Agglomeration economies are a persistent subject of debate among economists and urban planners. Their definition turns on whether or not larger cities and regions are more efficient and more productive than smaller ones. We complement existing discussion on agglomeration economies and the urban wage premium here by providing a sensitivity analysis of estimated coefficients to different delineations of urban agglomeration as well as to different definitions of the economic measure that summarises the urban premium. This quantity can consist of total wages measured at the place of work, or of income registered at the place of residence. The chosen option influences the scaling behaviour of city size as well as the spatial distribution of the phenomenon at the city level. Spatial discrepancies between the distribution of jobs and the distribution of households at different economic levels makes city definitions crucial to the estimation of economic relations which vary with city size. We argue this point by regressing measures of income and wage over about five thousands different definitions of cities in France, based on our algorithmic aggregation of administrative spatial units at regular cutoffs which reflect density, population thresholds and commuting flows. We also go beyond aggregated observations of wages and income by searching for evidence of larger inequalities and economic segregation in the largest cities. This paper therefore considers the spatial and economic complexity of cities with respect to discussion about how we measure agglomeration economies. It provides a basis for reflection on alternative ways to model the processes which lead to observed variations, and this can provide insights for more comprehensive regional planning.

Scaling laws are simple, easily usable and proven relevant models used in geography for validating various urban theories. These non-linear relationships may reveal physical constraints on the structure and evolution of complex systems, and underline the relationship between urban functions, size of cities and innovation cycles. In this contribution, we examine to what extent scaling laws are transferable towards urban theories and in which specific fields of urban geography these models may be relevant. We thus focus on the accuracy of scaling laws when exploring structures and processes of systems of cities, the diffusion of innovation, metropolization and intra-urban dynamics. We therefore use several examples taken in different regions of the world, embedded in various historical, political and economic contexts. However, in some cases, care must be taken not to over-interpret the results obtained from scaling laws and not to give scaling laws more explanatory power than they can describe. We illustrate this point by providing recommendations relying for instance on the sensitivity of measurements to the delineation of each object of the system under study and to the definition of the system itself. These recommendations can help to get robust results in order to understand the generic evolutionary mechanisms in urban systems.

Both theoretical and empirical studies have shown the ability of scaling laws to reveal processes of emergence in urban systems. Nevertheless, a controversy about the robustness of results obtained with these models on empirical cases remains, regarding for instance the definition of the ‘city’ considered or the way the estimations are performed. Another source of bias is highlighted in this contribution, with respect to the non-ubiquitous character of some urban attributes (i.e. their partial absence from several cities of the system). The problem with the zero count for cities where these attributes are absent is that the technical necessities of usual estimation procedures make the analysis ignore them altogether even when they represent some valid information. This could seriously impact the results. A precise exploration of the effects of this arbitrary filtering is conducted here, and several solutions are proposed to overcome this limitation. In a case study about foreign investment towards French cities, we show that some erroneous conclusions about a hierarchical diffusion could be drawn when adopting the classical ordinary least squares approach. The framework we suggest specifies how it is possible to avoid misinterpretations deriving from the exclusion of zero values by using methods of analysis which deal with zero values specifically. The conclusion of a diffusion of foreign investment in the French urban system is then rejected.