4. Spatial Structures of Regional Industrial Agglomerations
4.1. Spatial Econometric Models
The industrial variables in table 5 were constructed through the municipality-based aggregation of
data of local industrial units. A statistical model of imputation was developed to classify the companies
which are in two different data bases: PIA (Industrial Research by Sampling) and PINTEC
(Technological Innovation - Industrial Research), both IBGE industrial data bases. The classification of
the local units pursuant to innovation criteria defined by PINTEC followed the classification given to
the company: firms which innovate, differentiate products, are price markers, and export (type A),
firms that specialize in standardized products, are price takers, and that export (type B), firms which do
not differentiate products, are price takers, have lower productivity and do not export (type C). The
location quotients (QLA, QLB and QLC) for each of these categories were calculated based on the IVA
for each type. A municipality’s sector-based industrial structure is captured by variables that indicate
sector shares in the total IVA of that municipality. So BI denotes the participation of the intermediate
goods industry in the local IVA, BCD is the indicator for capital goods and durable consumer goods,
BCND for non-durable consumer goods and EXTRA for the extraction industry.5
The socio-economic variables listed in Table 5 are defined for each of the 5,507 Brazilian
municipalities, based on information available from different sources. Selected variables capture some
aspects of Brazil’s economic space structure, such as upper schooling levels (E25), serving to measure
educational qualifications across the municipality’s labor force; population (POP), a measure of the
scale of the local economy and/or market; percentage of the local population provided with sewage
connection to the sanitary system (ESGT), a measure of urban infrastructure availability; and the
classification of the municipality compared to certain metropolitan areas (NRM)6. Transportation cost
variables are determined by applying a linear programming procedure to calculate the lowest cost
incurred to travel from the center of a given municipality to the city of Sao Paulo and to the nearest
state capital (CTRPSP and CTRPCAP, respectively).7
5 The sum of these four variables for a given municipality is equal to 1, so that only three of them should be used in the
regressions (the one excluded is reflected in the constant).
6 The modeling effort covered 5179 non-metropolitan and 328 metropolitan municipalities, distributed among 13
metropolitan areas: Belém, Teresina, Fortaleza, Maceio, Natal, Recife, Salvador, Sao Lms, Goiânia, Brasilia, Vitoria, Belo
Horizonte, Rio de Janeiro, Sao Paulo, Campinas, Santos, Curitiba, Florianopolis and Porto Alegre.
7 Highway transportation costs are estimated as a function of the distance and cost of the paving type of federal and state
highways (see Castro et al., 1999).
14