The studies reviewed so far all have estimated a 1-dimensional wage
gradient. This implicitly assumes a single major center of employment, together
with ‘‘local’’ employment that uses only a ‘‘few’’ workers. An alternative, but
still 1-dimensional specification, has employment sharing land with residences
while commuting occurs along continuous radial routes. Theoretical models
confirm the existence of a wage gradient and show under a wide range of
conditions that employment density is more centralized than that of households
ŽOgawa and Fujita 19 . In the real world, however, production economies of .
scale and the irregularities of regional transportation systems have given rise to
‘‘clusters’’ of local employmentoften referred to as ‘‘edge cities.’’ Commuting to and around such subcenters has become a complex 2-dimensional
problem, in contrast to the simplistic radial commuting that gives rise to the
wage gradient approach White 28,29 , McMillen and McDonald 13 . If largeŽ .
secondary centers develop in the suburbs of an urban area, with smaller centers
or dispersed employment located in between the primary and suburban centers,
a 1-dimensional gradient will simply not capture the true variation in wages.
The second contribution of this paper is to treat employment location as
discrete within 2-dimensional space: firms and households can reside in one of
a finite number of zones, with a commuting cost matrix that is exogenous. This
allows for a less restrictive specification of employment location and generates
more general patterns of wage and land rent variation. It also matches with the
micro data set we will usewhich intrinsically treats space as discrete. In all
such data sets, individuals and firms are coded to some set of discrete zones
Ž .