spatially defined nodes as the unit of analysis, and analyse how skewed distributions in city size or land
values can emerge resulting from network economies resulting from a node’s connectivity (Andersson et
al., 2003, 2004). Again, one is interested in explaining observable skewed distributions from an
evolutionary process emerging endogenously from an initially spatially uniform world. Thus, although the
precise modelling techniques and underlying theoretical assumptions greatly differ, both evolutionary and
neoclassical approaches of economic geography use formal models to explain the emergence of uneven
distributions in an initially even world.
As explained earlier, the new economic geography remains firmly within the neoclassical framework,
as witnessed in assumptions like the assumption of profit/utility maximisation, representative agents,
given technology, and static market structures. In this, it differs greatly from evolutionary theory that is
based on a different set of assumptions. In short, evolutionary economic geography assumes that firms
compete on the basis of their knowledge, routines and competences that are built up in the past (and within
a particular local environment) and that are hard to imitate by competitors. Following this reasoning, the
emergence of spatial agglomeration is not analysed as stemming from rational location decision of firms
and consumers, but from historically grown concentration of geographically localised knowledge. This
knowledge is primarily embodied within the routine of firms, but also within a firm’s relationships with
other firms and other actors. While the new economic geography focuses attention on factor costs, and the
trade-off between transportation costs and internal economies of scale, the evolutionary approach focuses
dynamics on firm learning and knowledge spillovers between firms the two main sources of
agglomeration.
Put differently, the evolutionary approach is interested in exploring the explicit mechanisms of
knowledge production within the firm mainly resulting from R&D and learning-by-doing (Nelson and
Winter 1982; Klepper 1996), and of knowledge transfer, through imitation, spin-offs, social networks,
labour mobility, collaborative networking, etc. In doing so, evolutionary approaches assume that an
important part of agglomeration economies are caused by knowledge spillovers as a vehicle of diffusion of
routines and competences, and to what extent proximity facilitates these knowledge spillovers to happen
(Jaffe et al. 1993; Feldman 1999; Antonelli, 2000; Breschi and Lissoni, 2002). In this context, regional
differences are dependent on, though not fully explained by, place-specific institutions that affect the
incentives of agents to share knowledge and to engage in collaborative projects.
By contrast, the new economic geography does not deal with the specifics of industry dynamics,
knowledge spillovers and institutional factors when analysing the degree of spatial agglomeration, partly
because institutional factors are difficult to introduce in formal analysis (‘best left to the sociologists’, as
Krugman once put it). Due to their neglect of industrial dynamics, the new economic geography has not
the ambition to explain the location of particular industries (Krugman 1991). Therefore, Martin (1999) is
right that the new economic geography is “... unable to tell where it (industrial localization and
specialization) occurs, or why in particular places and not in others” (p. 78). We would argue that the aim
of an evolutionary approach is to do just that: it analyses the spatial evolution of a population of non-
identical firms with different routines in a particular industry, rather than the spatial pattern of economic
activities in general.
Apart from the theoretical differences between evolutionary and neoclassical approaches to economic
geography, the treatment of dynamics in both theories is different as well. Although the new economic
geography models are often interpreted as reflecting the formation of cities, it still bases its conclusions on
a static account of equilibrium. Model outcomes are derived by computing, for all individual agents, their
17