countries such as China and India industrialise, the resulting increase in the urban population is likely to create major pressures on the environment and infrastructure. Our finding that population growth is increasing in initial population density at intermediate densities and largely uncorrelated with initial population density at higher densities suggests that more densely-populated areas are likely to experience larger increases in the urban population. While our analysis suggests that some locations with higher initial population density will experience rapid population growth, other locations with lower initial population density will experience declines in relative and possibly absolute population. Therefore the rural depopulation that has afflicted many small towns in the American Midwest could also become a feature of the agricultural peripheries of currently developing nations. Taken together, the major changes in the distribution of population across space that our analysis points to will require major reorientations in public infrastructure, which are likely to provide considerable challenges for governments throughout the developing world.


Gabaix, Xavier (1999) Zipf’s Law for Cities: An Explanation, Quarterly Journal of Economics, 114(3), 739-767.
Michaels Guy, Ferdinand Rauch and Stephen Redding, Urbanisation and Structural Transformation, CEPR Discussion Paper, 7016, October 2008. This research was sponsored by the Centre for Economic Performance, London School of Economics.
Rosen, Kenneth T. and Mitchel Resnick (1980) The Size Distribution of Cities: An Examination of the Pareto Law and Primacy, Journal of Urban Economics, 8(2), 165-186.
Simon, Herbert A. (1955) On a Class of Skew Distribution Functions, Biometrika, 42(3/4), 425.440
Soo, Kwok Tong (2005) Zipf’s Law for Cities: A Cross-country Investigation, Regional Science and Urban Economics, 35(3), 239-263.

1,082 Reads